Did AI Write That Paper? What Teachers Can—and Can’t—Tell

A student turns in a paper.

It’s polished. The grammar is nearly perfect. The vocabulary seems more sophisticated than usual. Something about it just doesn’t sound like the student sitting in your classroom.

So you wonder: Did AI write this?

More teachers are asking that question. Unfortunately, there’s no simple test that will always give you the answer.

There are things worth noticing. There are questions worth asking. Tools may provide additional information.

But none of them, by themselves, prove that a student used artificial intelligence improperly.

Links in this post may be affiliate links. Purchases made using these links will not cost you more but may pay me an affiliate fee.

We are a participant in the Amazon Services LLC Associates Program, an affiliate advertising program designed to provide a means for us to earn fees by linking to Amazon.com and related sites. Thank you for using my links.

And that may mean we need to change the question. Instead of asking only, “How can I catch AI-written papers?”, perhaps we also need to ask: “How can I make student thinking more visible throughout the writing process?”

First: AI Detectors Aren’t Proof

It can be tempting to paste a suspicious paper into an AI detector and treat the resulting percentage as the answer.

That’s risky.

AI detection tools can produce false positives—identifying human writing as AI-generated—and false negatives. Turnitin warns educators that its AI-writing detection isn’t always accurate and shouldn’t be the sole basis for adverse action against a student. Its current system doesn’t even display numerical scores below 20% because false positives are more common in that range.

That doesn’t necessarily mean an AI detector is useless.

It means you should treat a detection result as one piece of information, not a verdict.

The same principle applies to teacher intuition. Knowing your students’ writing matters, but “This doesn’t sound like you” is a reason to investigate—not proof of misconduct.

What Might Make You Take a Closer Look?

No “AI writing style” proves a student used ChatGPT or another tool. Still, certain things may reasonably prompt a teacher to look more closely.

A Sudden Change in Writing Style

Perhaps a student who normally writes short, straightforward sentences suddenly submits sophisticated, highly structured prose. That’s worth noticing.

But students improve. They get help. They use spelling and grammar tools. Parents edit papers. Some students write differently when given more time. A change is a reason for a conversation, not an automatic conclusion.

Sources That Don’t Exist

Generative AI can produce incorrect information, including fabricated citations.

If a student cites a journal article, book, author, quotation, or website that you can’t locate, investigate further.

Ask the student to show you the source.

Quotations That Aren’t in the Source

The source may exist while the quotation does not.

A paper might contain a convincing-looking quotation attributed to a real author or book, yet the author never wrote those words.

This is one reason source checking remains important even in the age of AI.

Citations That Don’t Support the Claims

Sometimes all the citations are real—but when you follow them, they have little to do with what the paper says.

Again, that’s not automatic proof of AI use. It does tell you that the student needs to explain the research process.

Polished but Vague Writing

A paper may sound impressive while saying remarkably little.

It may contain lengthy introductions, broad generalizations, repeated ideas, or paragraphs that sound authoritative without providing concrete evidence.

Don’t assume polished-but-vague writing is AI-generated. Instead, ask the student to explain what a particular passage means.

Vocabulary the Student Can’t Explain

A sophisticated word isn’t evidence of cheating. But if a student repeatedly uses terminology and then cannot explain what those terms mean in the context of the paper, that’s worth discussing.

The Student Can’t Explain the Argument

This may be one of the most useful conversations a teacher can have.

Ask:

“Tell me what your main argument is.”
“Why did you choose this evidence?”
“What did this source contribute to your paper?”
“Which part of the paper was hardest to write?”
“What did you change between your first and final drafts?”

Pinterest graphic showing a teacher reviewing a student paper beside a laptop and books about research, writing, citations, and academic integrity, with the headline “Did AI Write That Paper? What Teachers Can—and Can’t—Tell” and Library Lady branding.

Turnitin’s current educator guidance similarly recommends asking students to explain the process they followed and discussing specific areas of concern rather than treating a detector result as an accusation.

A student who actually did the thinking behind a paper should usually be able to talk about that thinking—even if the writing itself isn’t perfect.

Be Careful With the “Signs of AI” Lists

Teachers have probably seen lists online claiming AI writing is easy to identify.

“AI uses too many transition words.”
“AI uses em dashes.”
“AI always writes in groups of three.”
“AI uses certain vocabulary.”
“AI writing is grammatically perfect.”

Those aren’t reliable tests.

Humans use em dashes. Students write lists of three. Strong writers use transitions and sophisticated vocabulary. Students may also use legitimate editing tools that improve grammar.

Conversely, AI-generated writing can contain errors, awkward sentences, casual language, and unusual formatting.

There is no single word, punctuation mark, sentence structure, or writing habit that proves AI use.

The Better Strategy: Make the Writing Process Visible

This is where teachers and librarians may have more control.

Instead of receiving only a finished 1,500-word research paper, build checkpoints into the assignment.

For example:

1. Topic proposal – Have the student explain what they want to investigate and why.

2. Research question – Require a working research question before students begin research.

3. Preliminary source list – Students identify several sources they expect to use.

4. Research notes – Have students keep notes showing what information came from which sources.

5. Annotated bibliography – Ask students to briefly explain why each source is useful.

6. Working thesis – Students submit an early version of their argument.

7. Outline – Even a simple outline provides evidence of how ideas are developing.

8. First draft – Don’t wait until the polished paper to see the student’s writing.

9. Revision – Have students explain what they changed and why.

10. Final reflection – Ask a few questions:

What part of this paper are you most satisfied with?
What was most difficult?
Which source was most helpful?
How did your argument change as you researched?
What would you do differently next time?

Now the teacher isn’t looking at one mysterious finished document. A trail of thinking leads to it.

Revision History Can Help Too

When students write in tools that preserve version histories, those records may provide useful context. If your school is a Google school, require your students to type their papers and reports in a Google document that you have access to. My daughter, a fifth grade teacher does that. And, she checks revision history if something sounds a bit off for a given student.

A document that develops gradually through outlining, drafting, deleting, rewriting, and revising tells you much more about the writing process than the final document alone.

But even here, avoid treating one particular pattern as definitive proof. Students have different writing habits and may draft in another program or offline before moving text into their final document.

Again, you’re gathering context rather than looking for a single “gotcha.”

Decide What AI Use Is Actually Allowed

There’s another important issue.

Before determining whether a student used AI improperly, students need to know what counts as improper use for that assignment.

Can they use AI to brainstorm possible topics?
Can they ask it to help develop research questions?
Can they use a grammar checker?
Can they ask AI to suggest revisions?
Can they use it to create an outline?
Can they have AI rewrite a paragraph?
Can they generate the entire paper?

Those aren’t the same thing.

Current educator guidance from Turnitin likewise emphasizes deciding when AI use is appropriate for particular assignments and communicating those expectations.

A simple assignment statement can eliminate considerable confusion: AI use for this assignment:

Allowed: Brainstorming topics and generating search terms.
Allowed with disclosure: Feedback on grammar or organization.
Not allowed: Generating sentences, paragraphs, analysis, or the final response that you submit as your own writing.

The exact rules depend on your school’s policy and the assignment’s learning objective.

What matters is that students know them.

When You Suspect AI Use, Start With a Conversation

If something seems wrong, the first step doesn’t have to be an accusation.

Try: “I’d like you to walk me through how you wrote this paper.”

Ask the student to show you research notes, sources, drafts, or revision history.
Choose a paragraph and ask the student to explain it.
Ask why a particular source was selected.
Ask what the student meant by a particular sentence.

If your school uses an AI detection tool, its results can be part of that conversation—but shouldn’t replace it. Turnitin itself recommends using detection information as a starting point for discussion and applying the school’s academic-integrity policies and educator judgment before determining whether misconduct occurred.

Maybe the Goal Isn’t Catching AI

AI has made authorship harder to determine. But it has also highlighted something educators have always known: The final paper isn’t the only thing that matters.

We want students to learn how to ask good questions.
Find reliable information.
Evaluate sources.
Take useful notes.
Develop an argument.
Support that argument with evidence.
Revise their thinking.
And communicate what they’ve learned.

When we see those steps along the way, we don’t have to rely quite so heavily on guessing whether the finished paper “sounds like AI.”

Instead, we can see the learning happen.

And ultimately, that’s far more valuable than any AI detector could ever be.

Get more information at https://www.turnitin.com/resources/topics/ai-writing

Library Lady

Sharing school library info with those who want to learn

Did AI Write That Paper?

Did AI Write That Paper? What Teachers Can—and Can’t—Tell

First: AI Detectors Aren’t Proof