How to Fact-Check

Pattern recognition before verification.

What you'll leave with

By the end of this lesson, you'll be able to recognise the warning signs of hallucination in AI output — so that potentially false information triggers a verification habit before you act on it.

Why this matters

You cannot verify everything AI tells you. That would defeat the purpose. What you can do is develop an eye for the output characteristics that signal something might be wrong — so you know where to focus your checking.

The idea

One of the subtlest risks is this: AI generates plausible text, and plausible often means what fits the pattern of what a good answer looks like. In high-stakes contexts, an answer that sounds right deserves exactly that much more scrutiny — not less.

This does not mean trusting AI less in general. It means developing a specific sensitivity to certain types of claims.

The teaching block

Red flag patterns:

• Suspiciously precise statistics: "Studies show that 67.3% of..." — precision without a source is a warning
• A very specific citation with no link and perfect details: title, author, publication, year — all suspiciously neat
• Confident claims about very recent events or news
• Detailed biographical information about someone who is not widely covered
• A technical process described with great authority in a domain AI may not have deep training on
• An answer that sounds exactly like what you wanted to hear — because AI also predicts what would satisfy you

The quick verbal check:

When you read something AI wrote and something feels slightly off, practice asking yourself: is this a fact, or is this language that sounds like a fact? Statistics, names, dates, citations, and specific claims are facts. Analysis, framing, and general descriptions usually are not.

Only facts need checking. Everything else can be evaluated on its own merits.

Example

Show two AI outputs side by side: one that is reliable (a drafted email or a general explanation), one that contains hallucination risk (a response about a specific study or a named statistic).

Walk through what triggers suspicion in the second example. Point out: "This is where I would slow down."

Try this now

Ask AI a question in a domain you know well and look for anything suspicious. You are not fact-checking yet — you are just practising the habit of noticing.

Make a note of what triggered your attention. Over time, this pattern recognition becomes automatic.

Save this

Quiet takeaway

Pattern recognition develops quickly. Once you have noticed a few red flags in AI output, the sensitivity becomes automatic — you slow down instinctively when something feels off.

Next

You can spot the warning signs. Now: what to actually do when you need to check.