How to check a student's essay for AI, fairly

Updated October 31, 2026 · 3 min read

Short answer

Use an AI detector as one signal, never as the verdict. Check a long enough text (300+ words), read the highlighted passages, compare the essay with the student's earlier work, check its sources, and talk with the student about how they wrote it before you decide anything.

That's what Turnitin itself advises: use its score "as a single data point rather than a definitive response." Some universities go further. Syracuse University won't accept a case backed only by detector output and asks for at least three independent pieces of evidence.

The process, step by step

Know your policy first. What AI use is allowed in this assignment, how must it be disclosed, and how does your institution handle suspected misconduct? Students can only break rules they were given.
Check the text with a detector, and note the details. Record the tool, the date and the score. Use at least 300 words; Turnitin won't score less. Read which passages are highlighted, not just the headline number. What scores mean in each tool.
Look for evidence beyond the score. Compare the essay with the student's earlier work. Check that the sources exist and say what the essay claims. Vanderbilt's guidance recommends both.
Talk with the student. The MLA-CCCC task force recommends meeting "in a spirit of supportive, collaborative inquiry," asking students "to describe their process in concrete terms" and asking "about the ideas expressed in the writing and their choices as a writer."
Ask for process evidence. Drafts, notes and version history show how the work came together. Students may choose to show you their Google Docs history; at Syracuse they aren't required to share it.
Decide on the whole picture, and write it down. Syracuse asks for "at least three independent points of documentation" and won't accept detector screenshots alone. Turnitin's own advice when evidence is unclear: "give students the strong benefit of the doubt."

What can an AI detector tell you?

Useful: which passages read like AI output, and how strongly. On long, unedited AI text, leading detectors are very accurate. Independent results.
Not reliable: short passages, edited or humanized AI text, and some non-native English writing. The research on non-native writers.
Never: whether the student broke your rules. Turnitin says its score "should not be used as the sole basis for adverse actions against a student," and that "no tool can replace the educator's judgment combined with other data points."

How much weight should each signal get?

Signal	How to use it
A high detector score on a long text	A reason to look closer and talk to the student. Not a finding on its own.
A low or asterisked score (under 20% in Turnitin)	Treat as no signal. Turnitin hides these numbers because false positives are more likely.
Highlighted passages in a different voice from the rest	Worth asking about: the student may have pasted in AI text, or polished one section with an AI tool.
Sources that don't exist or are misquoted	Strong evidence that needs explaining, whatever the detector says.
Big mismatch with the student's earlier work	Meaningful, but writing changes with effort, tutoring and topic. Ask about it.
The student can't explain their own argument	Strong evidence, if the conversation was fair and unhurried.
Version history showing the essay written over time	Strong evidence the student wrote it. A single large paste deserves a question, not a conclusion.

How can you prevent problems in the first place?

State the AI rules for each assignment, including what's allowed for brainstorming, editing and translation.
Ask for a process statement. MIT Sloan's teaching team suggests a short note in which students "briefly explain how they completed the assignment, including if and how they used any AI tools."
Collect drafts or outlines along the way, so the final essay isn't the only evidence you have.
Tell students how you'll use detectors. Explain that a flag starts a conversation, not a penalty. It lowers the stakes for honest students.

Why be this careful?

Because the cost of a false accusation is high and detectors do make mistakes. Turnitin estimates its own detector flags about 1 in 100 fully human documents. Several universities, including Vanderbilt, Waterloo and Curtin, have switched AI detection off. Others, like Melbourne, tell students that "an AI writing detection report alone is not sufficient evidence for an allegation." Universities that turned off AI detection.

Questions people also ask

What AI score should I act on?

None, on its own. Turnitin says its score should not be the sole basis for adverse action, and treats anything under 20% as too unreliable to show. A high score on a long, unedited text is a good reason to look closer and talk to the student, not a finding.

Can I require students to show their version history?

You can ask, and many students will gladly show it because it helps them. Some institutions limit this: Syracuse University notes that students are not required to share or send their history. Asking for drafts or a short process statement up front avoids the problem.

Is it OK to paste student work into a free online AI detector?

Check your school's rules on sharing student work with outside tools first. Also check what the tool does with the text: AI Checker Plus deletes free checks after a day and paid checks after 30 days, and never uses text to train AI. Other tools' policies differ.

What are the strongest signs of AI misuse besides a detector?

Sources that don't exist or don't say what the essay claims, a big mismatch with the student's other work, and a student who can't explain the ideas or choices in their own essay. Vanderbilt's guidance suggests comparing with other work, checking sources and talking to the student.

Sources

By the AI Checker Plus team. We research and draft these guides with help from AI tools, then check every claim against the sources listed above.