A straight answer
How do teachers check for AI in essays?
Teachers use five kinds of checks. They are not equal: some describe the finished text, others describe how it was written. Here is what each one shows.
In short
Teachers check for AI in five main ways: AI detector scores, reading for style and fit with earlier work, document version history, recorded writing sessions, and checking sources and talking with the student. Detector scores and style describe the finished text and are often wrong. Process records, sources and a conversation describe how the essay was produced, and those hold up better when a result is disputed.
The five checks
| Check | What it looks at | What it can show | Main weakness |
|---|---|---|---|
| AI detector score | The finished text | An estimate that text resembles machine output | False positives; no way to cross-examine |
| Style and earlier work | The finished text, against the student’s past essays | A change worth asking about | Students improve and get help legitimately |
| Document version history | Snapshots of a Google Doc or Word file | That a document existed and changed over time | No typed versus pasted; can be fabricated |
| Recorded writing session | Every keystroke, paste and pause, as it happens | How the text entered the editor | Not where the ideas came from |
| Sources and a conversation | Citations and the student’s own account | Fabricated sources; whether the student knows the work | Takes time; nerves can mislead |
1. AI detector scores
Turnitin, GPTZero and similar tools read the finished text and return a percentage. Many schools license one, so it is often the first check a teacher sees. Turnitin states a document-level false positive rate under 1 percent for documents with more than 20 percent AI writing. Independent research found GPT detectors flagged 61.3 percent of TOEFL essays by non-native English writers as AI-generated (Liang et al., Patterns, 2023), and several universities have turned Turnitin’s indicator off. A score is a reason to look, not a finding. More on the Turnitin accuracy page.
2. Reading for style and comparing with earlier work
Experienced teachers notice when an essay sounds unlike the student. That instinct is useful for deciding what to look at, and weak as evidence, because polished prose has no author signature. A comparison with the same student’s earlier essays is stronger than an impression, and still only a reason to ask.
3. Document version history
In Google Docs or Word, a teacher can open the version history and see how the document changed between saved states. Replay extensions such as Draftback and Revision History play that history back. It is useful, with limits: it records document states rather than keystrokes, it does not distinguish typed from pasted text, and auto-typer tools can replay pasted text into a Doc so the history looks written. See is version history proof.
Source: learn.microsoft.com Q&A on Word version history; phrasly.ai/human-auto-typer; Draftback and Revision History listings; checked 2026-09-08.
4. Recorded writing sessions
Some teachers have students write in an editor that records the session as it happens. The record shows each paste and its length, where the student paused, what they deleted and rebuilt, and how the essay grew over several sittings. Generated text usually arrives in a paste, or is retyped at a steady rate with little revision, which looks like copying in a replay. Tools in this category include Turnitin Clarity, Grammarly Authorship, Cursive, Process Feedback and EssayGuard; the comparison of alternatives sets them side by side.
5. Sources and a conversation
The oldest checks still work. A citation that does not exist, or a quotation that is not in the cited source, is concrete and anyone can verify it. A short conversation about the argument shows quickly whether a student knows their own paper. Teachers who use these checks well start with a question about the work, not an accusation.
Which checks hold up
When a result is disputed, the checks that describe how the essay was produced carry more weight than the ones that describe the text. In Newby v. Adelphi University (2026), a student’s 100 percent Turnitin flag did not survive review; the court found the finding “without valid basis.” A careful process asks for two independent kinds of evidence before any penalty, and a detector score plus a style impression count as one.
Where EssayGuard fits
EssayGuard is a recorded writing editor. Students open one link, sign in and write; every keystroke, revision, pause and paste is recorded with server-side timestamps, and the record is cross-checked against periodic snapshots so a fabricated history does not read as writing. Each essay is marked Approved, Needs Review or Flagged with a plain summary, and EssayGuard compares reading level and vocabulary with the same student’s earlier essays. It gives no AI percentage, and a teacher makes every decision. It is free for the rest of this semester; try it on one assignment.
A record shows how text entered the editor. It cannot show whose ideas it carries, which is why the conversation stays the last step.
Questions people ask
Can teachers tell if you used ChatGPT?
Not reliably from the prose alone. A teacher can tell much more from how the essay was written: whether it was typed or pasted, revised or produced in one pass, and whether the student can explain it. That is why more teachers are moving to process records rather than detector scores.
How do teachers check for AI on Google Docs?
Usually by reading the version history, sometimes with a replay extension. Version history shows snapshots of the document, does not separate typed from pasted text, and can be faked with auto-typer tools, so it is one signal rather than proof.
Is a detector score enough evidence?
No. Detector scores are estimates about prose and produce false positives, especially for some groups of writers. A score is a reason to look further, not a finding.
What is the most reliable check?
Two independent kinds of evidence that agree, such as a writing record that shows large unexplained pastes plus a student who cannot explain the argument. No single check is enough on its own.
Try it with one assignment
The rest of this semester costs nothing. Setup takes about three minutes.
Every course you teach, unlimited students. Students install nothing.