Systematic Reviews and AI Detection: Why Methodology Sections Get Flagged

Systematic reviews demand a standardized methodology section. You describe the databases searched, the search strings, the inclusion and exclusion criteria, and the screening process. The language is necessarily repetitive because the PRISMA checklist requires specific reporting elements. Turnitin's documentation describes text types that produce false positives:

"Sometimes false positives (incorrectly flagging human-written text as AI-generated), can include content without a lot of structural variation, text that literally repeats itself, or text that has been paraphrased without developing new ideas."

The next sentence: "If our indicator shows a higher amount of AI writing in such text, we advise you to take that into consideration when looking at the percentage indicated."

A methodology section hits all three criteria, which is why Turnitin flagged my whole methodology section is such a common complaint. The structural variation is low because every systematic review reports the same steps. The phrasing repeats ("we searched," "we identified," "we screened"). The text is paraphrased from protocol descriptions without developing new ideas. The documentation says to take the percentage into consideration when the text matches these patterns, not to accept it as definitive.

The detection pipeline breaks qualifying text into segments and scores each one:

"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. 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."

In a systematic review, the methodology section's repetitive patterns could produce high segment scores across multiple overlapping segments. Those scores aggregate into a high document-level percentage. An 800-word methodology section could pull up the score for an entire 5000-word review. If you end up revising it, what you can rephrase and what must stay exact marks the lines you cannot move.

The detector does not evaluate burstiness, perplexity, or other named metrics, a claim examined in does Turnitin use perplexity and burstiness to detect AI:

"Our model is not explicitly programmed to evaluate specific signals such as 'burstiness,' 'perplexity,' or other individual metrics sometimes referenced in public discussions."

The next sentence: "Instead, it learns statistical patterns from our training data."

The same source explains what the model actually does:

"Our classifiers are trained to detect these differences in word probability and are adept at the particular word probability sequences of human writers."

The detector evaluates word probability sequences. It checks whether the words in a segment match patterns learned from human writing or from AI-generated text. In a methodology section, the word sequence is driven by the reporting template. Phrases like "we searched the following databases" and "studies were included if" appear across thousands of papers. These sequences may align more closely with statistical patterns from training data than with an individual writer's word choices. That alignment can produce a high probability score without AI involvement.

The detector processes only qualifying text:

"This qualifying text includes only prose sentences, meaning that we only analyze blocks of text that are written in standard grammatical sentences and do not include other types of writing such as lists, bullet points (short non-sentence structures), or other non-sentence structures."

The next sentence: "This percentage is not necessarily the percentage of the entire submission."

Systematic reviews contain tables (PRISMA flow diagrams, extraction tables) and structured lists (inclusion and exclusion criteria). These are not qualifying prose. The percentage covers only prose sentences. If your results section is mostly tables, the percentage is weighted toward prose-heavy sections like methodology. The documentation also notes:

"The model does not reliably detect AI-generated text in the form of non-prose, or code, nor does it detect short-form/unconventional writing such as bullet points (short non-sentence structures)."

The next sentence: "This means that a document containing several different writing types would result in a disparity between the percentage and the highlights."

A systematic review is a multi-format document. The disparity between the percentage and the highlights is expected behavior.

Methodology sections are sometimes broken into short subsections. A search strategy subsection might be only 200 words. The documentation warns:

"In shorter documents where there are only a few hundred words, the prediction will be mostly 'all or nothing' because we're predicting on a single segment without the opportunity to overlap."

The next sentence: "This means that some text that is a mix of AI-generated and original content could be flagged as entirely AI-generated."

For a short subsection, the detector may predict on a single segment with no overlap to moderate the score. A human-written but template-following subsection could receive an all-or-nothing high score.

In 2023, Turnitin also addressed false positives at document boundaries:

"Since launch, we have observed a higher incidence of false positive detection in the first few or last few sentences of a document. Many times these sentences consist of introduction or conclusion content written in a generic way. As a result, we have changed our detection logic to help reduce these false positives."

The next sentence: "We also worked on making our segment boundaries detection more precise which could lead in some rare cases to change of boundaries compared with a previous version."

This was a 2023 improvement. Systematic review introductions and conclusions follow templates (background, gap, objective, findings summary). The logic change addressed this pattern.

Turnitin's documentation is explicit about the limitations:

"Our AI writing detection model may not always be accurate (it may misidentify human-written, AI-generated, and AI-paraphrased text), so it should not be used as the sole basis for adverse actions against a student."

The next sentence: "It takes further scrutiny and human judgment in conjunction with an organization's application of its specific academic policies to determine whether academic misconduct has occurred."

A systematic review with a high AI score on its methodology section is not proof of misconduct. The methodology section matches false-positive patterns. The detector is a word probability classifier. The score requires human judgment and institutional policy. What your instructor is told to do when your AI percentage is high is the guidance the marker is working from.

To summarize:

• Methodology sections match false-positive patterns: low structural variation, repetitive phrasing, and paraphrased text without new ideas.
• The detector is a word probability classifier, not a burstiness or perplexity evaluator.
• Only qualifying prose is analyzed. Tables, lists, and bullet points are excluded.
• Short methodology subsections face an all-or-nothing scoring problem.
• In 2023, Turnitin improved detection logic to reduce false positives at document boundaries.
• The score should not be used as the sole basis for adverse actions. It requires human judgment.

If you receive a Turnitin AI report on a systematic review and want to address the flagged passages, import the report and work on them. Eligible passages can be re-run at no charge.