Guide Library

Week 3 · Guide 1 of 6

Foundations

Check an AI Answer Before You Use It

Field Guide

  • 20–30 minGuided walkthrough

Your agent's answers can read as finished and still get a fact wrong, such as a date that moved or a pending decision written up as approved. In this guide you'll check the claims that matter in one answer against the material it should come from, and fix what doesn't hold up before anyone acts on it.

You’ll leave with

A checked answer you can use, with each important claim supported by your source, corrected or marked unconfirmed, and a check prompt you can reuse.

What You Need for This Guide

  • An AI answer you're about to send, forward or act on. Step 1 shows you how to make one if you don't have one.
  • The note, message or document the answer should be based on.
  • Your approved AI tool.

Choose the tool you’re allowed to use

Your approved tool

The tool your company approved.

Pick your approved tool above to see the steps and screenshots for it.

Start guide
6 steps · At your own pace
About saved progress

No account needed. Checkmarks save on this browser and device when browser storage is available. If storage is blocked or full, new checkmarks last until you reload or close this page. Progress doesn’t sync to other browsers or devices. Clearing this site’s browser data removes it, and private browsing may not keep it after you close the window.

Standard Data rules for AI at work

Data Rules

Data rules for AI at work

Check with your IT Team before you paste any work data into an AI tool. Avoid names, account numbers, customer data, or anything sensitive without explicit permission from IT. If you aren't sure, redact anything sensitive and replace it with generic information. AI makes mistakes, make sure to fact check your agents outputs even if they sound right.

Before you start

  • Check with your IT teamAsk what your company allows before you put anything from work into an AI tool. Your company policy is the one that counts, and it is the one this library cannot see.
  • Use the tool your company approvedChatGPT, Claude, Microsoft 365 Copilot Chat and Gemini are all covered here. Use the one your company approved, not the one you like best.
  • Sign in with your work accountA work account keeps your chats inside the agreement your company signed. A personal account in the same product sits outside it.
  • Connected appsAn email or file connection your IT team approved can read the names and addresses in what it opens. Everything else under “Never put this into an AI tool” still applies.
  • No approved account yetFollow the steps with made-up example text, or on paper, until IT gives you one.

Never put this into an AI tool

  • Customer or personal detailsNo names, email addresses, phone numbers, home addresses or account numbers. That covers colleagues as well as customers.
  • Regulated recordsHealth, financial, HR and legal records stay in the systems built to hold them, unless IT has told you in writing that this tool may see them.
  • Passwords, keys and access codesNo credentials of any kind, yours or anyone else’s, and nothing that grants access to a system.
  • Anything under an NDA or not yet publicUnreleased plans, contracts, pricing and source code need clearance before they go near a chat window.
  • Anything your company policy prohibitsYour company’s rules win over anything in this library. If policy keeps something out of AI tools, it stays out, whatever a guide suggests.

Do this instead

  • Swap real details for placeholdersWrite “Customer A”, “[client name]”, “$X”. You get the same quality of answer, and you put the real details back yourself.
  • Describe the situation rather than pasting the documentMost of what an agent needs is the shape of the problem, not the file it arrived in.
  • Trim a file before you upload or paste itTake out the identifying columns and pages first, or work from a short extract.

Check the answer before you use it

  • Read it as if a colleague had drafted itAn agent states wrong things with complete confidence. Your name goes on the work, so the check is yours.
  • Say where the agent helpedFollow your company rule on disclosing AI assistance, especially in client-facing or formal work.

When you are not sure

  • Ask before you paste or upload, not afterA question to your IT team takes a minute. Cleaning up after a data incident takes a great deal longer.
  • When in doubt, leave it outIf you would not put it in an email to someone outside the company, keep it out of the chat.
Open the full page
Download Guide

Step

Choose an answer and mark its claims

  1. Choose an answer your agent gave you that you're about to send, forward or act on, where a wrong date or status would send someone the wrong way. An update written from your meeting notes, a summary of a policy or long email you're about to forward, a reply that quotes a date, price or rule, or at home a trip plan drafted from booking emails would all work.

  2. No answer yet? Make a short one from your own notes. Open a new chat, copy the prompt below and replace its bracketed line with five to ten lines of notes from a recent meeting or call, following the data rules.

    Write a short update from my notes
    Write a three-sentence update from my notes below for a colleague who missed the meeting. Use only what my notes say. Do not add details, dates or decisions that are not in them.

My notes:
[paste five to ten lines of your own notes from a recent meeting or call]

    The update it writes is the answer you'll check, and your notes are its source. Keep the chat open to copy from.

  3. Copy the answer into a note or document and mark each claim someone could act on: a name, date, number, owner, decision, commitment or status. For a long answer, start with the part you're about to use.

Step

Find the source each claim should come from

Find the original material before you ask your agent anything, so you're checking against something you've read yourself.

  1. It's usually your own notes, the original email, message or document, written confirmation from whoever made the decision, or for a fact about the outside world, the publisher's own dated page.

  2. Check that the source applies. It should cover the same project, version and time period as the answer, because an earlier draft or last year's policy can't confirm this year's answer.

  3. If you can't find a source for a claim, note who or what could confirm it and carry on with the rest. If nothing covers the answer at all, ask the person who would know before you go further, or go back to step 1 and choose an answer whose source you have.

Step

Ask your agent to trace each claim to the source

  1. Open a new chat for the check, away from the conversation that wrote the answer. Copy the prompt below and replace its two bracketed lines: your source text under Source, and the answer under Answer to check. Paste only the parts of the source that cover your claims, following the data rules.

    Check each claim against its source
    Download
    Check the answer against the source. Both are pasted below. Use only the source. Do not use other files, earlier conversations or outside information, and do not rewrite the answer yet.

List every claim in the answer that someone could act on, such as a name, date, number, owner, decision, commitment or status. For each claim, give:
1. The claim, in a few words.
2. One label: SUPPORTED if the source states the whole claim, DIFFERENT if the source says something else, or NOT IN SOURCE if the source does not cover it.
3. The exact words from the source behind that label, in quotation marks. For NOT IN SOURCE, write "none".

Treat words like pending, proposed, draft and not confirmed as part of the fact. If the answer drops one, label the claim DIFFERENT.

After the list, tell me whether the source covers the same project or situation and the same time period as the answer, and anything that makes you unsure.

Source:
[paste the note, message or document text the answer should be based on]

Answer to check:
[paste the AI answer you want to check]

  2. Read the list it sends back. Each claim gets a label and the source's words behind it:

    • SUPPORTED: the source states it, so it's a source fact if the quote holds up in step 4.
    • DIFFERENT: the source says something else, or the answer dropped a word like pending or proposed.
    • NOT IN SOURCE: the answer goes beyond the source, with an inference or a filled gap, so it's unconfirmed for now.

    Check that the list includes every claim you marked in step 1. If one is missing, ask in the same chat for its label and the source's words. If the agent says the source covers a different project or time period, go back to step 2 for the right one.

Step

Read the evidence yourself

The labels are the agent's reading, and a confident, neatly formatted check can still be wrong. Here you decide what each claim really rests on.

  1. For each SUPPORTED claim, find the quoted words in your source. If they aren't there, the quote came from somewhere else, so treat the claim as NOT IN SOURCE and say so in the same chat.

  2. Check that the words support the whole claim. A quote on the same subject often covers only part of it, so the right label may be DIFFERENT or NOT IN SOURCE.

  3. If you disagree with a label, say so in the same chat: name the claim and the words in your source that limit it, and ask for a recheck without anything added. Read the reply against your source too, because an agent agreeing with you isn't evidence.

Step

Fix the claims that don't hold up

  1. Decide what happens to each claim that didn't hold up:

    • DIFFERENT: use what the source says, keeping its own words such as pending or not yet agreed.
    • NOT IN SOURCE: mark it unconfirmed, or take it out if the answer works without it. Don't swap in a detail that only sounds right.

  2. In the same chat, copy the prompt below and add your own decisions to its bracketed line, such as a label you changed in step 4, a claim the person responsible confirmed in writing or one to take out, or write none. For a short answer, you can make the changes yourself instead.

    Fix the answer from my check
    Rewrite the answer I asked you to check, using your check above and my decisions below. My decisions override your labels. Use only the source I gave you and my decisions, and add nothing else.

- Keep every SUPPORTED claim as it is.
- Replace each DIFFERENT claim with what the source says, keeping words like pending, proposed or not confirmed.
- For each NOT IN SOURCE claim, write [UNCONFIRMED: what's missing] in its place, unless my decisions confirm it or say to remove it.
- Keep the answer's purpose, order and tone.

My decisions: [any label you changed, any claim to take out, and any claim confirmed in writing, with what it should say and who confirmed it, by role, or write none]

After the rewrite, list each sentence you changed and why.

  3. Read the rewrite against your source once more, because a rewrite can drop a qualifier or move a date. Each claim should say what your source or a written confirmation says, and every unconfirmed claim should be marked.

Step

Save the checked answer and the check prompt

  1. Put the checked answer where you'll use it. If you swapped any details for placeholders before pasting, put the real ones back there yourself, outside the chat, and check each one against your source. Beside the answer, or in your own notes, keep the title or link and date of the source behind each important claim, so you can find the evidence again.

  2. Before anyone acts on it, ask whoever could confirm each unconfirmed claim, or leave the label in so readers know that point is open. A decision, a commitment to a customer, money or a policy question needs the responsible person's confirmation, whatever the check found.

  3. Download the check prompt from its card in step 3, or copy it into My Agent Files or another folder you can find again. Edit your copy to fit your work, and replace the source and the answer each time you use it.

Run it the next time an answer about to go out has a date, a number or a decision in it.

Check an AI Answer Before You Use It

Confirm your result

Confirm the outcomes below to finish this guide.

Final checks

You have a checked version of your answer, with each claim that matters supported by your source or a written confirmation, corrected or marked unconfirmed.

For each claim you kept, you can point to the words behind it in your source or a written confirmation, and for each unconfirmed one you can name who could confirm it.

Provide Feedback

Thanks. That went straight to the people who write these guides.

Next in Week 3

Turn Your Meeting Transcripts Into Something Useful

Continue in the suggested order, or choose another guide for your task.

View Week 3 guides