AI Productivity
How to Fact-Check an AI Answer in Three Steps
AI chatbots can sound sure and still be wrong. Ask for sources, open them, and cross-check one key fact before you trust or share an AI answer.
AI chatbots write in a calm, confident voice, whether the answer is right or not. They can mix up dates, misquote people, or name a study that doesn't exist. Most AI tools say so themselves with a small "can make mistakes" note under the chat box. You don't need to stop using them. You just need a quick habit for checking an answer before you rely on it or pass it on.
Here's the habit from our video: ask for sources, open them, and cross-check one key fact.
Step 1: Ask "What are your sources?"
After the chatbot answers, ask it directly:
What are your sources for this? Give me links I can open, and tell me which claim each link supports.
A few things to know:
- Tools that search the web while they answer, such as ChatGPT search, Perplexity, Google Gemini and Microsoft Copilot, can usually show the pages they used.
- If the tool isn't searching the web, it may still produce links or titles that look real but aren't. That's why Step 2 matters.
- If it can't give a source, treat the claim as unconfirmed, not as false. It just means you have to check it yourself.
Step 2: Open the links and read them
This is the step most people skip. A link existing doesn't mean it says what the chatbot claims.
When you open each source, check:
- Does the page load? A broken or made-up link is a red flag for the whole answer.
- Does it actually say that? Use Ctrl+F (Cmd+F on Mac) to find the exact number, name or date the chatbot gave you.
- Who published it? An official site, a known news outlet or a research paper carries more weight than an anonymous blog or a forum post.
- How old is it? Prices, laws, product features and "latest" anything change fast. Check the date on the page.
Step 3: Cross-check one key fact somewhere else
You don't need to verify every sentence. Pick the one fact the whole answer depends on, the number, date or claim you'd actually act on, and look it up somewhere the chatbot didn't point you.
Good places to cross-check:
- A plain web search on Google or Bing, looking at several independent results rather than the first one.
- The original source: the company's own page, the government site, the official documentation.
- Fact-checking sites for news and viral claims, such as Google Fact Check Explorer, AP Fact Check, Reuters Fact Check and Snopes.
- Wikipedia as a starting point, then the references at the bottom of the article.
Checking what other sources say about a claim, instead of digging deeper into one page, is often called lateral reading. It's part of the SIFT method from digital-literacy educator Mike Caulfield: Stop, Investigate the source, Find better coverage, Trace claims to the original.
A prompt you can save
Paste this after any AI answer you plan to use:
List the 3 most important factual claims in your answer. For each one, give a source I can open and say how confident you are. Mark anything you're unsure about.
Then do Steps 2 and 3 on the claim that matters most.
When to be extra careful
Spend more time checking when the answer is about:
- Health, medicine or anything you'd ask a doctor about
- Money, taxes or legal questions
- Recent news or events from the last few weeks
- Quotes, statistics and study results
- Anything you're about to post, send to your boss or hand in for school
For health, legal or financial decisions, use the AI answer as a starting point for questions, then confirm with a qualified person or an official source.
The takeaway
AI is a fast first draft, not the final word. Ask for sources, open them, and cross-check the one fact that matters. It takes a minute or two, and it's the difference between using AI and being fooled by it.
Follow @ai_tools_day for one practical AI tip a day.