What Percentage of AI Detection Is Acceptable?

You ran your essay through a checker and it came back 30% AI. Or your instructor mentioned a number in an email and now you can't stop thinking about it. So what percentage of AI detection is acceptable?

The honest answer is that there isn't one. Neither Turnitin's nor GPTZero's guidance names a "safe" score. What matters is what your course allows, what the specific tool's number actually measures, and whether you can show how you wrote the work. This guide walks through all three, then covers what to do at different score levels.

There's no universal "acceptable" AI score

Turnitin, the tool most students worry about, doesn't set a pass mark. Its own FAQ says Turnitin does not make a determination of misconduct and that the percentage "should not be used as the sole basis for action or a definitive grading measure by instructors." Its guidance for teachers on what to do when the percentage is high starts with "refer to your institution's policy" and describes the score as a starting point for a conversation with the student.

That shifts the real question. It isn't "what number is safe?" It's "what does my course allow, and can I show my work?" A course that bans AI entirely and a course that allows AI for brainstorming will treat the same score very differently. And a low score doesn't make prohibited AI use acceptable, just as a high score on work you wrote yourself doesn't prove anything.

What the numbers actually mean

Here's the part most score charts online skip: "% AI" means different things in different tools. A 40% in one detector is not the same claim as a 40% in another.

Turnitin: a share of your sentences

According to Turnitin's FAQ, its model gives each sentence a score between 0 and 1. Sentences above a threshold are labeled AI, and the document score is the percentage of sentences above that threshold. A few details matter a lot:

• It only counts "qualifying text." That means long-form prose sentences. Lists, bullet points, headers and references are left out, so the percentage isn't necessarily a share of your whole submission.
• It needs enough text. English submissions need at least 300 words of prose to get a score at all.
• Low scores show as an asterisk. Since a July 2024 update, results in the 1% to 19% range display as "*%" with no number and no highlights. Turnitin says it did this "to avoid potential incidence of false positives."
• The false positive target has conditions. Turnitin says it aims to keep false positives under 1% for documents with over 20% AI writing, which works out to roughly one in every 100 fully human documents. It also says it may miss some AI text to keep that rate low, giving the example that a document scored at 50% "could contain as much as 65% AI writing."

Turnitin also lists the kinds of human writing that can trip it up: content without much structural variation, text that repeats itself, and text that's been paraphrased without developing new ideas. If you write in a very formulaic style, that's worth knowing.

GPTZero: a classification with a confidence level

GPTZero works differently. It sorts a document into one of three classes, AI, Mixed or Human, and the percentage is how confident the model is in that classification. GPTZero's help article spells it out: "100% AI" means the model is confident the whole document is AI, not that every sentence was written by AI.

Each result also comes with a confidence label tied to an error rate. GPTZero says "highly confident" means an error rate under 2%, "moderately confident" around 10%, and "uncertain" 14% or higher. When it's uncertain, it says the document would likely be considered human. In its guide to reading scan results, GPTZero says it doesn't recommend using a detection result "as the only proof for academic punishment or discipline."

Why "30% AI" isn't one thing

Tool	What the percentage refers to	Source
Turnitin	Share of qualifying prose sentences above the AI threshold. 1% to 19% shows as *%.	Turnitin FAQ
GPTZero	Confidence in a document-level class (AI, Mixed or Human), plus a confidence label.	GPTZero help
Originality.ai	Earlier models gave confidence scores. Version 5.0 estimates how much of the text is Likely AI, Original or Mixed.	Originality.ai

So a "30% AI" from GPTZero and a "30%" from Turnitin aren't saying the same thing, and neither is a measure of how much you cheated. If you want the longer story on Turnitin specifically, read does Turnitin detect ChatGPT.

What universities actually say

Several universities have gone further than the detector companies. A few examples:

• The University of Pittsburgh's Teaching Center recommends against AI detection tools, saying they're "not accurate enough to prove" a policy violation, and it has disabled the AI detection tool in Turnitin.
• The University of Toronto's teaching centre says AI detection software is unreliable and biased against non-native English writers, and that U of T doesn't support the use of these tools.
• Cal State Fullerton's faculty development center describes the tools as unsuitable for use as evidence of academic misconduct.

The bias point comes up a lot, and it has research behind it. A 2023 study by Weixin Liang and colleagues, "GPT detectors are biased against non-native English writers", found that detectors consistently misclassified non-native English writing samples as AI-generated. If English is your second language and you've been flagged, you're not imagining the problem.

None of this means your school ignores detectors. Policies vary, even between departments, so find your institution's academic integrity policy and your course syllabus, and read what they actually say about AI use and about how flags are handled.

What to do at different score levels

These bands aren't official cutoffs from any tool or school. They're just a practical way to think about where you stand. No score guarantees a particular outcome.

Under 20% (or an asterisk)

On Turnitin, this range doesn't even show a number. Turnitin hides scores here to avoid false positives. You don't need to rewrite anything. Keep your drafts and notes like you normally would, and move on.

20 to 50%

This is the range where people panic most, and where the score tells you the least. Turnitin itself says writing with little structural variation or repeated phrasing can produce false positives, and research has found non-native English writing gets flagged more often. If an instructor raises it, they'll most likely want to talk about how you wrote it. Gather your version history and notes now so you're ready. Don't start swapping words around to chase a lower number. That can make your writing worse and your edit history look stranger.

Over 50%

A high score is more likely to start a conversation or a review, but it's still not a verdict. Turnitin, GPTZero and the universities above all say a score alone isn't proof. If you wrote the work yourself, stay calm, collect your evidence, and ask which tool was used and which passages were flagged. Our step-by-step guide for when your essay is flagged as AI covers the meeting and the appeal in detail.

If you did use AI in a way your course didn't allow, the honest path is to talk to your instructor and follow your school's process. Trying to edit or rewrite the text until it slips under a threshold doesn't fix the underlying problem, and Turnitin says it now looks for text run through AI paraphrasers and bypasser tools anyway.

How to document your writing process

The best protection against a bad score isn't a lower score. It's a clear record of how you wrote the work. Start this before anyone asks:

• Write in a tool with version history. Google Docs keeps a version history and lets you name versions, so you can mark points like "outline done" or "first full draft."
• Keep your messy stuff. Outlines, brainstorm notes, rejected intros, photos of handwritten plans. Rough material is hard to fake and easy to explain.
• Save your sources as you go. A reading list with dates, highlighted PDFs, or library search history shows the research behind your argument.
• Log any AI use your course allows. If you're permitted to use AI for brainstorming or grammar checks, note what you asked and what you kept. Being upfront beats being found out.
• Know your own argument. Be ready to explain why you made a claim, why you chose a source, or what you'd change. That conversation is often more convincing than any file.

Should you check your own score first?

It can help to see what a detector makes of your draft, and you can try our free AI detector for a rough read. Just keep its limits in mind. No outside checker, ours included, can tell you what your instructor's tool will say, because each one uses a different model and a different kind of percentage. Use it to spot passages that read flat or formulaic, then fix them for the reader's sake, not for the number.

WriteHumanly's other tools are built for AI-assisted drafts in courses and jobs that allow them: making the text read naturally in your own voice while keeping every fact. They aren't a way around your school's rules, and no tool can promise a particular detector result.

Frequently asked questions

Is 30% AI detection bad?

Not by itself. On Turnitin, 30% means about 30% of the qualifying prose sentences scored above its AI threshold. On GPTZero, a percentage is the model's confidence in a classification. Either way, the tools themselves say a score shouldn't be the sole basis for action. What matters is your course policy and whether you can show how you wrote it.

Is 50% AI bad?

It's more likely to get noticed, but it isn't proof. Turnitin's own FAQ uses 50% as an example of a score that could reflect as much as 65% AI writing, which shows how rough the number is, and it leaves the final decision to the instructor. If you wrote the work, gather your version history and notes and be ready to talk through your process.

Can I see my own Turnitin AI score?

Usually not. Turnitin's FAQ says only instructors and administrators can see the AI writing indicator. A number from another checker is a different measurement, not a preview of your Turnitin result.

Does a 0% score mean I'm in the clear?

A low score just means the tool didn't flag much. It says nothing about whether your AI use followed the rules. Your course policy decides what's allowed, not the detector.