The short answer
There is no universal acceptable percentage of AI. No number is accepted by every school, journal or employer, and no detector’s score is a rule. What makes work acceptable is whether you followed the AI policy that applies to it.
Three things follow from that:
- The percentage is an estimate. It is what one tool thinks about the text. It is not a measurement of what you did.
- The policy is the rule. A course that allows no AI use is not asking for a low score. It is asking you not to use AI.
- The same text gets different numbers in different tools, so a number only means something next to the tool that produced it.
What the percentage is
An AI score is a detector’s estimate of how likely a piece of text is to have been written by an AI model. The detector looks at patterns in the writing. It cannot see you typing, and it does not know which tools you had open.
It also helps to know that “percent AI” does not mean the same thing everywhere. Depending on the tool, a figure can describe:
- how confident the tool is that the whole text is AI-written;
- how much of the text the tool marked as likely AI-written;
- an average of scores given to individual sentences.
Those are different statements. “30%” in one report can mean something quite different from “30%” in another. That is one reason a sentence-level view is more useful than a single figure: it shows where the tool is looking, so a person can read those passages and judge. The Plagiarism Checker Plus AI detector gives each sentence a likelihood of being AI-written plus an overall score.
For the signals behind a score, see how AI detectors work.
What decides “acceptable”
The policy you are working under decides it, not the score. Policies are written about what you may do, and they vary a lot, even between courses at the same school.
| If the policy says | What is acceptable | What the score has to do with it |
|---|---|---|
| No AI use | Work you produced without AI tools. | A low score does not make AI use allowed. A high score does not show you used it. |
| AI for some tasks | The uses the brief names, such as brainstorming or grammar checks. | The score cannot tell an allowed use from one that is not allowed. |
| AI allowed with disclosure | AI use that you describe and cite as required. | A high score may be expected. What matters is that you disclosed it. |
| Nothing, or unclear | Not decided yet. Ask before you submit. | None until you know the rule. |
Publishers and employers set their own rules in the same way. Check the author guidelines or the workplace policy. For the common policy types and scenarios, see is using AI plagiarism?
Some tools do set their own display thresholds, and some institutions set internal guidance for when a report gets a closer look. Those are decisions about how a tool or an office works. They are not a general standard. For one documented example, our guide on whether Turnitin detects AI explains how that tool shows its score, with links to its documentation.
Why a low score is not proof of innocence and a high score is not proof of misconduct
Any detector can be wrong in both directions. It can flag writing a person produced, and it can miss text a model produced. So the number cannot settle the question on its own, whichever way it points.
A low score
A low score means the tool did not find the patterns it looks for. It does not show that no AI was used, and it does not make AI use acceptable where the policy forbids it. Teachers know this, and they also look at drafts, earlier work and how well a student can discuss what they wrote.
A high score
A high score means the tool found those patterns. It does not show how the text was written. Human writing can score high, particularly when it is short or very formal. A high score is a reason for someone to read the flagged passages and ask questions. It is not a finding.
Why different tools give different numbers
Different detectors are different products. Running the same essay through two of them and getting two numbers is normal. Common reasons include:
- They are built and trained differently. Each tool has its own model and its own examples of human and AI writing.
- They report different things. As above, one figure may be a confidence and another a share of the text.
- They set different cut-offs. Each tool decides how sure it must be before it marks a sentence.
- They handle the text differently. Length limits, and how quotations, references and headings are treated, vary between tools.
- They are updated at different times. The same tool can give a different result after an update.
So you cannot use one tool to predict another. A check you run yourself will not tell you what your instructor’s report will show, and a result from one tool cannot overrule a result from another. They are separate estimates.
For students: what to do with a score
- Find the rule first. Read the assignment brief and the syllabus. If they do not say what AI use is allowed, ask.
- Do not aim for a number. Follow the policy. If you used AI where it is allowed, say so and cite it. See how to cite ChatGPT.
- Keep your process. Write in a document that keeps version history, and keep your notes, outline and sources.
- Read a flag as a question, not a verdict. If your own writing is flagged, look at which sentences and keep your drafts ready. You do not need to rewrite honest work to change a score.
- Ask to see the report if your work is questioned, along with the policy it is being judged against.
The first check per tool on Plagiarism Checker Plus is free up to 1,000 words without an account, and a free account includes 3,000 words a month. Your text is never used to train AI. There is also a page on using the AI detector for essays.
For teachers: what to do with a score
- Write the policy in terms of use, not percentages. Say which uses of AI are allowed and how students should disclose them. A percentage rule invites arguments about the tool.
- Read the flagged passages. A sentence-level report shows where to look. The overall figure does not.
- Look for other evidence. Drafts, version history, earlier writing and a conversation tell you more than a number. See how teachers check for AI.
- Talk to the student before you conclude anything. Our guide on talking to a student about plagiarism covers the conversation.
- Follow your institution’s procedure, and keep a note of what you saw and what the student said.
Plagiarism Checker Plus gives a plagiarism check and a sentence-level AI check in one report. See the plagiarism and AI checker for teachers.