The short answer

This guide is the D in the ACED framework from Better AI prompts: after you write a prompt, Double-check the answer.

AI tools write fluent, confident text, and sometimes that text is wrong. They can state invented facts, cite sources that do not exist, or miss recent changes. You do not need to avoid AI, but you do need a routine for checking anything that matters before you use it.

Why AI gets things wrong

AI language tools predict likely wording. They are not looking facts up in a trusted database unless they are connected to search or documents, and even then they can misread what they find. This leads to a few common problems:

Problem What it looks like
Invented facts A plausible statistic, date or quotation that cannot be traced to any source
Fake references Realistic-looking authors, journal names and page numbers for papers that do not exist
Out-of-date information Rules, prices or product features that have since changed
Mixed-up details Real facts attached to the wrong person, company or year
Over-confident reasoning Arithmetic or logic mistakes explained in a convincing tone

Warning signs

Be extra careful when an answer contains:

  • Very specific numbers, percentages or quotations with no source.
  • References you cannot find with a quick search.
  • Claims about recent events, laws or prices.
  • Names of people, cases or studies you have not heard of.
  • Perfectly smooth answers on topics experts still disagree about.

A five-step checking routine

  1. Mark the claims. Read the answer and underline every fact, number, name, date and reference. These are what you must check.
  2. Find a trusted source. Use official websites, textbooks, journal databases or your library. Prefer original sources over summaries.
  3. Cross-check numbers. Recalculate sums and percentages yourself. Check that units and years are right.
  4. Open every reference. Confirm that it exists, then read enough of it to confirm it supports the claim.
  5. Check the date. Ask whether the information could have changed since the tool was trained, and look for the latest version.

Prompts that help you check

You can use the tool itself to help, but treat its replies as leads to verify, not proof.

List the factual claims in your answer that I should verify,
and say how confident you are in each.
Which parts of this answer might be out of date?
Give me search terms I can use to check these claims in a library database.

A quick example

Suppose an AI writes: “A 2019 study found that 73 percent of consumers prefer video content.” Before using it:

  1. Mark the claim: a year, a percentage and a type of study.
  2. Search for the study. Does it exist, and who published it?
  3. If you find it, check the exact wording, the sample and the year.
  4. If you cannot find it, leave the statistic out or find a verified source.

When to check more carefully

The higher the stakes, the more checking it deserves. Coursework, published articles, client advice and anything about health, law or money all need careful verification. For casual brainstorming, lighter checks are fine.

Be open about AI use

Follow your institution’s rules on AI in assessed work. Where you used AI, say so, and make sure you can explain and defend every claim in what you submit.

Try it yourself

Ask an AI tool a question about a topic you know well. Mark every claim in the answer, then check three of them using trusted sources. Note what you found, and which kinds of claim were most likely to be wrong.

This guide is shared under CC BY 4.0 . You're welcome to reuse it in your own teaching with credit to Surabil Sudarshan.