Field notes

How to spot when AI is wrong

By Ashley · Build Metacognition · · 3 min read

JudgmentFact checkingFor students
The short version

AI is wrong in a confident voice, which is what makes it dangerous for students. Teach the specific tells: facts with no source, citations that look real but aren't, and answers that get vaguer under follow-up questions.

AI can be wrong in a very convincing voice.

That is what makes it tricky.

A wrong answer from a person might sound uncertain. A wrong answer from AI can sound polished, organized, and completely confident.

Students need to learn this early:

Fluent can still be false.

Sign 1: the answer has facts but no sources

If AI gives dates, numbers, names, studies, quotes, or statistics, those claims need checking.

A student should ask:

"Where did that come from?"

or:

"What source confirms this?"

If the tool cannot provide a real source, do not treat the claim as proven.

Sign 2: the source looks real but is not

AI can invent citations.

It may create a title that sounds academic, list an author, and format it like a real source. The source may still be fake.

Students should click, search, and verify.

If the source cannot be found outside the AI answer, do not use it.

Sign 3: the answer is too confident

Real knowledge often has limits.

If AI gives a simple answer to a complicated question, pause.

Ask:

This teaches students to look for nuance.

Sign 4: it does not match what you already know

Students should not ignore their own knowledge.

If an AI answer contradicts the textbook, teacher, source document, or assignment instructions, check carefully.

AI may be wrong. Or the student may have misunderstood. Either way, the mismatch is a signal.

Sign 5: it uses vague language

Watch for phrases like:

Those phrases may be fine in a draft, but they are not evidence.

Students should ask:

"Which experts?"

"Which studies?"

"What exactly did they find?"

Sign 6: it skips the assignment

AI may answer a question that sounds similar but is not the actual assignment.

Students should compare the answer to the prompt:

A polished wrong-format answer is still wrong.

A student fact-check routine

Use this:

  1. Circle every factual claim.
  2. Mark which claims need a source.
  3. Search for reliable confirmation.
  4. Compare with the original assignment or source.
  5. Rewrite anything that was wrong or too vague.

That is AI literacy in action.

A useful prompt

Students can ask AI:

"List the claims in your answer that I should verify before trusting them."

That does not replace checking. It helps start the checking.

Why this skill matters

Students are growing up in a world where generated content will be everywhere.

They need to become better readers of information, not more passive ones.

AI should not train them to accept answers quickly.

It should train them to ask better questions.

The next step

Keep the thinking with the student.

Build Metacognition teaches students to spot when AI is wrong, verify claims, and use AI without surrendering judgment.

Join the founding waitlist

Sources