Turnitin can often flag text that ChatGPT wrote, especially long passages pasted with little change. As people edit the output, mix it with their own writing, or run it through other tools, the indicator becomes less certain, but less certain doesn't mean safe or permitted. Instructors weigh the score alongside your drafts, your usual writing, and whether you can discuss the work.
ZeroTell publishes this blog and makes an AI detector and humanizer. Turnitin is a separate company with its own institutional product. Everything here about Turnitin comes from what it has said publicly.
How Turnitin looks at ChatGPT text
Turnitin added an AI writing indicator to its similarity product in 2023. In broad terms, its model reads a submission in segments and estimates how likely each segment is to have been produced by a large language model. The report then shows an overall percentage and highlights the passages it considers likely AI-generated.
ChatGPT is the most widely used of those models, so it's the one most people ask about. The model behind the indicator isn't reading for "ChatGPT" as a brand. It's reading for the statistical habits that language models share: predictable word choices, even rhythm, and smooth but generic transitions.
For a broader look at the indicator itself and how schools use it, see our companion article: does Turnitin detect AI?
Scenario by scenario: what tends to happen
No one outside Turnitin can tell you exactly how its model will score a particular document. What we can describe is the general pattern detectors show, which Turnitin's own guidance broadly reflects.
| How the text was produced | What tends to happen | Policy reality |
|---|---|---|
| Pasted straight from ChatGPT | Most likely to be highlighted | Almost always a problem if AI writing isn't allowed |
| ChatGPT draft, lightly edited | Often still highlighted in large parts | Still AI-produced work in most policies |
| Mix of your writing and AI paragraphs | AI segments may be highlighted; yours often aren't | Depends on whether mixing is allowed and disclosed |
| Your draft, ChatGPT fixed grammar | Usually less affected, but not predictable | Often allowed; check whether disclosure is needed |
| Heavily rewritten AI text | Less certain; may or may not be highlighted | Still AI-originated ideas and structure |
| Your writing, no AI at all | Usually low, but false positives happen | You're in the clear; keep evidence anyway |
Raw output
Unedited ChatGPT text is where detectors are strongest. The writing is fluent, balanced and predictable at the word level, which is exactly what these models are trained to recognize.
Light edits
Swapping a few words or adding a personal sentence at the start rarely changes the overall pattern of a long passage. Detectors look across many sentences, not at one phrase.
Mixed documents
When a paper combines your own paragraphs with AI ones, the indicator may highlight some segments and not others. Instructors see that unevenness, and it often prompts a conversation.
Heavy rewriting
The more thoroughly a person rewrites AI text, the more it carries human patterns, and the less confident any detector becomes. But here's what matters: if the argument, structure and evidence came from ChatGPT, the work still isn't yours in the way most assignments require. A lower score doesn't change that.
Why the score is only one piece of evidence
Turnitin has said its AI indicator shouldn't be the sole basis for an academic integrity decision, and has acknowledged that false positives can occur. Instructors who suspect AI use typically look at:
- Your other work. Does this essay sound like your in-class writing and earlier assignments?
- Your process. Can you show drafts, notes, and version history?
- Your understanding. Can you explain your argument and sources in conversation?
- The sources. Do the citations exist and say what the paper claims? ChatGPT has a known tendency to produce convincing but fake references.
- The prompt. AI answers to generic prompts often sound alike across a class.
That last point catches many students off guard. If several people ask ChatGPT the same essay question, the responses often share structure and examples, which an instructor reading thirty papers will notice.
Gray areas students ask about
Not every use of ChatGPT is clearly in or out. These come up often:
Translation. You write your essay in your first language and ask ChatGPT to translate it into English. The ideas are yours, but the English sentences were produced by a model, and they may read that way to a detector. Some courses allow this with disclosure; others, especially language courses, don't. Ask.
Outlines. ChatGPT suggests a structure, and you write every sentence yourself. Many instructors are fine with this, some want it disclosed, and a few consider the structure part of the assessed thinking.
Rewording one paragraph. You wrote a clumsy paragraph and asked ChatGPT to tidy it. That paragraph may now read differently from the rest of your paper. If your policy allows editing help, keep your original version so you can show what you started with.
Feedback you then act on. Asking "what's weak about my argument?" and fixing it yourself is about as low-risk as AI help gets, because every sentence remains yours.
When you're unsure, a short message to your instructor describing exactly what you plan to do will get you a clear answer. Most appreciate being asked.
Using ChatGPT within the rules
ChatGPT can be a useful study partner. Here's how to keep it that way:
- Find your course's AI policy in the syllabus or on the course page. If it's missing, ask.
- Use it before you write: to understand a concept, generate questions, or test your thesis against objections.
- Write your own draft in a document with version history.
- Use it for feedback on clarity or structure, then make the changes yourself.
- Never use it as a source. Find, read and cite real sources.
- Disclose when required. A short note about how you used AI is often all that's needed.
If you want to see how your own writing reads to an AI detector before you submit, a checker can show which sentences sound formulaic. Use that as feedback on style, not as a target score.
AI DetectorSentence-level AI scoreIf you're flagged and you didn't use ChatGPT
False positives are real, and they tend to hit certain kinds of writing harder: short answers, formulaic lab reports, heavily templated assignments, and writing by people working in a second language.
If it happens to you:
- Don't panic or get defensive. A flag starts a conversation; it isn't a verdict.
- Gather your evidence: version history, outlines, notes, browser history of your research, and earlier drafts.
- Ask for a meeting and walk your instructor through how you wrote it.
- Offer to discuss the content or write a short response on the topic under supervision.
- Know your rights. Your school's integrity policy describes the process and how to appeal.
Our guide on AI detection false positives goes through this step by step.
If you did use ChatGPT more than allowed
It happens. The best move is usually honesty. Talk to your instructor before it becomes a formal case if you can. Explain what you used and why, and ask whether you can redo the work. Many instructors respond far better to candor than to denial, and a concealment attempt usually makes outcomes worse.
Key takeaways
- Turnitin often flags raw or lightly edited ChatGPT text; heavy rewriting makes scores less certain.
- A lower score doesn't make AI-generated work acceptable under your course policy.
- Instructors use drafts, writing history and conversation, not just the percentage.
- Use ChatGPT for understanding and feedback, write your own draft, and disclose when required.
- If falsely flagged, your version history and notes are your strongest evidence.
Open your current assignment's syllabus and find the AI section now. Knowing the rule is the single best protection you have.