How to Start Lowering a Turnitin AI Score: What to Do with the Report First

The short answer: do not open your document and start rewriting from the first paragraph. Open the AI writing report first. It shows you which passages were flagged, and that decides where your effort goes.

This page is for people whose score is high but whose writing is mostly their own. If your text started as AI-generated and you are trying to hide that, this is not the page for you — what follows is about making work you actually wrote read as yours, and about knowing what the score does and does not mean.

The AI writing report is a separate thing from the similarity report, and it answers a different question. The similarity report compares your text against published sources. The AI writing report estimates how much of your document reads like machine-written text, based on the statistical properties of the writing itself.

What you can find in it: a percentage for the whole document, and — when the percentage is at or above 20% — highlights on the specific passages the model judged likely to be AI-written. Below 20%, Turnitin shows an asterisk (*%) instead of a number, and no highlights, because the company says avoiding false positives is the reason no score is attributed in that range.

Whether you can see any of this depends on your school. Only instructors and administrators can see the indicator; students cannot. But instructors can download the AI report as a PDF and share it with you — that is the normal way a student ends up holding the report in their hands.

Do not just glance at the percentage. Read the report as a map of where your effort belongs.

First pass: the percentage. It tells you how big the problem is, not where it is. A 40% score suggests that a share of qualifying sentences roughly in that proportion sit in flagged segments — the specific ones are in the highlights, not in the number.

Second pass: the highlights. When the report shows a score of 20% or more, it highlights the passages the model considered likely AI-written. Those passages are your editing target set. Everything outside them is, at this moment, not what the model is flagging.

Third pass: the pattern. Look at what the highlighted passages have in common. Are they all introductions? All methodology? All paragraphs that start the same way? The pattern tells you what kind of edit matters — and the report is your evidence when you talk to your instructor, because it puts the flags next to the text that carried them.

Editing effort is not evenly valuable. The passages with the most highlights are the ones carrying the most weight in the pooled score, so changing them first gives you the largest return per hour.

This works because of how the score is built. Turnitin extracts sentences from the submission, segments them into overlapping sections, and gives each segment a probability between 0 and 1. Every qualifying sentence inherits the score of the segment it sits in. Because segments overlap, a sentence can carry several scores, which are pooled, and the document percentage is aggregated from those.

Two things follow. One: a sentence's score comes from the block of text around it, not from the sentence alone. Two: the paragraph — not the word — is the smallest unit you can meaningfully change.

So the order is: pick the most heavily flagged passage, rewrite it as a passage, then move to the next one. Do not scatter single-word edits across the document.

Turnitin publishes the features that produce false positives — text that is human-written but gets flagged anyway. That list is the closest thing to an official editing specification:

• text without a lot of structural variation
• text that literally repeats itself
• text that has been paraphrased without developing new ideas

Check the flagged passages against those three. Monotonous sentence lengths and parallel openings are structural variation problems. Repeated phrasing across sections is a repetition problem. Text that restates the argument without adding anything new is a no-new-ideas problem.

The company also says that when its indicator shows a higher amount of AI writing in text of this kind, it advises taking that into consideration when looking at the percentage. In other words: for writing in exactly these three shapes, the official position is that the score should be discounted. That sentence is worth remembering when you talk to your instructor.

If your submission is short — a few hundred words — two things change about how you should read the report.

First, the prediction tends to be mostly "all or nothing". Turnitin says that in shorter documents with only a few hundred words, the prediction will be mostly all-or-nothing, because the model is predicting on a single segment without the opportunity to overlap. The next sentence is the one that matters: some text that is a mix of AI-generated and original content could be flagged as entirely AI-generated. So a short flagged document is not necessarily mostly AI — and its highlights do not map onto "which parts are the problem" the way they do in a long document.

Second, the number itself is less reliable below 300 words. Turnitin's own release notes say results show accuracy increases with a little more text, so submissions that contain less than 300 words may result in an AI writing score that is likely less accurate.

For short documents, the paragraph-based editing order above still applies, but expect the flags to be coarse. Your best move is usually to rebuild the flagged passages rather than to fine-tune single sentences, then re-check.

The score does not update on its own. Turnitin says detection capabilities change over time, but for a submitted document the AI percentage changes only if it is re-submitted and processed again. If your school allows a re-submission before the deadline, that is the mechanism you use: revise, re-submit, read the new report.

If your school does not allow re-submission, the report is still useful — but as evidence for a conversation, not as a score to fix. A flagged report on a final submission is a starting point for talking to your instructor about how the score was produced, not a verdict on your integrity.

And remember the access rule from earlier: you see the report only if your instructor shares it. If they have not, ask for the PDF — it is a normal request, and it is the concrete thing that turns a scary number into a list of passages you can actually work with.

None of the editing above is about "getting away with something". It is about the fact that the percentage should not be treated as a judgment of your work.

Turnitin's own guidance is explicit that the AI writing indicator should not be used as the sole basis for action or a definitive grading measure by instructors. Another official page adds the same point from the other side: the AI writing detection model may not always be accurate, so it should not be used as the sole basis for adverse actions against a student — determining whether misconduct occurred takes further scrutiny and human judgment together with the institution's own policies.

And in the words of the official guidance itself, the score is not meant to provide definitive answers in isolation. When educators look at the AI writing score and use it as a single data point rather than a definitive response, it is being used as intended.

So a high percentage is a signal that your prose matches statistical patterns the model associates with machine writing — and those patterns also appear in careful human writing. What you do with the report — edit the flagged passages, know what the number means, and bring the report to the conversation — is the part that is up to you.

Read the report before you touch the document. The percentage tells you the size of the problem, the highlights tell you where it is, and the pattern tells you what kind of edit matters. Change the most-flagged passages first, check them against the three features Turnitin names for false positives, and expect short documents to behave differently. Re-submit to re-check, or use the report as the basis for a conversation. The score was never meant to be the verdict — that is Turnitin's own position, and it is the strongest thing you can bring to the discussion.