How to fact-check an AI answer in five minutes
AI can sound sure and still be wrong. A five-step routine for checking names, numbers, quotes and sources before you use an answer.

AI chats are trained to produce fluent, plausible text. Most of the time plausible and true overlap. Sometimes they don't — and the wrong answer sounds exactly as confident as the right one. That's the part that catches people out: there's no change in tone to warn you.
You don't need to distrust everything. You need a quick routine for the answers that matter. Here's ours.
Step 1: Decide how much it matters
Not every answer needs checking. Brainstorming gift ideas? Fine as-is. A number going into a report, a medical or legal question, a quote you'll attribute to someone, anything you'll send to a client — check it. A useful rule: the more it would cost to be wrong, the more you verify.
Step 2: Pull out the checkable claims
Skim the answer and list the specific things that could be true or false:
- Names — people, companies, products, laws.
- Numbers — dates, prices, percentages, statistics.
- Quotes — anything in quotation marks attributed to someone.
- Sources — studies, articles, books and links the AI cites.
These are where errors hide. General explanations are usually fine; the specifics are what go wrong.
Step 3: Check the riskiest claim against a primary source
Go to the original: the company's own website, the official statistics office, the law itself, the actual study. Not another AI, and not a random blog that may have been written by one.
Pay special attention to citations. AI models can produce references that look perfect — real-sounding authors, plausible journal names, tidy page numbers — for papers that don't exist. If you're going to rely on a source, open it and confirm it says what the AI claims.
Step 4: Ask the AI to show its uncertainty
A follow-up prompt can surface weak spots quickly:
Follow-up promptWhich parts of your answer are you least sure about? List any facts, numbers or sources I should verify, and say why.This isn't proof — the model can be wrong about its own uncertainty too — but it's a fast way to find where to look first.
Step 5: Cross-check with a second model or a search
For important answers, ask the same question in a fresh chat or with a different model and compare. If two answers disagree on a fact, that's your signal to dig. If they agree, that's reassuring but not conclusive — both may have learned the same mistake — so the primary source in step 3 still wins.
Red flags worth knowing
- Very precise numbers with no source ("73.4% of users…").
- Recent events — the model may not know about anything after its training data ends.
- Quotes that sound a little too perfect.
- Answers that simply agree with how you phrased the question. "Isn't it true that…?" invites a yes.
Five minutes, not five hours
In practice the routine is quick: decide if it matters, find the two or three specific claims, check the riskiest one at the source. That small habit is the difference between using AI and being misled by it.


