How to Check an AI Chatbot Answer Before You Trust It

What you need

• An AI answer you want to rely on, such as a list of facts, a summary or a short piece of writing.
• A web browser and about fifteen minutes for a first run.
• A trustworthy place to check claims, such as an official document, the original study or a named publisher.
• A notepad or document where you can record what you checked and what you found.

Step by step

1. Pick the claims that matter most. Skip the friendly wording and underline only the facts you would act on. Example: if you are planning a trip to Lisbon, the opening hours and entry rules matter far more than the background history.

2. Trace each important claim to its source. If the answer names a report, article or author, search for its exact title. Check the publisher, the author’s record and the DOI (digital object identifier) if one is given. If the source leaves no trace at all, treat the citation as unreliable.

3. Open the cited page and find the exact detail. Look for the precise figure, quotation or conclusion the answer attributes to it. Read the text around it. Ask whether the source really says what the answer claims, not just whether it covers the same general topic.

4. Split the claim into parts. Separate it into the person or organisation, the event or feature, and the date or version. Check each part on its own, then confirm that the original source connects all three. Example: an email writing tip that says “a company changed its policy in 2024” has three parts to check.

5. Trace any numbers to the original study. Percentages, prices and survey results need extra care. Confirm the date, the sample or dataset, the method and the figure itself. If the original reports a different scope, timeframe or calculation, the AI may be overstating what the data shows.

6. Check that the information is current. Look at when the source was published and last updated. Product features, prices, laws and leadership roles change quickly, so confirm the claim with a current primary source such as an official page.

7. Ask the same question again in neutral words. Start a fresh question without repeating the wording you used before, then compare only the details that can be checked. If two answers disagree on a date, author or total, treat that as a lead to investigate, not as proof that either answer is false. Confirm the detail with an independent source.

You can also ask the tool to show its working. Here is a copyable prompt:

For the claim below, tell me the original source, its publication date, and when it was last updated. If you cannot name a checkable source, say so clearly.

Claim: [paste the one claim you want to check]

Tips and common mistakes

• Do not treat a broken link as proof of a made-up answer. Links can fail for temporary reasons. The stronger warning signs are a publisher with no record of the article, a DOI that points to a different paper, or an author with no documented link to the work.
• Polished writing is not evidence. A confident tone and precise numbers can make weak claims look researched. Precision should make a claim easier to trace, so check it harder, not less.
• A credible source can still be out of date. Check the date before you rely on it.
• Do not stop after one check. A source can exist and still not contain the statement attributed to it, so always compare the wording.
• Keep a short record. Note which claims you checked, where you found the source and whether it matched. This saves time next time.

Use it responsibly

Always check AI answers before you act on them, especially for money, health, legal matters, schoolwork or work decisions. Never paste private information into a chatbot, such as passwords, bank details, medical records, or personal data about other people. Follow the rules of your school or workplace about using AI tools, and follow the service’s own terms. Keep your account secure by using a strong, unique password and not sharing your login with anyone.