What AI Percentage Is Acceptable on Turnitin and GPTZero?
There's no universal acceptable AI percentage. See what Turnitin and GPTZero scores mean, how the 1% to 19% asterisk works and what to do if yours is high.

Key takeaways
- No AI detector sets an acceptable percentage. Your course's AI policy decides what's allowed.
- Turnitin shows a share of qualifying prose and GPTZero shows a probability for the whole document, so the same number means different things.
- Turnitin hides scores from 1% to 19% behind an asterisk (*%) because it says false positives are more likely there.
- Turnitin and GPTZero both say a score shouldn't be the only basis for discipline.
- If your score is high, ask what was flagged, gather drafts and version history, and talk to your instructor.
In this guide
- Is there an acceptable AI percentage?
- What does the AI percentage mean on each detector?
- What does Turnitin's asterisk (%) mean?
- How do GPTZero's confidence levels work?
- What do instructors typically do with a high AI score?
- Why can't schools agree on one threshold?
- What should you do if your AI score is high?
- How can you keep your AI score low without gaming it?
- The bottom line
There's no universally acceptable AI percentage. Turnitin, GPTZero and other detectors don't set a pass mark, and their numbers measure different things. Your school or instructor decides what counts as a problem, usually by reading the flagged text and looking at your process. What matters most is whether your AI use followed the course policy.
That can feel unsatisfying when you're staring at a number. So here's what each detector's percentage means, how Turnitin's asterisk works, what instructors typically do with a score, and what to do if yours is high.
Is there an acceptable AI percentage?
Not in any official sense. Turnitin says it "does not make a determination of misconduct" and that its percentage "should not be used as the sole basis for action or a definitive grading measure by instructors" (Turnitin's AI writing FAQ). GPTZero says it doesn't recommend "using an AI detection result as the only proof for academic punishment or discipline" (GPTZero).
So "acceptable" is set by people, not tools. Your syllabus might allow AI for brainstorming but not drafting, allow it with disclosure, or ban it for a specific assignment. Under a strict no-AI rule, the acceptable amount of AI use is zero, whatever a detector says. Under an open policy, a high score on properly disclosed work may not matter at all. Our guide to academic AI policies covers the common versions.
What does the AI percentage mean on each detector?
The same "40%" means different things depending on the tool.
| Detector | What the percentage shows | How to read a 40% result |
|---|---|---|
| Turnitin | Share of qualifying prose likely generated by AI or modified by AI tools | About 40% of the evaluated sentences were flagged |
| GPTZero | The model's probability that the whole document is AI or human | An estimate about the document, not 40% of the text |
| ZeroGPT | Meant to show how much of the text it considers AI | Sentences making up about 40% of the text were highlighted |
| QuillBot | Percentages for AI-generated, AI-refined and human-written text | Read each category, not just the total |
| Grammarly | Percentage of scanned text that's likely AI-generated | Grammarly calls it an average estimate |
| Originality.ai | Share of text that's Likely AI, Likely Original or Mixed | Check which sections are highlighted |
For tool-by-tool detail, see the Turnitin AI score explained, is GPTZero accurate, is ZeroGPT accurate and is QuillBot's AI detector accurate.
What does Turnitin's asterisk (*%) mean?
Since July 2024, Turnitin hasn't shown AI scores between 1% and 19%. If its detector flags some text but less than 20%, the report shows an asterisk (*%) with no number and no highlights. Turnitin's AI writing report guide says its testing found "a higher incidence of false positives" in that range, and the asterisk is there to reduce misinterpretation.
| What you see | What it means |
|---|---|
| 0% | No qualifying text was flagged as likely AI |
| *% | Some text was flagged, but under 20%, so no score or highlights appear |
| 20% to 100% | The share of qualifying prose flagged, with highlights |
| Gray, with no number | The file couldn't be processed, for example because it had fewer than 300 words of prose |
Two things follow. An asterisk isn't the same as zero, and it isn't evidence of anything either. Turnitin designed it so low, unreliable results don't get treated as findings. Reports generated before July 2024 may still show numbers below 20%.
The score can also undercount. Turnitin says it accepts missing some AI writing to keep false positives low, and that a document scored at 50% "could contain as much as 65% AI writing." For short documents of only a few hundred words, Turnitin says the prediction is mostly "all or nothing."
How do GPTZero's confidence levels work?
GPTZero's percentage is "our model's probability that your document was written by either AI or human," not a share of the text (GPTZero). A "4% AI" result doesn't mean 4% of your essay is AI. GPTZero also labels each result with a confidence level:
- Highly confident: GPTZero says its error rate is less than 2%.
- Moderately confident: around 10%.
- Low confidence: 14% or higher.
GPTZero says that when it isn't confident something is AI, it leans toward "human" with a low confidence score. So a low-confidence AI label is weak evidence, whatever the percentage looks like.
What do instructors typically do with a high AI score?
Policies differ, but Turnitin's own guidance points instructors toward judgment, not cutoffs. Its guide to reviewing the AI writing report says, "More important than any tool is the educator who sees the score and makes decisions," weighing it against what they know about their students and their institution's policy.
A careful review usually involves some mix of these steps:
- Reading the highlighted passages to see whether they sound like your other work.
- Comparing with earlier assignments and in-class writing.
- Asking about your process, including drafts, notes and version history.
- Talking with you before deciding anything.
- Following the school's integrity process if concerns remain.
Some schools have gone further. Vanderbilt University disabled Turnitin's AI detector in 2023, estimating that a 1% false positive rate could have wrongly flagged around 750 of the 75,000 papers it submitted the year before (Vanderbilt).
Why can't schools agree on one threshold?
- Different tools measure different things. A Turnitin share and a GPTZero probability can't share a cutoff.
- Length changes results. Short texts swing more, and Turnitin won't score anything under 300 words of prose.
- Mixed writing is hard to score. Turnitin says that in documents mixing human and AI text, "it can be difficult to exactly determine where the AI writing begins and original writing ends."
- Some writers get flagged more. A Stanford study found seven detectors labeled an average of 61% of TOEFL essays by non-native English speakers as AI (Stanford HAI).
- Assignments allow different AI use. A permitted, disclosed use and a banned one can produce the same score.
For the bigger picture, see why AI detectors have false positives and why is my writing flagged as AI.
What should you do if your AI score is high?
- Stay calm. A score is an estimate, not a finding, and most processes start with a conversation.
- Find out exactly what was flagged. Ask which tool was used, what the score was and which passages were highlighted.
- Gather your evidence. Version history from Google Docs or Word, outlines, notes, sources and earlier drafts carry more weight than a screenshot from another detector.
- Check the policy. If you used AI in an allowed way, explain what you did and show your disclosure.
- Be honest. If you used AI in a way that wasn't allowed, saying so early usually goes better than defending a story you can't support.
- Ask for a conversation. Offer to talk through your argument and how you wrote it.
Our guides on how to prove you didn't use AI and what to do if you're falsely accused of using AI go step by step.
How can you keep your AI score low without gaming it?
The goal isn't to hit a number. It's to make sure your writing reads like you, so a detector has less reason to misread it.
- Write the first draft yourself in a tool with version history.
- Vary your rhythm. Mix long and short sentences. Our sentence length checker shows when your rhythm is flat.
- Cut stock phrases. The AI word checker highlights wording that's common in AI output.
- Add specifics from class, your reading and your own experience.
- Run a quick check. Our free AI detector estimates AI-likeness from writing patterns. It can't predict your Turnitin score, but it shows which passages read as stiff.
If your course allows AI-assisted drafts, The Humanizer rewrites rhythm, structure and word choice while keeping your meaning and citations. It's designed to read as human, though results vary by text, and Turnitin now looks for text modified by humanizers, as we explain in can Turnitin detect AI humanizers. More tips are on our page for students.
The bottom line
No AI percentage is officially acceptable, because detectors don't set thresholds and their numbers mean different things. On Turnitin, scores from 1% to 19% show only as an asterisk, and even high scores are meant to start a review, not end one. What counts is your course's AI policy and the evidence of how you wrote.
If you want help making allowed AI-assisted writing read naturally, try The Humanizer free with 350 words every month.
Frequently asked
Questions, answered
Written by
The Humanizer Team
Editorial


