Why Are Different Sections of My Paper Flagged as Different AI Percentages? Turnitin's Detection Pipeline Explained

Different sections of the same paper get different AI scores because Turnitin scores the document one sentence at a time, then pools those sentence scores into the overall percentage. Turnitin's own guide describes the pipeline: sentences are extracted, cut into overlapping sections, each section is scored, each sentence inherits a score, and the sentence scores are aggregated into the document number. A short, dense section and a long, conversational one will not look the same in the Submission Breakdown — that is the mechanism working as designed, not a fault in the report.

The official "Using the AI Writing Report" guide describes how the detection works:

"When a paper is submitted to Turnitin, sentences from the submission are extracted and segmented into overlapping sections for prediction analysis."

"Each segment is classified by the AI detection model and given a value between 0 and 1, denoting the probability of the text being likely human or AI-generated."

Turnitin's own blog adds the granularity detail:

"The submission is first broken into segments of text that are roughly a few hundred words (about five to ten sentences). Those segments are then overlapped with each other to capture each sentence in context."

So the report you are reading was built from the smallest unit upward: sentence-level scores, then pooled into section-level views, then aggregated into the single percentage at the top.

The official guide continues:

"Each qualifying sentence within these segments inherits the segment's score. Since segments overlap, some sentences may have multiple scores, which are then pooled into a single score. These sentence scores are further aggregated and used to compute the overall document AI writing score."

Three consequences fall out of this mechanism:

• Sections are scored by what the sentences in them look like — a section full of short, template-like sentences scores differently from one with long, varied prose.
• Section length matters: the Submission Breakdown shades each area of your document by how much of its qualifying text was flagged. A 5-page section can visibly carry more highlighted text than a 1-page one, even when the per-sentence pattern is similar.
• Overlap pooling means a sentence sitting at the boundary of two sections can carry a blended score — so the exact border of a highlight is not a precise verdict boundary.

None of this is a report error. It is the aggregation working the way Turnitin documents it.

One of the most common misreadings is comparing the AI "score" to the similarity percentage as if they were two readings of the same thing. Turnitin's guide states the separation explicitly:

"The percentage generated by Turnitin's AI writing detection model is different from and independent of the similarity score."

"AI writing highlights are not visible in the Similarity Report."

A section with 40% similarity and an AI score of 0% is not contradictory: one number counts matched text, the other reflects the detection model's per-sentence judgment. Different sections of a document can therefore differ in AI percentage for the same reason the AI and similarity percentages can differ from each other — they are answering different questions.

The official guide describes the interactive bar at the top of the report:

"The overall percentage of text likely detected as AI is detailed in the Submission Breakdown."

"The bar is interactive and allows you to select each highlight to bring the corresponding text and page of the submission into focus."

Two practical notes when you read your own report:

• The highlight color changed in August 2026: Turnitin merged the AI-generated and AI-paraphrased categories into a single blue color, so newer reports show one color where older reports showed two.
• Only qualifying prose participates in the calculation — lists, bullet points, tables, poetry, scripts, and code do not count. The official guide says: "The model does not reliably detect AI-generated text in the form of non-prose, such as poetry, scripts, or code, nor does it detect short-form/unconventional writing such as bullet points, tables, etc."

So an "empty" section is sometimes not empty of text — it is empty of qualifying prose. The report tells you nothing about bullet points or tables because it never counts them.

Turnitin's guide on indicator states covers the percentages that do not behave like a plain number:

"To avoid potential incidence of false positives, no score or highlights are attributed for AI detection scores above 0% and below the 20% threshold in the report. When this occurs, it is now indicated with an asterisk (*%) and no percentage is attributed."

"0% detected as AI … our testing has found that there is a higher incidence of false positives when the percentage is between 0 and 19."

This changes how you read a section-level difference: a section that shows 0% may be a true zero or a sub-20% score the report refuses to surface — the report itself cannot tell you which. That is by design, not by accident.

A Turnitin AI report that shows different percentages across sections is the detector's pipeline doing exactly what Turnitin documents: sentences are scored individually, overlapping segments pool the scores, and the overall number is an aggregate — not a uniform verdict across your document. Sections differ because their sentence character and length differ. The AI percentage is independent of the similarity score, and highlights only cover qualifying prose, so bullet points, tables, and code stay out of the calculation entirely. When you read your own report: treat the highlights as a lead to investigate, not as a fixed label on each paragraph.