How Turnitin Handles Math Equations and Non-Prose Content

Math equations, formulas, and code blocks are not analyzed by Turnitin's AI detector. The FAQ states: "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 model also "does not reliably detect AI-generated text in the form of non-prose, or code." The FAQ adds: "In addition, we are not pursuing ChatGPT code detection at this time." This means your equations and formulas do not directly contribute to your AI score. The score reflects only the prose portions of your document. If your paper is mostly equations with brief explanatory text, the AI detector is evaluating a small amount of text, which can make the score less meaningful.

The FAQ defines what the model examines:

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

For a math or engineering paper, this means equations themselves are not evaluated. Only the prose sentences around them, the explanations, introductions, and discussions, are subject to AI detection. If you have a page full of equations with two sentences of explanation, those two sentences are what the detector evaluates. A rewriting pass raises the mirror-image question: what happens to citations, tables and equations in an AI humanizer.

The FAQ is direct about non-prose content:

"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 FAQ also states: "In addition, we are not pursuing ChatGPT code detection at this time."

This has a practical implication for STEM papers. If you used an AI tool to generate code or equations, the AI detector is not designed to catch that. The detector evaluates the surrounding prose, not the code or math itself.

The detection model processes qualifying text by splitting it into overlapping segments:

"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."

Segments overlap, which means sentences can receive multiple scores that are pooled together. In a paper with dense equations and sparse prose, the overlap advantage diminishes because there are fewer prose sentences to segment. This can make the score less stable, as there are fewer data points for the model to work with.

A release note from August 2023 describes how the model handles tables:

"We are now able to process long-form prose text in tables."

The next sentence: "Resubmit to reprocess existing submissions that contain tables."

This means if your table contains long-form prose sentences, those sentences are now analyzed. But raw data, numbers, and short labels in tables are not prose sentences and fall outside the qualifying text definition. Bibliographies are also excluded: the same release note says "Bibliographies are now excluded when processing the AI writing report." The next sentence again requires resubmission for existing submissions, which is why an older report can still look like Turnitin flagged your references.

One common source of confusion is whether math content flagged in a similarity check also affects the AI score. It does not. The FAQ states:

"The Similarity score and the AI writing detection percentage are completely independent and do not influence each other."

If your equations match a source in the similarity database, that affects your similarity score, not your AI score. The AI detector evaluates only the prose for AI-generated patterns, regardless of what the similarity check finds; reading the two scores as the same thing is the mistake to avoid here.

To summarize what we have covered:

• Math equations and formulas are not analyzed by the AI detector. Only qualifying prose sentences are.
• The FAQ says the model does not reliably detect non-prose or code, and is not pursuing code detection.
• Tables with long-form prose are now processed, but raw data and short labels are not.
• Bibliographies are excluded from AI detection processing.
• The AI score and similarity score are completely independent and do not influence each other.
• Papers with dense equations and sparse prose have less text for the model to evaluate, which can make the score less stable.

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