What Content Does Turnitin AI Skip? Non-Prose, Code, and More
Not everything in your document gets analyzed by Turnitin's AI detector. The FAQ says the model only analyzes qualifying prose sentences. One help article lists poetry, scripts, and annotated bibliographies as types it does not reliably detect. The FAQ also says Turnitin is not pursuing code detection. Here is what counts as qualifying text and what does not.
HumanPen Team
· 4 min read
The Short Answer
Turnitin's AI detector does not analyze every part of your document. 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." One Turnitin help article provides a broader list: "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, or annotated bibliographies." This means your AI score reflects only the prose sentences in your document, not the full content.
What Counts as Qualifying Text
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."
The percentage you see is not the percentage of your entire submission. It is the percentage of qualifying prose text only. If your document is 50% equations, 20% bullet points, and 30% prose, the AI score reflects only that 30%. How to read a Turnitin AI writing report covers where that gap shows up on the page.
Code and Non-Prose Content
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 does not establish that every code block is excluded from processing. The official claim is narrower: AI detection is not reliable for code, and Turnitin says it is not pursuing ChatGPT code detection at this time. Surrounding long-form prose can still be qualifying text. A rewriting tool has to respect the same document boundaries: what happens to citations, tables and equations in an AI humanizer.
Poetry, Scripts, and Annotated Bibliographies
One Turnitin help article provides a broader list of content types the model does not reliably detect:
"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, or annotated bibliographies."
The next sentence: "This means that a document containing several different writing types would result in a disparity between the percentage and the highlights."
This list comes from a single help article, not the FAQ. For poetry, script-style dialogue and annotated bibliography entries, the published claim is unreliable detection, not a promise that every such section is completely excluded.
Tables and Bibliographies
A release note from August 2023 describes updates:
"We are now able to process long-form prose text in tables."
The next sentence: "Resubmit to reprocess existing submissions that contain tables."
The same release note also says: "Bibliographies are now excluded when processing the AI writing report." Again, the next sentence requires resubmission for existing submissions.
This means long-form prose in tables can be processed. Short labels and raw values do not become qualifying prose merely because they sit inside a table. Bibliographies are excluded from AI-report processing after that model update; an older submission has to be resubmitted to pick up the change. Why Turnitin may appear to flag references separates that case from Similarity Report matches.
Excluded, processed, or unreliable are three different statuses
The official pages use three different verbs. Keeping them separate prevents “not reliably detected” from being rewritten as “never processed.”
| Content shape | Current official wording | What can safely be concluded | Source date / checked |
|---|---|---|---|
| Standard grammatical prose sentences | The capabilities FAQ says qualifying text “includes only prose sentences” in standard grammatical blocks. | These are the text units included in the AI percentage | Current page / 28 August 2026 |
| Lists, bullet points and short non-sentence structures | The same FAQ says qualifying text does not include these writing types. | They are outside the qualifying-text denominator described by the FAQ | Current page / 28 August 2026 |
| Code | The FAQ says the model “does not reliably detect” AI-generated code and that Turnitin is “not pursuing ChatGPT code detection at this time.” | Detection is not reliable; the page does not promise categorical parser exclusion | Current page / 28 August 2026 |
| Poetry, scripts and annotated bibliographies | The AI Writing Report guide places them in its list of forms the model does not reliably detect. | Do not treat an AI result as reliable coverage of these forms | Current page / 28 August 2026 |
| Long-form prose inside tables | The model change log says, “We are now able to process long-form prose text in tables.” | It can be processed; older submissions need resubmission for the August 2023 change | 9 August 2023 / 28 August 2026 |
| Bibliographies | The same change log says, “Bibliographies are now excluded when processing the AI writing report.” | This is an explicit AI Writing Report exclusion; older highlighted reference sections need resubmission | 9 August 2023 / 28 August 2026 |
Which exclusions belong to which report
Turnitin's two reports use different exclusion rules. The report name in the source is the boundary:
| Item | What the official page says | Report and control |
|---|---|---|
| Bibliographies | The AI model change log says, "Bibliographies are now excluded when processing the AI writing report." It tells readers to resubmit older files to reprocess them. | AI Writing Report; this is the one explicit AI-report exclusion in the pages checked |
| Quoted text and block quotes | The Similarity Report filter guide says its filter ignores text in quotation marks or formatted as block quotes. | Similarity Report filter; not an AI Writing Report exclusion |
| In-text citations and their associated sentences | The same Similarity Report guide says the cited-text filter removes a recognised citation and its associated sentence. It also says this setting is unavailable in the classic report. | New Similarity Report filter; availability depends on the report version |
| Footnotes and endnotes | The three official pages checked for the AI Writing Report do not document either term as an exclusion. The current Similarity Report filter guide also does not list a dedicated footnote or endnote switch. | No published basis to treat them as automatically excluded from the AI Writing Report |
For the negative check on 28 August 2026, the fixed corpus was Turnitin's "Using the AI Writing Report" guide, its "AI writing detection capabilities FAQs," and its "AI writing detection model" change log. For each page we took `document.querySelector('article').innerText` without normalising whitespace, then summed the three raw lengths rather than joining the pages first: 6,492 + 30,987 + 9,233 = 46,712 characters. Case-insensitive substring counts for quotation, quote, footnote, endnote, citation, verbatim, statute, law and legal were all zero. The same extraction returned prose 13 times, qualifying 15 times and bibliograph 4 times, so the zero was not an empty-page result. This does not prove how an undocumented parser treats every layout. It proves only that the public AI-report pages do not promise those exclusions.
How the Mechanism Works on Qualifying Text
The detection model processes qualifying prose 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."
Only qualifying prose sentences are extracted and segmented. Non-prose content is not included in this process. The document-level percentage is computed from the qualifying prose segments only.
What This Means for You
To summarize what we have covered:
- The AI detector only analyzes qualifying prose sentences written in standard grammatical sentences.
- Lists, bullet points and short non-sentence structures are outside the qualifying-text definition.
- For code, the official claim is unreliable detection and no current pursuit of ChatGPT code detection, not a categorical exclusion promise.
- The FAQ says Turnitin is not pursuing code detection.
- One help article lists poetry, scripts, and annotated bibliographies as not reliably detected.
- Long-form prose in tables can be processed; short labels and raw values still have to meet the qualifying-prose rule.
- Bibliographies are excluded from AI detection processing.
- The percentage reflects only qualifying prose, not your entire submission.
If you receive a Turnitin AI report and want to address the flagged passages, import the report and work on them. Eligible passages can be re-run at no charge.
KEEP READING