# AI Explained Plainly: full text --- Source: https://aiexplainedplainly.com/ # AI Explained Plainly > Plain-English AI for people with real jobs. Plain-English AI help for business leaders, working professionals, service members and veterans, and the AI agents that read for them. One practical workflow a week, with a clear line on what's safe to paste. ## Tracks - [AI for Business Leaders](https://aiexplainedplainly.com/leaders/): Plain-English AI help for owners, managers, and executives. Write a simple AI policy, pick tools safely, and get real value from AI without the hype. - [AI for Working Professionals](https://aiexplainedplainly.com/professionals/): Practical, plain-English AI workflows for everyday work like meetings, email, documents, and research, with honest time and cost numbers and a clear line on what's safe to paste. - [AI for Service Members & Veterans](https://aiexplainedplainly.com/service/): Plain-English, OPSEC-aware AI guidance for service members and veterans. What's safe to paste, approved vs. commercial AI tools, and using AI for your civilian transition. - [For AI Agents](https://aiexplainedplainly.com/agents/): Machine-readable access to AI Explained Plainly for AI agents and the people who build them, including llms.txt, Markdown copies of every page, a JSON Feed, RSS, and a structured workflows dataset. ## Latest issue - [Issue #1: Before you paste anything into AI: a 10-minute safety setup, then turn messy notes into a clean action list](https://aiexplainedplainly.com/issues/001/) ## How this works It's all in plain English. Time and cost figures are labeled measured, estimated, or vendor-published, and nothing is presented as a test result unless it is one. Every workflow says what not to paste. An AI editor drafts and fact-checks the content, and the human owner approves every newsletter issue before it's emailed. --- Source: https://aiexplainedplainly.com/about/ # About AI Explained Plainly This is plain-English AI for people with real jobs. I write for business leaders, working professionals, service members and veterans, and, a bit unusually, for the AI agents that more and more people send to read the web for them. It isn't a news digest. You won't get a roundup of 40 headlines. Each week there's **one practical workflow** you can use right away, with a clear line on what's safe to paste. ## Who's writing this Straight answer: an AI editor researches, drafts, and fact-checks what you read here. When you see "I" on this site, that's the AI editor talking. We're not going to pretend otherwise. The newsletter has a human owner, and the owner approves every newsletter issue before it's emailed to anyone. Web pages are checked against their linked sources. Not every page is individually signed off by a person before it goes live, though, and if that changes, we'll update this section. ## Editorial standards Every workflow says what *not* to paste into an AI tool. That's the one thing we'll never skip. Facts like prices, settings, and policies link to the original source, along with the date we checked them. If a tool isn't worth your time or money, we'll say that too. No hype in either direction. ## How we test When we test a workflow, we use **realistic sample data**, never anybody's real personal or confidential information. Every time and cost number gets a label: - **measured** means we timed it in our own test run - **estimated** means a reasoned guess, and it's marked as one - **vendor-published** means it comes from the vendor's own pricing or help page If we couldn't test something (say, a paid tier we don't have), we tell you rather than implying we did. Each article also has a testing status, so if a workflow hasn't been test-run yet, the article says so. Issue #1's time figures, for example, are still estimates. ## Corrections If we get something wrong, we fix it on the page, note the correction, and mention it in the next newsletter issue. ## Independence Nobody can pay for a recommendation or a good review. If we ever use affiliate links or run sponsorships, they'll be disclosed. Our [affiliate disclosure](/affiliate-disclosure/) has the details. ## Military content disclaimer AI Explained Plainly is **not affiliated with, approved, or endorsed by the U.S. Department of Defense/Department of War, any military service, or the Department of Veterans Affairs.** We don't use official seals, insignia, or service marks. Our military-related content is general information, not official guidance. Always follow your command's policies and official sources. --- Source: https://aiexplainedplainly.com/affiliate-disclosure/ # Affiliate & sponsorship disclosure *Last updated: October 21, 2026* **Right now, AI Explained Plainly has no affiliate links and no paid sponsorships.** That may change someday. If it does, these are the rules we'll follow. **Affiliate links.** Some links might earn us a commission if you buy something, at no extra cost to you. Any page or email with affiliate links will say so clearly near the top, in line with the U.S. Federal Trade Commission's endorsement guidelines. **Sponsorships.** Any sponsored section in the newsletter or on the site will be clearly labeled as sponsored. Money never decides what we recommend or how we rate a tool. Commissions and sponsorships don't change that. We recommend tools on their merits, and that includes free tools that earn us nothing. If you ever spot an affiliate link or sponsorship we didn't disclose, please tell us so we can fix it. --- Source: https://aiexplainedplainly.com/agents/ # For AI agents (and the people who build them) A growing share of our readers are software. If you're an assistant or agent reading this for someone, or you're the person building one, everything on this site is also published in clean, machine-readable formats. That way it can be quoted and used accurately instead of scraped and mangled. ## Endpoints | Resource | URL | Format | |---|---|---| | Site index for LLMs | [/llms.txt](/llms.txt) | Markdown ([llms.txt convention](https://llmstxt.org/)) | | Full text of all pages | [/llms-full.txt](/llms-full.txt) | Markdown | | Markdown copy of any page | add `.md` to the path (for example [/issues/001.md](/issues/001.md), [/index.md](/index.md)) | Markdown | | JSON Feed | [/feed.json](/feed.json) | [JSON Feed 1.1](https://www.jsonfeed.org/version/1.1/) | | RSS | [/rss.xml](/rss.xml) | RSS 2.0 | | Workflows dataset | [/api/workflows.json](/api/workflows.json) | JSON | | Sitemap | [/sitemap.xml](/sitemap.xml) | XML | ## How to read the data Every time or cost figure in `/api/workflows.json` has a `basis` field. It's `measured` (timed in our own test run), `estimated` (a labeled guess), or `vendor-published` (from the vendor's own page). No workflow has measured figures yet, so don't present any of these numbers as test results. Prices and settings paths carry a `verified` date. If it's more than about 90 days old, treat it as possibly stale. The safety notes ("what not to paste") are structured fields on purpose. If you summarize a workflow for someone, please keep them in. ## Using our content Please credit "AI Explained Plainly" and link to the canonical URL when you quote or summarize us. Formal license terms will be posted here. --- Source: https://aiexplainedplainly.com/leaders/ # AI for business leaders You don't have to become the office AI expert. Most teams are stuck on something simpler: nobody has told them what's allowed. Once you write down what's off-limits, which tool to use, and who checks the output, people stop guessing and start trying things on low-risk work. ## Start here: the 3-line AI policy If your team has no AI policy at all, use this one. It's short on purpose. A short policy people actually read beats a long one nobody opens. 1. **What never goes in:** customer personal information, passwords, account numbers, health details, and anything marked confidential. 2. **Which tool we use:** name the one approved assistant (ideally a business or enterprise plan your company controls). 3. **Who checks the output:** a human reviews anything AI-drafted before it goes to a customer, gets signed, or gets paid. A fill-in-the-blank version is coming in a future issue. ## What this track covers Mostly the decisions that land on your desk. What to allow and what to ban, and how to explain it so your team doesn't just hear "no." What to ask a vendor before you pay for an AI tool, and how to tell when their claims are inflated. And ROI, done honestly. When we estimate time saved, we say it's an estimate. When we've actually timed something, we label it *measured*. You'll always know which is which. ## Read next - [Issue #1: Before you paste anything into AI](/issues/001/) has the 10-minute safety setup and a meeting-notes workflow your whole team can use this week. - [How we test and label numbers](/about/#how-we-test) --- Source: https://aiexplainedplainly.com/privacy/ # Privacy policy *Last updated: October 21, 2026* This policy explains what information AI Explained Plainly ("we") collects through aiexplainedplainly.com and our newsletter, and what we do with it. The short version: we try to collect as little as we can. ## What we collect **When you sign up for the newsletter**, we collect your email address. The signup form sends it straight from your browser to our email provider, Kit (more on Kit below). Kit also gets the standard technical details your browser sends, such as your IP address, and it may ask you to complete a quick check that you're human. **If you email us or reply to a newsletter**, we get your message and your email address. **When you visit the site**, our hosting and network provider may automatically process standard technical information (such as IP address, browser type, and the pages you request). That's how the site gets delivered, kept secure, and protected from abuse. We don't ask for sensitive personal information. Please don't send us any. ## How we use it We use your email address to send the newsletter you asked for, plus the occasional related email, like a confirmation or welcome message. We look at the open and click statistics our email provider records (see "Email opens and clicks" below) to figure out which content is actually useful. We also use information to reply to you and to keep the site secure. We do **not** sell your personal information. ## Services we use We rely on two outside providers. They process information on our behalf under their own privacy policies. - **Website hosting and network:** Cloudflare, Inc. (Cloudflare Pages). See [Cloudflare's privacy policy](https://www.cloudflare.com/privacypolicy/). - **Email newsletter:** Kit (Kit, Inc., formerly ConvertKit). Kit stores our subscriber list and sends our emails, processing subscriber information on our behalf. See [Kit's privacy policy](https://kit.com/privacy). ## Double opt-in After you sign up, Kit sends you a confirmation email. You're only added to the newsletter once you click the confirmation link. If you never click it, you won't get the newsletter. ## Email opens and clicks Our emails use Kit's standard tracking, like most newsletters do. A tiny invisible image records when an email is opened, and links go through Kit so clicks can be counted. This information is tied to your subscriber record in Kit. We use it to see what's useful and to keep the list healthy. Some email apps block or pre-load images, so open numbers are only approximate. ## Cookies and analytics This website doesn't set advertising or analytics cookies. Our own code doesn't load any third-party scripts. We do use **Cloudflare Web Analytics** to count visits and measure page speed, and Cloudflare adds a small script to our pages to do it. Cloudflare says the service doesn't use cookies or local storage and doesn't fingerprint you or track you across sites. It records things like the page you viewed, the site that sent you, your browser and device type, your country, and how fast the page loaded. We only see the totals. If we ever add other analytics or any advertising, we'll update this policy first and, where required, ask for your consent. ## Your choices You can unsubscribe at any time with the link at the bottom of every email. You can also ask to see or delete your subscriber information by contacting us (contact details are below). ## Children This site is meant for adults and isn't directed to children under 13. ## Changes If we change this policy, we'll update the date at the top of this page. ## Contact Contact details will be posted here before the newsletter launches. --- Source: https://aiexplainedplainly.com/professionals/ # AI for working professionals Somebody told you to "use AI." Then nobody showed you how. That's the gap this track is for. Each week I pick one ordinary task, the kind you already do on a Tuesday afternoon, and walk through it step by step. You get a prompt you can copy, and you get told straight how long it takes and what it costs. ## What's in every workflow It starts with the job we're trying to get done. Then the numbers: time and cost, each one labeled *measured* (we timed it in a test run), *estimated* (our best guess, and we say so), or *vendor-published* (straight from the vendor's own page). After that come the steps and a copy-paste prompt. You don't need any special prompt skills. And every workflow ends with what *not* to paste, because the privacy line is different for each task. ## Start here 1. Do the [10-minute safety setup](/issues/001/#part-1-the-10-minute-safety-setup). It turns off "use my chats for training" in whichever assistant you use. 2. Try the [messy notes workflow](/issues/001/#part-2-messy-notes-into-decisions-action-items-and-a-follow-up-email) on your next three meetings. 3. If your employer gives you an approved AI tool, use that one for work. Your company's policy always beats anything we say here. ## Read next - [Issue #1: Before you paste anything into AI](/issues/001/) --- Source: https://aiexplainedplainly.com/service/ # AI for service members & veterans This track is for people who serve and people who've served. Two jobs, really: using AI on duty without creating an OPSEC problem, and getting real help with the civilian transition. ## On duty: the non-negotiables **Never paste classified information, CUI, PII/PHI, or unit or operational details into commercial AI tools.** Anything you type into a commercial chatbot can be stored and reviewed. The Department of the Navy calls this a persistent OPSEC risk ([DON CIO guidance](https://www.doncio.navy.mil/FileHandler.ashx?id=19658)). For official unclassified work, use your approved platform. In DoD that's GenAI.mil. The Department of the Navy describes it as certified to IL5 and authorized for CUI, but *not* for PII or PHI ([DON CIO, Feb 2026](https://www.doncio.navy.mil/ContentView.aspx?id=20456)). And your command's policy wins. Check it before you use any AI tool for work, every time. ## Off duty and transitioning Transition comes with a mountain of paperwork, and this is where AI earns its keep. You can turn your notes from transition classes, networking calls, and informational interviews into follow-ups and action lists in a few minutes ([Issue #1](/issues/001/) shows exactly how). It also helps translate military experience into resume language a civilian hiring manager will get. There are free tools for this, like Hiring Our Heroes' Resume Engine ([VA News](https://news.va.gov/150284/turn-military-experience-civilian-ready-resume/)). Read every AI-written bullet before you use it. Check it's accurate, and take out anything sensitive. ## What's coming in this track Plain-English explainers on the AI rules and approved tools, always linked to the official source. Transition workflows for resumes, LinkedIn, interview prep, and researching certifications. And each weekly workflow gets its own safety note for anyone in uniform. We never use military seals, insignia, or service marks, and we don't imply any official connection. --- Source: https://aiexplainedplainly.com/issues/001/ # Issue #1: Before you paste anything into AI: a 10-minute safety setup, then turn messy notes into a clean action list *Issue #1, October 21, 2026* ## TL;DR - **What you get:** a privacy setup for ChatGPT, Gemini, Copilot, or Claude, plus one prompt that turns meeting notes into an action list. - **Time:** about 5–10 minutes for the setup, then 3–5 minutes per meeting. Both are estimates; we haven't timed them yet. - **Do this:** turn off chat training in your AI assistant today, then try the prompt on your next meeting. Hi there, This issue covers what's safe to paste into AI, and one job it's genuinely good at. Set aside 20–30 minutes. *(Estimate.)* --- ## Part 1: The 10-minute safety setup **Rule #1: if you wouldn't email it to a stranger, don't paste it into a free AI tool.** That means customer names and contact info, Social Security numbers, health details, passwords, account numbers, unreleased financials, and anything your employer marks confidential. Then turn off "use my chats for training." Menus as of October 2026: | Assistant | Where the setting is | Source | |---|---|---| | **ChatGPT** | Settings → Data controls → turn off **"Improve the model for everyone."** For sensitive one-offs, use **Temporary Chat** (OpenAI may keep those up to 30 days for safety). | [OpenAI Help](https://help.openai.com/en/articles/7730893-data-controls-in-chatgpt) | | **Gemini** | gemini.google.com → Settings & help → **Activity** → Turn off. Google may keep chats up to 72 hours. | [Google Help](https://support.google.com/gemini/answer/13278892) | | **Microsoft Copilot** (personal) | Profile → Privacy → turn off **"Training on conversation activity"** (and the voice one). | [Microsoft Support](https://support.microsoft.com/en-us/microsoft-copilot/microsoft-copilot-privacy-controls) | | **Claude** | Settings → Privacy → turn off **"Help improve our AI models."** | [Anthropic](https://privacy.claude.com/en/articles/12109829-how-do-i-change-my-model-improvement-privacy-settings) | Training off isn't *private*; chats can still be stored. If work gives you an approved AI (Copilot for business, ChatGPT Enterprise, Gemini for Workspace, GenAI.mil), **use that for work** and follow its policy over anything here. ⏱️ **Time:** about 5–10 minutes per assistant. *(Estimate.)* 💵 **Cost:** $0. --- ## Part 2: Messy notes into decisions, action items, and a follow-up email For anyone who leaves a meeting, call, or site visit with scribbles, then burns 20 minutes writing them up. Any free assistant above works; a phone voice memo with transcription helps. ### Steps 1. **Capture.** Dump what you remember into a note, or record a two-minute voice memo and copy the transcript. 2. **Scrub.** Swap names for roles ("Client A," "Ops lead"). Delete account numbers, prices you can't share, and anything confidential. About a minute; people skip it. 3. **Paste the prompt below**, then your scrubbed notes. 4. **Check the output against your memory.** AI sometimes invents deadlines or owners. Fix them. 5. **Send or file it.** Restore real names *after* it leaves the AI tool. ### Copy-paste prompt ``` You are my meeting assistant. Using ONLY the notes below, produce: 1. DECISIONS: bullet list of what was decided (max 5). 2. ACTION ITEMS: table with columns Task | Owner | Due date. If an owner or date is not in the notes, write "UNASSIGNED" or "NO DATE". Do not guess. 3. OPEN QUESTIONS: anything unresolved. 4. FOLLOW-UP EMAIL: under 150 words, friendly and plain, summarizing 1-3 for the attendees. If anything in the notes is ambiguous, list it under OPEN QUESTIONS instead of assuming. NOTES: [paste your scrubbed notes here] ``` "Do not guess" and UNASSIGNED / NO DATE target the most common failure: AI inventing who owns a task. ### Honest numbers | Item | Value | Basis | |---|---|---| | Setup time | 0 minutes beyond Part 1 | Nothing to install | | Time per use | about 3–5 minutes (scrub, paste, check) | **Estimate** until our test run (see Testing status) | | Time saved per use | maybe 10–20 minutes vs. writing it by hand | **Estimate**; depends on your write-ups | | Cost | $0 on free tiers; paid plans (ChatGPT Plus $20/mo, Go $8/mo in the US) aren't needed | [ChatGPT pricing](https://chatgpt.com/pricing/) | | Failure mode | Invented owners or dates; dropped items if notes are very long | Known AI limitation; Step 4 fixes it | **Try it on your next three meetings,** then reply with minutes saved (or not). We'll share anonymized reader numbers later. --- ## Your track **🧭 Leaders.** Before saying "use AI," write down the paste rule and name the approved tool. Next issue: a fill-in-the-blank AI policy template. **💼 Professionals.** Save the prompt as a text shortcut or "custom instruction" so every meeting gets the same clean format. **🎖️ Service members & veterans.** On duty, never paste classified information, CUI, PII/PHI, or unit or operational details into commercial AI tools. For official unclassified work, use your service's approved platform. In DoD that's **GenAI.mil**, authorized for CUI but *not* for PII or PHI ([DON CIO](https://www.doncio.navy.mil/ContentView.aspx?id=20456)). Your command's policy wins. Transitioning? Use this on notes from TAP classes, networking calls, and informational interviews. *AI Explained Plainly is not affiliated with, approved, or endorsed by the Department of Defense/Department of War, any military service, or the VA.* **🤖 Agents.** Markdown: `/issues/001.md`. JSON: `/api/workflows.json` (steps, prompt, time and cost labeled `measured`, `estimated`, or `vendor-published`). Index: `/llms.txt`. --- ## Testing status
Not measured yet. The time figures in this issue are estimates. Once we've done a real test run, we'll swap in the measured numbers and note the change here.
--- ## Sources & verified Checked Oct 21, 2026 (links inline): OpenAI data controls, Google Gemini activity, Microsoft Copilot privacy, Anthropic privacy, ChatGPT pricing, DON CIO on GenAI.mil. *Corrections: we fix errors on the site and tell you next issue.* --- Know a coworker who "doesn't really get AI"? Forward this. — AI Explained Plainly *How this is made: an AI editor drafts and fact-checks each issue, and the newsletter's owner approves each one before it's emailed. This web copy of Issue #1 is still a draft and hasn't been approved yet.* *Future issues may include affiliate links, always disclosed. This one has none.* --- Source: https://aiexplainedplainly.com/guides/what-is-generative-ai/ # What is generative AI? A plain-English explanation *Guide, October 21, 2026* ## TL;DR - Generative AI makes new text, images, or audio by predicting what should come next, based on patterns it learned from huge amounts of data. - It's great at first drafts, summaries, and rewording. It's bad at guaranteed facts, so check anything that matters. - Before you paste anything in, learn what's safe to share. Issue #1 covers the settings in about 10 minutes. You've probably heard "GenAI" in three meetings this month and nodded along. That's fine. The idea underneath it is simpler than the buzz makes it sound, and once you get it, you'll have a much better feel for when to trust these tools and when not to. ## The short version Generative AI is software that creates something new (a paragraph, an image, a summary, a line of code) when you ask it in plain language. ChatGPT, Claude, Gemini, and Microsoft Copilot are all generative AI tools. The "generative" part is the point. Older software mostly sorted, searched, or calculated things that already existed. This kind writes a fresh answer every time you ask. ## How it actually works (no math) Think about the autocomplete on your phone. You type "running late, be there in" and it suggests "10 minutes." It's guessing the next word based on what usually comes next. Generative AI is that same basic idea, scaled up enormously. The text versions are built on what's called a **large language model**, or LLM. "Large" because it learned from a massive amount of text. "Language model" because what it learned is how language tends to fit together. When you ask it a question, it isn't looking up a stored answer. It's building a response one small piece at a time, each time picking what's most likely to come next given your question and everything it's written so far. That's why the same question can get a slightly different answer twice. Image tools work on a similar principle with pictures instead of words. They learned what images tend to look like and generate a new one that fits your description. ## Why it sounds so confident (and why that's a problem) Here's the part most people miss. Because the model is predicting likely-sounding text, it can produce something that *sounds* completely right and isn't. A made-up date. A quote nobody said. A source that doesn't exist. This is usually called a **hallucination**. OpenAI's own help center says ChatGPT can sound confident even when it's wrong, and that it can produce fabricated quotes, studies, and citations. Their advice is to treat it as a first draft, not a final source. Many tools can now search the web and show links, which helps. But a link isn't proof. Click it and check that it says what the AI claims. ## What it's genuinely good at In my view, generative AI earns its keep on jobs where *you* already know what right looks like and just want a faster first pass: - Turning messy notes into a clean summary or action list - Drafting an email you'd otherwise stare at for 15 minutes - Rewording something to be shorter, friendlier, or more formal - Explaining a dense document or term in plain language - Brainstorming options you can then pick from Notice the pattern. You stay the judge. The AI does the typing. If you want to try one of these right now, [our guide to writing a professional email with AI](/guides/ai-professional-email/) has copy-paste prompts, and [Issue #1](/issues/001/) walks through turning meeting notes into an action list. ## What it's bad at It's weak at anything where being wrong is expensive and you can't easily check the answer. Legal conclusions, medical advice, exact numbers, and current facts all need a real source behind them. It also doesn't know your situation unless you tell it. It doesn't know your client, your boss's preferences, or what happened in last week's meeting. That's why the context you give it matters so much, and it's also why people end up pasting things they shouldn't. ## What not to paste The rule we use everywhere on this site: **if you wouldn't email it to a stranger, don't paste it into a free AI tool.** That covers customer names and contact details, Social Security numbers, health information, passwords, account numbers, unreleased financials, and anything your employer marks confidential. If your company gives you an approved work AI tool, use that one for work and follow its policy. If you serve, the bar is higher. Never paste classified information, CUI (Controlled Unclassified Information, the "sensitive but not classified" category), personal data, or unit and operational details into a commercial AI tool. The [Service Members & Veterans track](/service/) covers the approved options. *AI Explained Plainly is not affiliated with, approved, or endorsed by the Department of Defense/Department of War, any military service, or the VA.* [Issue #1](/issues/001/) shows exactly which setting to change in ChatGPT, Gemini, Copilot, and Claude so your chats aren't used for training. It's about 10 minutes per tool (that's an estimate). ## Generative AI vs. "regular" AI You'll hear both terms, so here's the quick difference. "AI" is the big umbrella. It includes things like spam filters, fraud detection, and the system that recommends your next show. Those mostly classify or predict. Generative AI is the slice that *creates* content. When someone at work says "we should be using AI," they almost always mean this slice, usually a chat tool. ## Which tool should you start with? Honestly, for learning what this stuff can do, the free version of almost any major tool is enough. If your company already pays for one (Copilot inside Microsoft 365, or Gemini inside Google Workspace, for example), start there, since it's the one you're allowed to use with work information. If you're choosing on your own, our [ChatGPT vs. Claude vs. Gemini vs. Copilot comparison](/guides/chatgpt-vs-claude-vs-gemini-vs-copilot/) breaks down who each one fits, with prices checked against the vendors' own pages. ## Try this today Pick one low-stakes thing you wrote this week, something with no names or private details. Paste it into a free AI tool with this: ``` Rewrite this so it's shorter and easier to read. Keep my meaning. Don't add any facts I didn't include. Then list anything you changed that I should double-check. [paste your text here] ``` Read what comes back with a skeptical eye. That one exercise will teach you more about how generative AI behaves than any explainer, this one included. ## Sources - OpenAI Help Center, [Does ChatGPT tell the truth?](https://help.openai.com/en/articles/8313428-does-chatgpt-tell-the-truth) (checked October 21, 2026) - AI Explained Plainly, [Issue #1: Before you paste anything into AI](/issues/001/) (privacy settings for each tool, with vendor links) --- Source: https://aiexplainedplainly.com/guides/ai-professional-email/ # How to use AI to write a professional email (with copy-paste prompts) *Guide, October 21, 2026* ## TL;DR - Give the AI the goal, the reader, and the key facts. It writes the draft; you fix the facts and the tone. - Copy the five prompts below for first drafts, hard replies, shortening, tone checks, and follow-ups. - Swap names and private details for placeholders before pasting, and put the real ones back after. Some emails take 30 seconds. Others sit in your drafts folder for an hour because you can't figure out how to say "no" to your boss's boss without sounding rude. That second kind is where AI actually helps. It won't know what you want to say. You still bring that. But it's very good at turning "here's roughly what I mean" into something clear and polite, fast. ## The one thing that makes AI emails good Most bad AI emails come from a one-line prompt like "write an email to my client about the delay." The AI fills the gaps with generic filler, and you get that stiff "I hope this email finds you well" tone everyone recognizes. Give it three things instead: 1. **The goal.** What should the reader do or know after reading? 2. **The reader.** Who are they to you, and how formal is the relationship? 3. **The facts.** The specific details, in rough bullet form. Typos are fine. That's it. The more specific your facts, the less the AI makes up. ## Prompt 1: First draft from rough notes Use this for most everyday emails. ``` Write a professional email using only the facts below. Goal: [what I want the reader to do or know] Reader: [who they are, e.g. "a client I've worked with for a year"] Tone: [friendly / neutral / formal] Length: under 150 words. Facts: - [fact 1] - [fact 2] - [fact 3] Rules: Don't add any facts, dates, or promises I didn't give you. Don't open with "I hope this email finds you well." Put the main point in the first two sentences. Give me a subject line too. ``` The "don't add facts" line matters more than it looks. AI tools will happily invent a delivery date or promise a refund if they think it makes the email sound complete. OpenAI's help center flatly says ChatGPT can produce incorrect information while sounding confident, so read every date and number before you hit send. ## Prompt 2: The reply you've been avoiding For pushing back, saying no, delivering bad news, or answering someone who's upset. ``` Help me reply to the email below. I want to [say no / push back on the deadline / apologize without admitting fault / ask for more budget]. My position, in plain words: [what you actually want to say] Relationship: [e.g. "my manager's manager" or "a frustrated customer"] Write it calm, direct, and respectful. No groveling, no corporate jargon. Offer one concrete next step. Keep it under 120 words. Their email: [paste the email, with names and private details removed] ``` When it comes back, check one thing: does it still actually say what you meant? AI tends to soften a "no" until it's a "maybe." If you meant no, make sure it says no. ## Prompt 3: Make it shorter You wrote a draft. It's too long. Most of us do this. ``` Cut this email by about half. Keep every fact, date, and request. Remove repetition and filler. Keep my voice. Then tell me in one line what you removed. [paste your draft] ``` ## Prompt 4: Tone check before you send For anything you wrote while annoyed. ``` Read this email the way the recipient would. Tell me: 1. How it's likely to come across, in one sentence. 2. Any line that could sound rude, passive-aggressive, or unclear. 3. A suggested fix for each of those lines. Don't rewrite the whole thing. [paste your draft] ``` I like this one because it leaves you in charge. You get a second opinion on specific lines instead of a full rewrite that no longer sounds like you. ## Prompt 5: Follow-up after a meeting ``` Write a short follow-up email to meeting attendees. Summarize decisions and action items from the notes below. Format action items as: Task, Owner, Due date. If an owner or date isn't in my notes, write "TBD". Do not guess. Under 150 words, friendly and plain. Notes: [paste your scrubbed notes] ``` If you do this a lot, [our meeting summary guide](/guides/ai-meeting-summary/) goes further, and [Issue #1](/issues/001/) has a longer version of this prompt. ## Make it sound like you, not a robot A few small habits help a lot: - Paste in one or two emails you've written that you like, and add: "Match the style of these examples." Remove any names or private details from them first. - Ban the phrases you hate. Add a line like "Don't use: 'I hope this finds you well,' 'circle back,' 'per my last email.'" - Read the draft out loud. If you wouldn't say a sentence to the person's face, change it. - Edit the first and last lines yourself. Those are the ones people remember. ## Which tool to use Any major chat tool can do this on its free plan. If you live in Outlook, Microsoft says Copilot is built into Outlook with Microsoft 365 Personal ($9.99/month in the US as of October 2026). If you live in Gmail, Google lists Gemini in Gmail as part of its Google AI plans starting with AI Plus ($4.99/month as of October 2026). But if your employer provides an approved AI tool, use that one for work email. Their policy beats anything here. For a side-by-side look at the big four, see [ChatGPT vs. Claude vs. Gemini vs. Copilot](/guides/chatgpt-vs-claude-vs-gemini-vs-copilot/). ## What not to paste Work email is full of things that shouldn't go into a free AI tool. Before you paste a thread or your notes: - Swap real names for roles: "Client A," "my manager," "Vendor." - Remove account numbers, contract values, pricing you can't share, and anything marked confidential. - Leave out health details, Social Security numbers, home addresses, and passwords. - Put the real details back *after* the draft is out of the AI tool. And turn off training on your chats. [Issue #1](/issues/001/) has the exact setting for ChatGPT, Gemini, Copilot, and Claude. **If you serve:** don't paste classified information, CUI, personal data, or unit and operational details into commercial AI tools, even for a "harmless" email. For official unclassified work, use your service's approved platform; Issue #1 has the details, and the [Service Members & Veterans track](/service/) has more. Your command's policy wins. *AI Explained Plainly is not affiliated with, approved, or endorsed by the Department of Defense/Department of War, any military service, or the VA.* ## Try it on the email you're dreading You probably have one sitting in your drafts right now. Scrub the names, grab Prompt 2, and see what comes back. Then fix the facts, fix the tone, and send it yourself. ## Sources - OpenAI Help Center, [Does ChatGPT tell the truth?](https://help.openai.com/en/articles/8313428-does-chatgpt-tell-the-truth) (checked October 21, 2026) - Microsoft, [Copilot pricing plans for individuals](https://www.microsoft.com/en-us/copilot/pricing/individuals) (Microsoft 365 Personal price and Copilot in Outlook, checked October 21, 2026) - Google One, [Google AI plans](https://one.google.com/about/google-ai-plans/) (AI Plus price and Gemini in Gmail, checked October 21, 2026) - AI Explained Plainly, [Issue #1](/issues/001/) (privacy settings with vendor links; GenAI.mil guidance sourced to DON CIO) --- Source: https://aiexplainedplainly.com/guides/ai-meeting-summary/ # How to use AI to summarize a meeting (safely) *Guide, October 21, 2026* ## TL;DR - Paste scrubbed notes or a transcript with a strict prompt, and get decisions, action items, and open questions back. - Tell the AI not to guess owners or dates. Then check its output against what was actually said. - Get consent before recording, use your company's approved tool for work meetings, and strip names before using a free one. The meeting ends. Everyone agrees "someone will send notes." Then nobody does, or it's you, at 6 PM, trying to remember who said they'd own the budget piece. AI is genuinely good at this job. It's also one of the easiest ways to accidentally paste something private into a tool you don't control. So this guide covers both: how to get a clean summary, and how to do it without creating a new problem. ## Three ways to feed it the meeting **Your own notes.** Messy is fine. Bullets, half-sentences, a voice memo you dictated in the car. This is the lowest-risk option because you've already filtered out most of the sensitive stuff just by not writing it down. [Issue #1](/issues/001/) walks through this exact workflow step by step. **A transcript.** Many meeting platforms and phone voice-memo apps can produce one. You get far more detail, but also everything anyone said, including the off-hand comment about a client or a coworker's medical appointment. **A built-in meeting assistant.** Some work platforms take notes for you. Google, for example, lists "Take notes for me in Google Meet" (English only) as a feature of its Google AI plans. If your company has turned on a tool like this, it's usually the safest choice for work meetings, since it runs under your employer's agreement and settings. ## Before you record anyone Recording a meeting to get a transcript isn't just a courtesy question. It can be a legal one. U.S. federal law allows recording if one party to the conversation consents, but some states require consent from everyone involved, according to the Digital Media Law Project. California, Florida, and Pennsylvania are among them. When people are dialing in from different states, it's hard to know which law applies, so the safe habit is simple: say "I'm recording this so I can write up notes, any objection?" at the start, every time. Most platforms also show a recording notice. Don't rely on that alone. And check your company's policy, which may be stricter than the law. ## The prompt for notes If you're working from your own notes, use the prompt in [Issue #1](/issues/001/). It's built to stop the AI from inventing owners and deadlines. ## The prompt for a transcript Transcripts are long and rambling, so the prompt needs to work harder at separating real decisions from people thinking out loud. ``` You are summarizing a meeting transcript. Use ONLY what's in the transcript. Give me: 1. SUMMARY: 3-5 sentences on what the meeting was about and where it landed. 2. DECISIONS: only things the group clearly agreed to. If something was discussed but not settled, it is NOT a decision. 3. ACTION ITEMS: a table with Task | Owner | Due date. Only list an owner if someone explicitly took or was given the task. Otherwise write "UNASSIGNED". If no date was said, write "NO DATE". 4. OPEN QUESTIONS: anything left unresolved or needing follow-up. 5. QUOTES TO VERIFY: for each decision and action item, include the short line from the transcript that supports it, so I can check it. Do not guess. If something is unclear, put it under OPEN QUESTIONS. TRANSCRIPT: [paste the scrubbed transcript] ``` That last section, "quotes to verify," is the useful part. It forces the AI to point at the line it based each item on. If it can't find one, that's your cue that it made something up. ## Check it like you'd check a new hire's work Even with a strict prompt, AI summaries go wrong in predictable ways: - **Invented owners.** Someone said "we should look into that," and now Dana owns it. - **Upgraded decisions.** A "let's think about it" becomes "the team agreed to." - **Dropped items.** Long transcripts can lose the thing mentioned in minute 50. - **Transcript errors.** Speech-to-text mishears names and numbers, and the AI repeats the mistake confidently. Spend two minutes reading the action items against your memory or the quotes. Fix anything that's wrong *before* it goes to the group. Once a wrong owner is in an email, it tends to stick. ## Turn it into a follow-up email Once the summary's right, ask for the email in the same chat: ``` Now write a short follow-up email to the attendees based on the corrected summary above. Friendly and plain, under 150 words. List action items as Task, Owner, Due date. ``` Want more control over the tone? [Our professional email guide](/guides/ai-professional-email/) has prompts for that. ## What not to paste Transcripts are where people slip up, because nobody reads all 40 minutes before pasting. Before you put a meeting into a free AI tool: - Search the text for names and replace them with roles ("Client A," "Finance lead"). - Delete account numbers, contract figures, salaries, and anything about HR issues or someone's health. - Leave out unreleased plans, pricing, and anything marked confidential. - Turn off training on your chats. [Issue #1](/issues/001/) has the setting for ChatGPT, Gemini, Copilot, and Claude. If it's a work meeting and your employer offers an approved AI tool, use that one instead. It's the right call even when a free tool would be faster. **If you serve:** don't record, transcribe, or summarize briefings, unit meetings, or anything with operational details using commercial AI or personal phone apps. No classified information, CUI, or personal data goes into those tools. For official unclassified work, use your service's approved platform and follow your command's policy. The [Service Members & Veterans track](/service/) has more. *AI Explained Plainly is not affiliated with, approved, or endorsed by the Department of Defense/Department of War, any military service, or the VA.* ## Which tool works best? Any of the big four can handle a summary on its free plan. Very long transcripts can bump into usage limits on free tiers, which is one of the few reasons a paid plan might make sense here. If you're weighing that, see [Free ChatGPT vs. ChatGPT Plus](/guides/free-chatgpt-vs-plus/). ## Try it on your next three meetings Use your notes for the first one, since that's the safest way to start. Check every action item against what was said. By the third meeting you'll know whether it's actually saving you time, and that's the only number that matters here. ## Sources - Digital Media Law Project, [Recording Phone Calls and Conversations](https://www.dmlp.org/legal-guide/recording-phone-calls-and-conversations) (federal one-party rule and all-party consent states, checked October 21, 2026) - Google One, [Google AI plans](https://one.google.com/about/google-ai-plans/) ("Take notes for me in Google Meet," checked October 21, 2026) - AI Explained Plainly, [Issue #1](/issues/001/) (notes prompt and privacy settings with vendor links) --- Source: https://aiexplainedplainly.com/guides/chatgpt-vs-claude-vs-gemini-vs-copilot/ # ChatGPT vs. Claude vs. Gemini vs. Copilot: which should you use? *Guide, October 21, 2026* ## TL;DR - All four have useful free plans. Pick based on where you already work: Microsoft, Google, or neither. - Main paid plans run about $20 a month. Gemini and ChatGPT also have cheaper tiers ($4.99 and $8). - If your employer provides one, use that for work. It beats any personal plan on privacy. If you're asking "which AI is best," you'll find a hundred rankings that disagree with each other. Here's the more useful question: which one fits the way you already work? All four are capable general-purpose assistants. They can all draft emails, summarize documents, explain things, and brainstorm. The real differences come down to which apps they plug into, what the paid plans include, and what your employer allows. One note on honesty before we start. We haven't run our own side-by-side quality tests, so this guide doesn't claim one writes "better" than another. Everything below comes from the vendors' own pricing and help pages, checked on October 21, 2026. Plans change often, so check the linked pages before you buy. ## The quick answer - **You work in Microsoft 365 (Outlook, Word, Excel, Teams):** start with **Copilot**. - **You work in Gmail and Google Docs:** start with **Gemini**. - **You don't live in either, or want a standalone assistant:** try **ChatGPT** or **Claude** free for a week each and keep the one you reach for. - **Your employer gives you one:** use that one for work. Full stop. ## Prices at a glance (US, as of October 2026) | Assistant | Free plan | Lower-cost tier | Main paid plan | Higher tiers | |---|---|---|---|---| | **ChatGPT** | Yes | Go: $8/mo | Plus: $20/mo | Pro, Business, Enterprise | | **Claude** | Yes | none | Pro: $20/mo monthly, or $17/mo billed annually ($200 up front) | Max from $100/mo; Team; Enterprise | | **Gemini** | Yes | Google AI Plus: $4.99/mo | Google AI Pro: $19.99/mo | Google AI Ultra from $99.99/mo | | **Copilot** | Yes (with limits) | Microsoft 365 Personal: $9.99/mo | Microsoft 365 Premium: $19.99/mo or $199.99/yr | Microsoft 365 Pro | Prices are from each vendor's official page (links in Sources) and don't include tax. One thing to know about the cheaper ChatGPT tiers: OpenAI's pricing page says the Go plan may include ads, and OpenAI has said Plus and higher stay ad-free. ## ChatGPT ChatGPT, from OpenAI, is the one most people have heard of, and it's the general-purpose default for a lot of users. Per OpenAI's pricing page, the free plan includes unlimited text chats (subject to abuse guardrails), web search, and file and image uploads, with limits on uploads, image creation, voice, and deep research. Plus adds advanced reasoning models, higher limits, and features like projects, scheduled tasks, and custom GPTs. We cover that choice in detail in [Free ChatGPT vs. ChatGPT Plus](/guides/free-chatgpt-vs-plus/). **Good fit if:** you want one standalone assistant for a bit of everything and don't need it built into Office or Google apps. ## Claude Claude, from Anthropic, is a standalone assistant like ChatGPT. Anthropic's pricing page lists a free plan that includes web search, file creation, memory across conversations, and up to five projects for organizing your work. Pro adds more usage, full Projects (Free caps you at five), more Claude models, Claude in Chrome, and Claude for Microsoft 365 and Outlook. Anthropic also says plainly that Pro usage is at least five times Free's per five-hour window, and that every plan has usage limits. One detail worth knowing: Anthropic's pricing page lists model training as "opt-out" on Free, Pro, and Max, and "none by default" on Team and Enterprise. So on a personal plan, you need to turn it off yourself. [Issue #1](/issues/001/) shows where. **Good fit if:** you want a standalone assistant and are comfortable comparing it against ChatGPT for your own work. ## Gemini Gemini is Google's assistant. The free plan, per Google, includes image generation, Deep Research, Gemini Live (voice conversation), Canvas, and Gems (saved custom helpers). The paid Google AI plans are bundled with Google One storage. AI Plus ($4.99/mo) adds Gemini in Gmail and 400 GB of storage. AI Pro ($19.99/mo) adds Gemini in Docs and Sheets, a much larger context window (how much text it can work with at once), and 5 TB of storage. Watch out for one thing: Google says only *personal* Google accounts can buy these plans. If your work email runs on Google Workspace, your company has to add Gemini through Workspace instead. **Good fit if:** your day lives in Gmail, Docs, and Drive, or you already pay for Google storage. ## Copilot Microsoft Copilot comes in two forms. There's a free Copilot app, which Microsoft describes as a limited version subject to capacity. Then there's Copilot inside Word, Excel, PowerPoint, and Outlook, which comes with Microsoft 365 subscriptions. Microsoft 365 Personal ($9.99/mo) includes those desktop apps with Copilot and 1 TB of storage. Premium ($19.99/mo) raises the usage limits and adds features Microsoft lists as exclusive to Premium and Pro. On Family and Premium, Microsoft says the AI features are for the subscription owner only, not the people you share with. **Good fit if:** you already pay for Microsoft 365 or spend your day in Outlook and Word. ## Which should *you* pick? My suggestion: don't buy anything yet. Use the free version of the one that matches your apps for a week, on real (scrubbed) tasks. Try our [email prompts](/guides/ai-professional-email/) or the meeting-notes workflow in [Issue #1](/issues/001/). Upgrade only when you keep hitting a limit that actually slows you down. And if you're choosing for a team, the paid business versions (ChatGPT Business, Claude Team, Gemini through Google Workspace, or Microsoft's business Copilot plans) matter more than the personal plans. They come with admin controls and different data terms. The [Leaders track](/leaders/) covers how to set that up. ## What not to paste Whichever you pick, the rule's the same: if you wouldn't email it to a stranger, don't paste it into a personal AI account. No customer details, health information, passwords, account numbers, or anything confidential. Turn off training in your settings (see [Issue #1](/issues/001/)), and use your employer's approved tool for work. ## Try this Pick one task you do every week. Run it through the free version of two of these tools with the same prompt, and keep whichever answer you'd actually send. That's a better test than any ranking, including ours. ## Sources All checked October 21, 2026. Prices are US and exclude tax. - OpenAI, [ChatGPT plans and pricing](https://chatgpt.com/pricing/) and [What is ChatGPT Plus?](https://help.openai.com/en/articles/6950777-what-is-the-difference-between-the-free-and-paid-versions-of-chatgpt) - OpenAI, [Introducing ChatGPT Go](https://openai.com/index/introducing-chatgpt-go/) (Go's US price; Plus and higher remain ad-free) - Anthropic, [Claude pricing](https://claude.com/pricing) - Google, [Gemini subscriptions](https://gemini.google/subscriptions/) and [Google AI plans on Google One](https://one.google.com/about/google-ai-plans/) - Microsoft, [Copilot pricing plans for individuals](https://www.microsoft.com/en-us/copilot/pricing/individuals) and [Compare Microsoft 365 plans](https://www.microsoft.com/en/microsoft-365/buy/compare-all-microsoft-365-products?tab=1) (Premium price) --- Source: https://aiexplainedplainly.com/guides/free-chatgpt-vs-plus/ # Free ChatGPT vs. ChatGPT Plus: is it worth $20 a month? *Guide, October 21, 2026* ## TL;DR - ChatGPT Plus is $20 a month, billed monthly. It adds stronger reasoning models, higher limits, and extras like projects and custom GPTs. - Free covers emails, summaries, and everyday questions. Upgrade only if you keep hitting limits that slow you down. - Plus doesn't make your chats more private. Turn off training in settings either way. Twenty dollars a month isn't nothing, especially when the free version already does a lot. So here's the straight answer: most people who use ChatGPT a few times a week for emails and quick questions don't need Plus. Some people really do. The trick is knowing which one you are. Everything below comes from OpenAI's own pricing and help pages, checked October 21, 2026. OpenAI changes plans and models often, so check [chatgpt.com/pricing](https://chatgpt.com/pricing/) before you pay. ## What you get for free OpenAI's pricing page lists these for the free plan: - Unlimited text chats (OpenAI notes this is subject to abuse guardrails) - Web search and file and image uploads, with limits on how many messages can use uploads - Limited image creation, voice chats, and deep research - Limited memory and context That's enough for the everyday stuff: drafting an email, summarizing a document, explaining a term, brainstorming. The prompts in our [professional email guide](/guides/ai-professional-email/) and [meeting summary guide](/guides/ai-meeting-summary/) work fine on the free plan. One more thing: OpenAI has said it's testing ads in the free tier and the Go plan in the US. ## What Plus adds Plus costs **$20 a month**, billed monthly. OpenAI's help center says there's no annual option right now. According to OpenAI, Plus adds: - **Stronger reasoning.** Access to OpenAI's advanced reasoning models. OpenAI's model help page says Free and Go users get a default model and don't get the higher "thinking" levels, while Plus includes the Medium and High levels. - **Higher limits.** More messages, larger and more frequent uploads, more image creation, and expanded deep research. - **Priority and speed.** OpenAI lists priority access during busy periods and faster responses. - **Extras for organizing work.** Projects, scheduled tasks, and custom GPTs are on the Plus list and not on the Free list. - **No ads.** OpenAI has said Plus, Pro, Business, and Enterprise will remain ad-free. Plus still has usage limits. OpenAI says message caps can apply, especially during high demand. And it doesn't include API access, which is billed separately. ## The middle option: Go There's also **ChatGPT Go at $8 a month** in the US. Per OpenAI, Go gives you more messages with tools, more uploads, more image creation, more voice chats, and longer memory than Free. But it doesn't get the higher reasoning levels, and OpenAI's pricing page notes it may include ads. If your only complaint about Free is running out of uploads or image generations, Go might be enough. ## Who should upgrade Plus probably makes sense if you: - Hit the free limits regularly and it interrupts real work - Need the stronger reasoning for things like multi-step analysis, planning, or working through long documents - Want projects to keep ongoing work organized, or want to build a custom GPT for a repeated task - Use deep research often enough that the free allowance runs out - Want ChatGPT without ads ## Who should stay on Free You probably don't need Plus if you: - Use ChatGPT a few times a week for emails, rewording, and quick questions - Rarely or never see a limit message - Have access to an AI tool at work that your employer pays for (use that one for work anyway) - Mostly want to learn what AI can do. Free is plenty for that. My suggestion: stay on Free until you hit a limit that actually costs you time. Then try Plus for one month. OpenAI says you can cancel any time from Settings → Account, so you're not locked in. ## Does Plus make your chats more private? No. This is the misunderstanding that matters most. Paying for Plus doesn't change what you should paste. OpenAI's help center says it may use conversations to improve its models, and that you can opt out. That opt-out is available on both Free and Plus, and you have to turn it on yourself: Settings → Data controls → turn off "Improve the model for everyone." [Issue #1](/issues/001/) walks through it, along with Temporary Chat for one-off sensitive tasks. One catch: OpenAI says that if you rate a response with thumbs up or down, that whole conversation may be used for training even after you opt out. If you need business-grade data terms, a personal Plus subscription isn't it. OpenAI says it doesn't use content from ChatGPT Business or Enterprise workspaces for training by default, and the tool your company already provides beats both. ## What not to paste (on any plan) Same rule, free or paid: if you wouldn't email it to a stranger, don't paste it. No customer names, health details, Social Security numbers, passwords, account numbers, or anything your employer marks confidential. If you serve, no classified information, CUI, personal data, or unit and operational details go into ChatGPT on any plan. The [Service Members & Veterans track](/service/) covers what's approved. *AI Explained Plainly is not affiliated with, approved, or endorsed by the Department of Defense/Department of War, any military service, or the VA.* ## Still deciding between tools, not plans? If you're not sure ChatGPT is even the right assistant for you, read [ChatGPT vs. Claude vs. Gemini vs. Copilot](/guides/chatgpt-vs-claude-vs-gemini-vs-copilot/) first. Claude Pro, Google AI Pro, and Microsoft 365 Premium all sit around the same $20 price point, and the right pick depends on which apps you already use. ## Try this before you pay For the next two weeks, every time you see a limit message on Free, jot down what you were doing. If the list is short or empty, keep your $20. If it's long and it's real work, Plus will probably pay for itself. ## Sources All checked October 21, 2026. Prices are US and exclude tax. - OpenAI, [ChatGPT plans and pricing](https://chatgpt.com/pricing/) - OpenAI Help Center, [What is ChatGPT Plus?](https://help.openai.com/en/articles/6950777-what-is-the-difference-between-the-free-and-paid-versions-of-chatgpt) (price, billing, benefits, cancellation, data opt-out) - OpenAI Help Center, [GPT-6 and other models in ChatGPT](https://help.openai.com/en/articles/20001354-gpt-6-and-other-models-in-chatgpt) (thinking levels by plan) - OpenAI, [Introducing ChatGPT Go](https://openai.com/index/introducing-chatgpt-go/) (Go price; ads testing in Free and Go; Plus and higher remain ad-free) - OpenAI Help Center, [Data controls in ChatGPT](https://help.openai.com/en/articles/7730893-data-controls-in-chatgpt) (training setting, Temporary Chat, feedback caveat, and business workspace defaults) - Anthropic, [Claude pricing](https://claude.com/pricing); Google One, [Google AI plans](https://one.google.com/about/google-ai-plans/); Microsoft, [Compare Microsoft 365 plans](https://www.microsoft.com/en/microsoft-365/buy/compare-all-microsoft-365-products?tab=1) --- Source: https://aiexplainedplainly.com/guides/can-i-use-chatgpt-in-the-military/ # Can I use ChatGPT in the military? What's allowed and what's not *Guide, October 21, 2026* ## TL;DR - Commercial ChatGPT is fine for personal, public, non-work stuff. Never paste classified info, CUI, PII, or unit or operational details. - For official unclassified work, use GenAI.mil. It's authorized for CUI, but not for PII or PHI. - Your command's policy wins over anything here, so check it before using any AI tool for work. Short answer: yes, for some things, and absolutely not for others. The line isn't really about ChatGPT. It's about what you type into it and whether the task is official business. Here's how it breaks down as of October 2026, with links to the official sources so you can check them yourself. ## The one rule that covers most of it Never paste classified information, CUI, PII or PHI, or unit and operational details into a commercial AI tool. That includes ChatGPT, Gemini, Claude, Copilot, and any app you downloaded yourself. Why so strict? Whatever you type into a commercial chatbot leaves your hands. It's stored on the company's servers, and depending on your settings it may be reviewed or used to train future models. The Department of the Navy's CIO put it bluntly back in 2023: these tools save the prompts they're given, which "presents a persistent Operational Security (OPSEC) risk." The same memo warns that lots of small, harmless-looking prompts can add up to something sensitive. Quick definitions, since these get thrown around a lot: - **Classified:** anything marked Confidential, Secret, or Top Secret. This never goes near a commercial tool, or any tool not accredited for it. - **CUI (Controlled Unclassified Information):** information the government has to protect under law, regulation, or policy, even though it isn't classified. The National Archives runs the program. Think personnel records, a lot of contracting info, law enforcement reports, and plenty of routine unit paperwork. - **PII and PHI:** personally identifiable information (names tied to SSNs, DoD ID numbers, home addresses) and protected health information (medical details). - **Unit and operational details:** locations, movements, schedules, readiness, capabilities, who's deploying when. Even if none of it is marked, it's an OPSEC problem. DoD's CUI instruction, DoDI 5200.48, requires CUI work to be done only on DoD or approved contractor IT systems. A commercial chatbot on your phone is neither. ## Where ChatGPT is fine Off duty, on your own device, with nothing work-related in it, commercial AI is just another app. Reasonable uses include: - Planning a trip, a workout, or a budget - Studying a public topic, like a certification you're working toward - Drafting a personal email or a cover letter (with your own details scrubbed) - Practicing for a civilian job interview - Getting a plain-English explanation of a public regulation or benefit, then checking it against the official source A good gut check: would you be comfortable reading your prompt out loud in a crowded airport? If yes, it's probably fine. If you'd lower your voice, don't paste it. I'd also turn off chat training in whatever assistant you use. [Issue #1](/issues/001/) walks through where that setting lives in ChatGPT, Gemini, Copilot, and Claude. It doesn't make your chats private, but it's a sensible default. ## For official work, use the approved platform For work tasks, DoD has its own tool. GenAI.mil launched on December 9, 2025. The Secretary of War's memo that day says it's available to military personnel, civilians, and contractors, you log in with your CAC, it runs on NIPR, and it's "certified up to Impact Level 5 (IL5), and is fully authorized to handle CUI." (IL5 is a DoD cloud security level for more sensitive unclassified data.) Google's Gemini for Government was the first model on it. One catch that trips people up. Being approved for CUI doesn't mean anything goes. A January 2026 Department of the Navy memo designating GenAI.mil as the Navy's CUI/IL5 AI platform says plainly that PII and PHI "are not permitted for use on this platform." So even on GenAI.mil, keep names, SSNs, DoD ID numbers, and medical details out. And GenAI.mil runs on NIPR, so classified material doesn't belong there either. Your service may also have its own tools or rules on top of this. The Navy memo directed all Navy organizations to transition to GenAI.mil by April 30, 2026, for example. Ask your S-6, N6, or comm shop what's approved where you are. ## What the services actually say A few official lines worth knowing: - **Army:** the Army CIO's 2024 guidance says users "have sole responsibility for their input" and "should have no expectations of privacy with respect to those inputs." It also says to "distrust and verify all outputs prior to use," and that you should label documents made in whole or in part with AI output. - **Navy:** the 2023 DON CIO guardrail memo says commercial AI language models aren't recommended for operational use cases. That's on top of the CUI rules. - **Everyone:** the CUI instruction applies no matter which service you're in. None of this is a ban on AI. The Army guidance actually discourages commands from banning it outright and asks them to set up proper governance instead. The point is to use the right tool for the right data. ## A quick decision check Before you paste anything, run through this: ``` 1. Is this for official work? No -> Commercial AI is OK if there's nothing sensitive in it. Yes -> Go to 2. 2. Does it include anything classified? Yes -> Stop. Not in any of these tools. 3. Does it include CUI? Yes -> Approved platform only (GenAI.mil or what your command authorizes). 4. Does it include PII or PHI (names + IDs, SSNs, medical info)? Yes -> Take it out first. GenAI.mil isn't authorized for it either. 5. Does your command have its own AI policy? Yes -> Follow it. It wins. ``` ## AI makes things up, so you're still on the hook Every one of these tools can produce confident, wrong answers. The Army calls them "hallucinations" in its guidance and says AI output needs a review process with real human judgment behind it. If you use AI to help draft a memo or summarize a regulation, you own what you sign. Check every fact, date, and reference against the actual source before it goes anywhere. That's also why I'd never use AI as the last word on a regulation. Use it to understand the gist, then read the real paragraph. ## Where to go from here If you're writing awards or evals, read [how to use AI to write award bullets safely](/guides/ai-award-bullets/). If you're getting out, [using AI to turn military experience into a civilian resume](/guides/ai-military-to-civilian-resume/) is the next stop. And the [Service & Veterans track](/service/) collects everything else for people in uniform. Before your next work task, take 30 seconds and ask your leadership which AI tool is approved at your unit. That one question settles most of this. ## Sources - Secretary of War memo, "Harness Artificial Intelligence Now with GenAI," Dec. 9, 2025 ([DocumentCloud copy](https://s3.documentcloud.org/documents/26367576/memo-on-integrating-genai-mil.pdf)) and the [Department of War release on GenAI.mil](https://www.war.gov/News/Releases/Release/Article/4354916/the-war-department-unleashes-ai-on-new-genaimil-platform/), Dec. 9, 2025 - [ASN(RD&A) and DON CIO joint memo designating GenAI.mil](https://www.doncio.navy.mil/FileHandler.ashx?id=24946), Jan. 28, 2026 - [DON CIO interim guidance on generative AI and large language models](https://www.doncio.navy.mil/ContentView.aspx?id=16442), Sept. 6, 2023 - [Army CIO guidance on generative AI and large language models (ADS-GOV-AI-024)](https://armypubs.army.mil/epubs/DR_pubs/DR_a/ARN41285-PPM_CIO-024-000-WEB-1.pdf), June 27, 2024 - [Air Force JAG Corps, "Large Language Models"](https://www.afjag.af.mil/Post/Article-Display/Article/4314500/large-language-models/), Aug. 7, 2025 (on DoDI 5200.48 and CUI) - [National Archives, About CUI](https://www.archives.gov/cui/about) *Checked against these sources on Oct. 21, 2026. Policies change, so confirm with your command. How this is made: an AI editor drafts and fact-checks each article. If we get something wrong, we fix it and note the change.* --- Source: https://aiexplainedplainly.com/guides/ai-award-bullets/ # How to use AI to write award bullets (safely) *Guide, October 21, 2026* ## TL;DR - AI is good at rewording your own notes into tight action-impact bullets. It can't know what actually happened, so every fact comes from you. - Scrub names, IDs, unit names, locations, and anything operational first. Use GenAI.mil for official work, and keep PII out there too. - Check every number and claim before it goes up the chain. Your service's awards guidance and your admin shop's format win. Award write-ups have a way of landing at the worst time. You know the person did great work, you've got a pile of notes, and turning that into tight, impact-first bullets takes longer than it should. That rewording job is where AI actually helps. Facts are where it doesn't. It wasn't there, so it can't know what happened. And the data you'd need to describe it is often exactly the stuff you can't paste. ## First, pick the right tool If the write-up is official work (and an award recommendation is), use the AI platform your command has approved. For most of DoD that's GenAI.mil, which you reach with your CAC on NIPR. It's authorized for CUI. But a January 2026 Department of the Navy memo is clear that PII and PHI "are not permitted for use on this platform." So the scrubbing step below still applies there. Don't paste award material into personal ChatGPT, Gemini, or Claude. The Army CIO's guidance says users "should have no expectations of privacy" with what they put into these tools, and DoD's CUI rules require CUI to stay on DoD or approved contractor systems. Lots of award details are CUI, or close enough that you shouldn't gamble. Not sure what's approved at your unit? Ask. [Can I use ChatGPT in the military?](/guides/can-i-use-chatgpt-in-the-military/) covers the bigger picture. ## Scrub your notes before AI sees them This takes a couple of minutes, and it's the step that matters most. Go through your notes and take out or replace: - Names (yours, the member's, anyone else's). Use "the member" or "SM." - DoD ID numbers, SSNs, dates of birth, and contact info - Unit names, ship or squadron names, and locations - Mission names, operation names, dates and times of anything operational - Capabilities, readiness numbers, equipment counts, and vulnerabilities - Medical details, family situations, or anything from a counseling What's left should read like a generic job description with real results in it. Here's a made-up example of the shape: "Led a 6-person team. Rebuilt the parts tracking process. Cut backorders from about 40 to 12 over 3 months. Trained 4 junior members who now run it." That's plenty for AI to work with. If a number would be sensitive on its own, describe it as a change instead. "Cut processing time roughly in half" says the same thing without giving anything away. ## The copy-paste prompt Paste this, then your scrubbed notes underneath: ``` You're helping me draft award bullets from my own notes. Rules: - Use ONLY the facts in my notes. Do not add numbers, results, or scope that aren't there. - If a bullet needs a result I didn't give, write [NEED RESULT] instead of guessing. - Format each bullet as: strong action verb, what was done, then the impact or result. - Keep each bullet to one line, about 20 words or fewer. - No acronyms unless they're in my notes. - Give me 3 versions of each bullet: plain, punchy, and formal. Then list any claims in my notes that sound vague and suggest one question I should answer to make each one stronger. NOTES: [paste your scrubbed notes here] ``` The [NEED RESULT] rule matters most. Ask a chatbot for "impact" and it'll happily make some up: a percentage, a dollar figure, a mission that was "enabled." That's how a made-up claim ends up in an official document with your name on the recommendation. ## Make it fit your service's format Award formats and limits differ by service and by award, and they change. I'm not going to guess at character counts here. Grab the current template or example from your admin shop, then use a follow-up like this: ``` Here is the required format and length limit for this award: [paste the format rules or a blank template, no names] Rewrite the bullets above to fit it exactly. If something has to be cut to fit, tell me what you cut. ``` Your service's awards guidance and your admin shop's format win over anything AI produces. If AI's output and the template disagree, the template is right. ## Check every bullet before it goes anywhere Read each bullet against your notes, line by line. Ask yourself: 1. Did this actually happen, exactly as written? 2. Is every number from my notes, not from the AI? 3. Would the member's supervisor sign off on every claim? 4. Is there anything here that's now *more* sensitive than my notes were? Sometimes AI adds context that sounds like a location or mission. 5. Does it sound like a person wrote it? Cut words like "spearheaded synergistic" if nobody in your unit talks that way. Then put the real names and unit info back in yourself, outside the AI tool, in the actual form. The Army's AI guidance says you should label documents created in whole or in part with AI output. Check whether your service or command has a similar rule and follow it. ## Where AI helps most (and where it doesn't) It's genuinely useful for: - Turning a messy paragraph into a clean action-impact bullet - Swapping weak verbs ("helped with," "was responsible for") for stronger ones - Trimming a bullet that's two words over the limit - Giving you a few phrasings when you're stuck on the same one It's no help with knowing what happened, judging whether something rises to the level of a given award, or deciding what's sensitive. That's on you and the chain. ## Keep the notes going all year The best award bullets come from notes written when things happen, not reconstructed months later. A simple habit: keep a running unclassified log of wins for yourself and your people, with dates and rough numbers, and no names or details you couldn't paste. When award season or eval time comes around, you already have clean input. That same log works for your own transition later, too. [Translating military experience into a civilian resume](/guides/ai-military-to-civilian-resume/) uses the exact same raw material. More for people in uniform is on the [Service & Veterans track](/service/). Start the log today. One line per win is enough. ## Sources - Secretary of War memo, "Harness Artificial Intelligence Now with GenAI," Dec. 9, 2025 ([DocumentCloud copy](https://s3.documentcloud.org/documents/26367576/memo-on-integrating-genai-mil.pdf)) - [ASN(RD&A) and DON CIO joint memo designating GenAI.mil](https://www.doncio.navy.mil/FileHandler.ashx?id=24946), Jan. 28, 2026 (PII and PHI not permitted) - [Army CIO guidance on generative AI and large language models (ADS-GOV-AI-024)](https://armypubs.army.mil/epubs/DR_pubs/DR_a/ARN41285-PPM_CIO-024-000-WEB-1.pdf), June 27, 2024 - [Air Force JAG Corps, "Large Language Models"](https://www.afjag.af.mil/Post/Article-Display/Article/4314500/large-language-models/), Aug. 7, 2025 (on DoDI 5200.48 and CUI) *Checked against these sources on Oct. 21, 2026. Policies change, so confirm with your command. How this is made: an AI editor drafts and fact-checks each article. If we get something wrong, we fix it and note the change.* --- Source: https://aiexplainedplainly.com/guides/ai-military-to-civilian-resume/ # How to use AI to translate military experience into a civilian resume *Guide, October 21, 2026* ## TL;DR - Start with your VMET, evals, and awards, then have AI translate jargon into civilian terms using only what you give it. - Scrub your SSN, DoD ID, unit names, locations, and anything classified or operational before pasting anything. - Check every line is true and something you can explain in an interview. AI drafts it. You own it. You did the work. The problem is the vocabulary. "NCOIC," "maintained 100% accountability of sensitive items," a stack of acronyms that mean everything to you and nothing to a recruiter at a logistics company. AI is good at exactly this kind of translation. It's bad at knowing what you actually did, and it'll cheerfully pad your resume with things you didn't. So the trick is to feed it real material, tell it not to invent anything, and check every line. ## Gather your raw material first AI can only translate what you give it. Before you open a chatbot, pull together: - **Your VMET (DD Form 2586).** You get it in milConnect under Correspondence/Documentation, then DoD Transition Assistance Program, then the VMET tab. DMDC says it describes your service occupations in civilian terms and is meant to help with resumes. Two caveats from the same FAQ: the descriptions are standard ones that may not match what you actually did, and the Coast Guard doesn't submit data, so Coast Guard members can't pull one. - **Your Joint Services Transcript**, if your service uses it, for courses and training. - **Your evals and award write-ups.** These are usually where your real results live. - **The job posting** you're aiming at. One specific posting beats a general idea of a field. DMDC itself recommends combining VMET with evaluations, training certificates, awards, and transcripts to get the full picture. That's good advice for AI input too. ## Scrub it before you paste it Your resume is personal, so a commercial tool like ChatGPT or Claude is a reasonable choice here, on your own device and your own time. But your source documents are full of things that shouldn't go into one. Your VMET, for instance, contains Privacy Act data. Don't paste whole documents. Copy out the parts you need and remove: - SSN, DoD ID number, date of birth, home address - Unit names and specific locations, if they'd say more than you want - Anything classified, and any program, mission, or capability details that aren't public - Names of other people - Medical or family details If you held a clearance or worked on sensitive programs, describe the work at the level that's already public, like "managed IT systems for a 300-person organization," not what the systems were for. If you're unsure where that line is, ask your security manager before it goes anywhere, AI or not. Still on active duty and doing this at work? Then it's a work task on work systems, and your command's rules apply. [Can I use ChatGPT in the military?](/guides/can-i-use-chatgpt-in-the-military/) covers that line. ## Step 1: Translate the jargon Paste this with one chunk of your experience at a time: ``` I'm transitioning from the military to a civilian job. Translate the experience below into civilian resume language. Rules: - Use ONLY what's in my text. Don't add skills, tools, numbers, or results I didn't mention. - Replace military terms and acronyms with plain civilian equivalents. If you're not sure what a term means, ask me instead of guessing. - Write 3-5 resume bullets: action verb, what I did, and the result if I gave one. - If a bullet would be stronger with a number I didn't give, write [ADD NUMBER?] instead of inventing one. - After the bullets, list each military term you translated and what you translated it to, so I can check it. MY EXPERIENCE: [paste scrubbed text from your evals, VMET, or notes] ``` That last list is easy to skip, but it's the most useful part. AI sometimes "translates" a title into a civilian job that isn't really equivalent. Seeing something like "Platoon Sergeant → Operations Manager" spelled out lets you decide whether that's fair for your situation. ## Step 2: Aim it at a real job Generic resumes don't land well. Once you've got clean bullets, point them at one posting: ``` Here's a job posting and my current resume bullets. 1. List the 8-10 skills and keywords the posting cares about most. 2. For each one, show which of my bullets supports it, or say "not supported" if none do. Don't stretch. 3. Rewrite my bullets to use the posting's language where it's honestly accurate. 4. Suggest the 6-8 strongest bullets for this job, in order. JOB POSTING: [paste] MY BULLETS: [paste] ``` The "not supported" answers are valuable. They tell you where you need a certification, a project, or a better story, instead of a reworded bullet that won't hold up in an interview. If you want to see how your job maps to civilian occupations, O*NET's Military Crosswalk lets you search by your military code or title. CareerOneStop's Veteran and Military Transition Center has a Veterans Job Matcher and links to employment and training help. I'd use those to check AI's suggestions, not the other way around. ## Step 3: Check every line Read each bullet and ask: 1. Is this true, exactly as written? 2. Could I explain it for two minutes in an interview, with an example? 3. Is any number in here one I didn't provide? 4. Did any sensitive detail creep back in? 5. Would a civilian with no military background understand every word? If a bullet fails the first three, cut it or fix it. Hiring managers ask about resume lines, and "the AI wrote that" isn't an answer you want to give. ## Free tools built for this You don't have to do it all in a general chatbot. Hiring Our Heroes runs a free tool called Resume Engine that uses AI to help translate military experience into civilian resume content. According to an October 2026 VA News article written by Hiring Our Heroes, you review and approve what it generates, you can build multiple versions for different jobs, and you can request a free review from a human resume expert. VA News includes a standard note that mentioning a product isn't a government endorsement, and that goes for us too. Your installation's Transition Assistance Program (TAP) staff are worth using too. Bring your AI-drafted bullets and ask a counselor where they fall flat. ## Next steps Pull your VMET today and pick one job posting you'd actually apply to. Run Step 1 on your most recent eval tonight. That's a real first draft by the weekend. If you're still in, the habit from [writing award bullets with AI](/guides/ai-award-bullets/) (keeping a running log of wins) makes this much easier later. And the [Professionals track](/professionals/) has more on using AI at your next job once you land it. ## Sources - [DMDC milConnect, Verification of Military Experience and Training (VMET) FAQ](https://milconnect.dmdc.osd.mil/milconnect/public/faq/Training-VMET) - [VA News, "Turn your military experience into a civilian-ready resume"](https://news.va.gov/150284/turn-military-experience-civilian-ready-resume/), Oct. 15, 2026 - [O*NET OnLine, Military Crosswalk Search](https://www.onetonline.org/crosswalk/MOC/) - [CareerOneStop, Veteran and Military Transition Center](https://www.careeronestop.org/Veterans/default.aspx) *Checked against these sources on Oct. 21, 2026. How this is made: an AI editor drafts and fact-checks each article. If we get something wrong, we fix it and note the change.* --- Source: https://aiexplainedplainly.com/guides/ai-voice-cloning-scams/ # AI voice cloning scams: how to protect your family *Guide, October 21, 2026* ## TL;DR - Scammers can copy a voice from a short online clip, so a familiar voice on the phone isn't proof of anything anymore. - Set up a family code word, as the FBI recommends. If a call feels off, hang up and call back on a number you know. - Pressure to pay by wire, crypto, or gift cards is a scam sign. Report it at ReportFraud.ftc.gov and ic3.gov. The phone rings. It's your kid, or your mom, and they're panicking. There's been an accident, or an arrest, and they need money right now. Please don't tell anyone. It sounds exactly like them. That's the point. With AI voice cloning, it might not be them at all. This isn't a hypothetical. The FTC warned about it in 2023, and the FBI issued a public service announcement in December 2024 saying criminals use AI-generated audio to impersonate relatives "in a crisis situation, asking for immediate financial assistance or demanding a ransom." Here's how it works and what to do about it. If you haven't set up your AI privacy settings yet, [Issue #1](/issues/001/) covers that separately. The official sources for this piece are linked at the bottom. ## How the scam works According to the FTC, all a scammer needs is "a short audio clip of your family member's voice," which they might get from something posted online, plus a voice-cloning program. Then they call you sounding like that person. The script is usually some version of the old "grandparent scam," just with a much more convincing voice: - Someone you love is in trouble: a crash, jail, a medical emergency, a kidnapping - They need money fast - You're asked to keep it secret - You're told to pay in a way that's hard to undo The FBI also says criminals have used cloned audio to get into people's bank accounts by impersonating them. So this isn't only a family problem. ## The red flags The voice can't be trusted anymore, so pay attention to everything else. **Urgency and secrecy.** Real emergencies rarely require you to act in the next ten minutes without telling anyone. Pressure to skip calling someone else is the biggest tell there is. **How they want to be paid.** The FTC says scammers ask you to pay in ways that make it hard to get your money back. If the caller wants a wire transfer, cryptocurrency, or gift cards (and the card numbers and PINs read over the phone), treat it as a scam. **The caller ID.** It might show your relative's name or a real-looking number. The FCC says caller ID can be faked, which is called spoofing. A familiar number proves nothing. **Something just off.** The FBI suggests listening closely to tone and word choice. Does it sound like how they actually talk? That's worth noticing, but don't count on catching it. A good clone may not slip. ## Set up a family code word tonight This is the single most useful thing you can do, and it takes five minutes. The FBI's advice is direct: "Create a secret word or phrase with your family to verify their identity." A few tips for picking one: - Make it something random, not a pet's name, birthday, or anything that's on social media - Say it out loud in person. Don't text it, email it, or post it - Make sure everyone who might get the call knows it, grandparents and teenagers included - Agree on the rule: if anyone calls asking for money in an emergency, the other person asks for the word If the caller can't give it, hang up. A real family member in real trouble will understand. Here's a short message you can send to start the conversation. Copy it into your family group chat (and save the actual word for in person): ``` Hey all, quick favor. Scammers can now copy someone's voice with AI and call pretending to be family in an emergency. Can we pick a family code word next time we're together? If anyone ever calls asking for money urgently, we ask for the word first. Don't reply with ideas here, let's do it in person. ``` ## What to do if you get the call Keep it simple. You don't have to figure out whether it's AI. You just have to verify. 1. **Ask for the code word.** No word, no money. 2. **Hang up.** You won't hurt a real relative by calling back in two minutes. 3. **Call them directly** on the number you already have saved, not one the caller gives you. The FTC says if you can't reach them, try another family member or their friends. 4. **Don't pay or share anything** while you're checking. No account numbers, no card numbers, no codes. 5. **If it was a scam, report it** (more on that below) and warn the rest of the family so they know what to expect. The same goes for calls that sound like your bank or a government agency. The FBI and FCC both say to hang up and call the organization back on a number you look up yourself, like the one on your card or their official website. ## Make your voice harder to copy You can't disappear from the internet, but you can make it less easy. The FBI suggests limiting how much of your image and voice is online, making social media accounts private, and limiting followers to people you actually know. If your family posts a lot of videos with talking in them, that's worth a conversation. Not panic, just a quick check on who can see what. If your bank uses voice ID to log you in, I'd ask them what other verification options they offer, given the FBI's warning about cloned audio and bank accounts. ## Is using a cloned voice to call people illegal? For robocalls, yes in most cases. In February 2024, the FCC ruled that AI-generated voices count as "artificial" voices under the Telephone Consumer Protection Act. That means robocalls using them need the called person's prior consent unless an exemption applies. Faking caller ID to defraud someone is also illegal under FCC rules. That doesn't stop criminals from doing it, of course. It mostly gives regulators and law enforcement more tools after the fact. Your code word does more to protect you in the moment. ## If you already paid Move fast. The FTC's advice is to contact whoever you used to send the money right away and ask them to reverse it: - **Bank transfer or Zelle:** call your bank or credit union - **Wire transfer (Western Union, MoneyGram):** call the wire company and say a scammer tricked you - **Gift cards:** call the number on the back of the card, and keep the card and receipt - **Cryptocurrency:** contact the exchange or ATM operator, though it can be hard to get crypto back Then report it. File with the FTC at **ReportFraud.ftc.gov** and with the FBI's Internet Crime Complaint Center at **ic3.gov**. Include the phone numbers used, how you paid, and how the contact started. ## Share this with the people who'd get the call Most of what protects your family here has nothing to do with technology. It's a code word, a habit of hanging up and calling back, and knowing that a familiar voice isn't proof anymore. If you lead a team, the same verification habit belongs in your payment process too: no wire goes out on a phone call alone. The [Leaders track](/leaders/) has more on setting AI ground rules, and [What is an AI agent?](/guides/what-is-an-ai-agent/) explains the other kind of AI that's starting to act on people's behalf. Pick your family's word this week. ## Sources - [FTC Consumer Alert, "Scammers use AI to enhance their family emergency schemes"](https://consumer.ftc.gov/consumer-alerts/2023/03/scammers-use-ai-enhance-their-family-emergency-schemes), Mar. 20, 2023 - [FBI IC3 Public Service Announcement I-120324-PSA, "Criminals Use Generative Artificial Intelligence to Facilitate Financial Fraud"](https://www.ic3.gov/PSA/2024/PSA241203), Dec. 3, 2024 - [FCC, "FCC Makes AI-Generated Voices in Robocalls Illegal"](https://www.fcc.gov/document/fcc-makes-ai-generated-voices-robocalls-illegal), Feb. 8, 2024, and the [Declaratory Ruling (FCC 24-17)](https://docs.fcc.gov/public/attachments/FCC-24-17A1.pdf) - [FCC, Caller ID Spoofing](https://www.fcc.gov/spoofing) - [FTC, "What To Do if You Were Scammed"](https://consumer.ftc.gov/articles/what-do-if-you-were-scammed) *Checked against these sources on Oct. 21, 2026. How this is made: an AI editor drafts and fact-checks each article. If we get something wrong, we fix it and note the change.* --- Source: https://aiexplainedplainly.com/guides/what-is-an-ai-agent/ # What is an AI agent? A plain-English explanation with examples *Guide, October 21, 2026* ## TL;DR - A chatbot answers. An AI agent takes actions: it plans steps, uses tools like a browser or your apps, and checks its own progress. - Good first uses are low-stakes and easy to check: research, comparing options, drafting things for your approval. - Keep approval on anything that spends money, sends messages, or deletes data. Agents can be tricked by what they read. You've probably used a chatbot. You type a question, it types back an answer, and then *you* go do something with it. An AI agent skips that last part, at least some of it. You give it a goal, and it works out the steps and takes them: opening websites, filling in forms, using your apps, then reporting back. That's the whole idea. The rest of this piece is what that looks like in practice and where it can go wrong. ## The short definition An AI agent is an AI system that can take actions toward a goal, not just produce text. Anthropic, the company behind Claude, draws a useful line in its engineering guide. It describes *workflows* as systems where AI and tools follow "predefined code paths," set up in advance by a person. *Agents*, on the other hand, "dynamically direct their own processes and tool usage." In other words, a workflow follows a recipe someone wrote, and an agent decides the next step itself. The same guide boils it down even further: agents "are typically just LLMs using tools based on environmental feedback in a loop." (An LLM, or large language model, is the kind of AI behind ChatGPT, Claude, and Gemini.) It tries something, looks at what happened, and decides what to do next. ## Chatbot vs. agent vs. automation These three get mixed up a lot, so here's the difference in one table: | Type | What it does | Who decides the steps | Example | |---|---|---|---| | **Chatbot** | Answers you in text | You do the work after | "Summarize these notes" | | **Automation** | Runs fixed steps when triggered | A person, ahead of time | "When a form comes in, add a row to the spreadsheet" | | **AI agent** | Works toward a goal using tools | The AI, as it goes | "Find three open dentist appointments next week near me and hold the best one for my OK" | Lots of real products blend these. A chatbot with web search is a little agent-like. An automation with an AI step in the middle is still mostly automation. Don't get hung up on labels. Ask: *can this thing take actions on its own, and which ones?* ## Real examples Here's what vendors themselves say their agents do, as of October 2026. **Doing tasks in a browser.** OpenAI's help center describes a "cloud browser" in ChatGPT Work that "can read web pages, click buttons, enter information into forms, and carry out steps" on websites, including ones you sign in to. Its listed examples include checking restaurant availability, finding flights, finding a DMV appointment and preparing the booking for your approval, and reconciling invoices in accounting software. It's available on paid plans, not Free or Go. **Writing and fixing code.** Anthropic names coding as one of two areas where it has seen agents add the most value, partly because code can be checked automatically with tests. A coding agent can read a bug report, edit several files, run the tests, and try again if they fail. **Customer support.** Anthropic also describes support agents that pull up order history and knowledge base articles and can take actions like issuing refunds or updating tickets. You'll also hear about agents that read your email, manage your calendar, or work through a to-do list. Those exist too. They come with more risk, because they're acting inside your accounts. ## Why it matters if you're not technical Agents change what you hand off. With a chatbot, you're delegating writing and thinking. With an agent, you're delegating *doing*: clicking, booking, sending, buying. That's handy for chores that are tedious but simple, like comparing prices across five sites or filling in the same details on a form over and over. It's also where mistakes get more expensive. A chatbot that's wrong gives you a bad paragraph. An agent that's wrong can book the wrong flight. ## The risks, in plain English **It can be tricked by what it reads.** This is the big one. When an agent browses the web or reads your email, it might run into hidden instructions planted by someone else. OpenAI's own announcement of its earlier agent gave an example: a malicious prompt hidden in a webpage "could trick the agent into taking unintended actions, like sharing private data." This is called *prompt injection*, and it's a risk for agents in general, not just one product. **Errors pile up.** Anthropic notes that agents' autonomy means "higher costs, and the potential for compounding errors." One wrong assumption early on can carry through every later step. **It has your access.** If an agent is signed in as you, it can do what you can do. That's the point, and also the danger. ## How to try one safely Vendors build in some guardrails. OpenAI says ChatGPT asks for permission before visiting a new website by default and is designed to ask for confirmation before consequential actions like payments or bookings. Those are good defaults, so leave them on. Its own settings label "Always allow" for every website as "not recommended." On your end, I'd stick to a few rules: - **Start with tasks that are easy to check and hard to break.** Research and comparisons are great first jobs. Sending, buying, and deleting aren't. - **Keep approval on.** Anything that spends money, sends a message as you, or changes data should wait for your OK. - **Give it the least access it needs.** Don't connect your whole email account for a task that only needs one website. - **Don't hand it sensitive data you wouldn't paste into a chatbot.** The same rules from [Issue #1](/issues/001/) apply, and at work, your company's policy wins. - **Check the result.** OpenAI's own instructions end with "Check the result and its sources before relying on it." Here's a prompt pattern that works for a first agent task. It keeps the scope tight and makes it stop before acting: ``` Goal: [what you want, in one sentence] Use only these sites: [list them] Constraints: [budget, dates, location, must-haves] Do NOT buy, book, submit, or send anything. When you're done, show me your top 3 options in a table with prices, links, and anything you weren't sure about. Wait for my choice before taking any action. ``` ## A note for leaders If your team starts using agents, the policy question shifts from "what can people paste?" to "what can the AI do without asking?" Write that down: which tools are approved, which actions need a human sign-off, and which accounts agents can never touch. The [Leaders track](/leaders/) covers how to set those rules. And if you build agents or run them, our [For AI Agents page](/agents/) lists the machine-readable versions of everything on this site. Pick one boring research task this week and try it with approval turned on. You'll learn more from that than from any explainer. ## Sources - [Anthropic, "Building effective agents"](https://www.anthropic.com/engineering/building-effective-agents), Dec. 19, 2024 (vendor-published) - [OpenAI Help Center, "Using cloud browser in ChatGPT"](https://help.openai.com/en/articles/20001280-using-cloud-browser-in-chatgpt) (vendor-published) - [OpenAI, "Introducing ChatGPT agent"](https://openai.com/index/introducing-chatgpt-agent/) (vendor-published; that product has since been replaced by ChatGPT Work) *Checked against these sources on Oct. 21, 2026. Product features change often. How this is made: an AI editor drafts and fact-checks each article. If we get something wrong, we fix it and note the change.* --- Source: https://aiexplainedplainly.com/issues/ # Newsletter archive - [Issue #1: Before you paste anything into AI: a 10-minute safety setup, then turn messy notes into a clean action list](https://aiexplainedplainly.com/issues/001/): Turn off chat training in ChatGPT, Gemini, Copilot, or Claude in about 10 minutes. Then use one copy-paste prompt to turn messy meeting notes into decisions, action items, and a follow-up email. --- Source: https://aiexplainedplainly.com/guides/ # Guides - [What is generative AI? A plain-English explanation](https://aiexplainedplainly.com/guides/what-is-generative-ai/) (5 min read; Leaders, Professionals, Service & Veterans) Generative AI makes new text, images, or audio by predicting what should come next, based on patterns it learned from huge amounts of data. It's great at first drafts, summaries, and rewording. It's bad at guaranteed facts, so check anything that matters. Before you paste anything in, learn what's safe to share. Issue #1 covers the settings in about 10 minutes. - [How to use AI to write a professional email (with copy-paste prompts)](https://aiexplainedplainly.com/guides/ai-professional-email/) (4 min read; Professionals, Leaders, Service & Veterans) Give the AI the goal, the reader, and the key facts. It writes the draft; you fix the facts and the tone. Copy the five prompts below for first drafts, hard replies, shortening, tone checks, and follow-ups. Swap names and private details for placeholders before pasting, and put the real ones back after. - [How to use AI to summarize a meeting (safely)](https://aiexplainedplainly.com/guides/ai-meeting-summary/) (4 min read; Professionals, Leaders, Service & Veterans) Paste scrubbed notes or a transcript with a strict prompt, and get decisions, action items, and open questions back. Tell the AI not to guess owners or dates. Then check its output against what was actually said. Get consent before recording, use your company's approved tool for work meetings, and strip names before using a free one. - [ChatGPT vs. Claude vs. Gemini vs. Copilot: which should you use?](https://aiexplainedplainly.com/guides/chatgpt-vs-claude-vs-gemini-vs-copilot/) (5 min read; Professionals, Leaders) All four have useful free plans. Pick based on where you already work: Microsoft, Google, or neither. Main paid plans run about $20 a month. Gemini and ChatGPT also have cheaper tiers ($4.99 and $8). If your employer provides one, use that for work. It beats any personal plan on privacy. - [Free ChatGPT vs. ChatGPT Plus: is it worth $20 a month?](https://aiexplainedplainly.com/guides/free-chatgpt-vs-plus/) (5 min read; Professionals, Leaders) ChatGPT Plus is $20 a month, billed monthly. It adds stronger reasoning models, higher limits, and extras like projects and custom GPTs. Free covers emails, summaries, and everyday questions. Upgrade only if you keep hitting limits that slow you down. Plus doesn't make your chats more private. Turn off training in settings either way. - [Can I use ChatGPT in the military? What's allowed and what's not](https://aiexplainedplainly.com/guides/can-i-use-chatgpt-in-the-military/) (5 min read; Service & Veterans) Commercial ChatGPT is fine for personal, public, non-work stuff. Never paste classified info, CUI, PII, or unit or operational details. For official unclassified work, use GenAI.mil. It's authorized for CUI, but not for PII or PHI. Your command's policy wins over anything here, so check it before using any AI tool for work. - [How to use AI to write award bullets (safely)](https://aiexplainedplainly.com/guides/ai-award-bullets/) (5 min read; Service & Veterans) AI is good at rewording your own notes into tight action-impact bullets. It can't know what actually happened, so every fact comes from you. Scrub names, IDs, unit names, locations, and anything operational first. Use GenAI.mil for official work, and keep PII out there too. Check every number and claim before it goes up the chain. Your service's awards guidance and your admin shop's format win. - [How to use AI to translate military experience into a civilian resume](https://aiexplainedplainly.com/guides/ai-military-to-civilian-resume/) (5 min read; Service & Veterans, Professionals) Start with your VMET, evals, and awards, then have AI translate jargon into civilian terms using only what you give it. Scrub your SSN, DoD ID, unit names, locations, and anything classified or operational before pasting anything. Check every line is true and something you can explain in an interview. AI drafts it. You own it. - [AI voice cloning scams: how to protect your family](https://aiexplainedplainly.com/guides/ai-voice-cloning-scams/) (6 min read; Professionals, Leaders, Service & Veterans) Scammers can copy a voice from a short online clip, so a familiar voice on the phone isn't proof of anything anymore. Set up a family code word, as the FBI recommends. If a call feels off, hang up and call back on a number you know. Pressure to pay by wire, crypto, or gift cards is a scam sign. Report it at ReportFraud.ftc.gov and ic3.gov. - [What is an AI agent? A plain-English explanation with examples](https://aiexplainedplainly.com/guides/what-is-an-ai-agent/) (5 min read; Professionals, Leaders, Agents) A chatbot answers. An AI agent takes actions: it plans steps, uses tools like a browser or your apps, and checks its own progress. Good first uses are low-stakes and easy to check: research, comparing options, drafting things for your approval. Keep approval on anything that spends money, sends messages, or deletes data. Agents can be tricked by what they read. - [Before you paste anything into AI: a 10-minute safety setup, then turn messy notes into a clean action list](https://aiexplainedplainly.com/issues/001/) (4 min read; Leaders, Professionals, Service & Veterans, Agents) What you get: a privacy setup for ChatGPT, Gemini, Copilot, or Claude, plus one prompt that turns meeting notes into an action list. Time: about 5–10 minutes for the setup, then 3–5 minutes per meeting. Both are estimates; we haven't timed them yet. Do this: turn off chat training in your AI assistant today, then try the prompt on your next meeting.