A polished essay that doesn't sound like the student who handed it in is a common puzzle now. This guide walks through the habits ChatGPT's default output tends to show, what to compare an essay against, and how to follow up fairly when a draft feels off.
Key takeaways
- Polish and good grammar are not evidence of AI use on their own.
- The strongest signal is a mismatch with the student's earlier writing and with your class material.
- Check one or two sources: invented or vague citations are a common giveaway.
- Use sentence highlights to decide where to look, then ask the student about the work.
Why ChatGPT essays are hard to spot by feel
Default ChatGPT output is fluent, grammatical and well organized. That is exactly what many teachers ask for, so polish alone tells you very little. Strong writers, students who use a grammar checker and students who worked hard on a final draft can all produce clean prose. What usually gives an AI-written essay away is not quality but sameness: it reads as if it could have been written for any class, by anyone. The reverse matters too: a student whose writing has improved over a term is not automatically suspect.
Signs in the writing itself
None of these is decisive. Look for several together, and look across the whole essay rather than at one paragraph.
- A five-part shape with no surprises. An introduction that restates the prompt, three body paragraphs of similar length, and a conclusion that opens by summing up.
- Balanced to a fault. Every argument gets a counterpoint, and the essay rarely commits to a clear, debatable position.
- Stock transitions and phrasing. Openers like 'It is important to note' and words such as 'crucial', 'pivotal' and 'foster' used far more often than a teenager would use them.
- Confident but empty claims. Lines like 'studies show' or 'experts agree' with no study or expert named.
- Lists disguised as prose. Paragraphs that read like bullet points joined with commas, often in neat groups of three.
Signs that come from context
Context is usually more telling than style, because ChatGPT doesn't know what happened in your classroom.
- The essay ignores the readings, discussions or specific framing you gave in the prompt.
- It cites books, page numbers or quotations that don't exist, or that don't say what the essay claims. Checking one or two takes minutes.
- The voice, vocabulary or pattern of mistakes is very different from the student's in-class writing, earlier drafts or emails.
- The argument drifts toward a general version of the question rather than the one you actually set.
Here's a typical case. You ask students to compare two characters using a scene the class acted out together. The essay you get compares them competently but never mentions that scene, quotes a line that isn't in the book, and ends with 'Ultimately, both characters remind us of the importance of empathy.' Any one of those could be innocent. Together, they're worth a conversation.
Checking the essay with a detector
Paste the essay into the free AI detector to get an overall score and sentence-level highlights. Red sentences read as likely AI. Yellow means mixed signals, which often points to human writing that was polished or paraphrased by a tool. Unhighlighted sentences read as human, and you can hover over any sentence to see its probability.
Use the highlights as a map, not a verdict. If the red sentences cluster in the body paragraphs while the introduction reads as human, that tells you where to focus your questions. Winston AI reports whether text is likely machine-written; it doesn't say which tool was used, so the same check covers ChatGPT and other chatbots.
What to do when an essay feels off
- Write down the specific signs you noticed, with examples quoted from the essay.
- Look at what you already have: earlier drafts, in-class work, and version history if the essay was written in a shared document.
- Invite the student to talk through the essay. Ask how they chose the thesis, what a source said, or why the paragraphs are in that order.
- Hold off on accusations in writing or in front of classmates until you've had that conversation.
- Follow your school's academic integrity policy, and keep any score as one piece of evidence among several.
Most students who did the work can explain it without much trouble. For a fair way to run that conversation, see how to talk to a student about suspected AI use, and for a full review routine, our page for teachers.