Does Turnitin Detect AI in Poetry, Scripts, and Annotated Bibliographies?

One Turnitin help article addresses this topic directly. It describes what the AI detection model does and does not cover:

"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."

That sentence names the formats the model does not reliably handle. Poetry, scripts, and annotated bibliographies are listed explicitly.

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

We will unpack both sentences below. The first one defines the coverage gap. The second one explains the visible symptom of that gap, which is the mismatch you may have seen in your own report.

The second sentence in the help article is the one that explains the confusion most students run into. When a document contains several different writing types, the percentage and the highlights will not line up.

Here is what that means in practice. The percentage is computed over the text the model actually analyzed. The highlights are the specific segments the model flagged. If a document mixes prose with poetry, script-style sections, or annotated bibliography entries, those non-prose parts do not get analyzed the same way. The percentage reflects the analyzed portion, and the highlights reflect what got flagged within it.

So you can end up with a report where the percentage suggests one thing and the highlighted passages suggest another. The article says this is the expected outcome for mixed-type documents, not a glitch. Reading the two against each other is the subject of how to read a Turnitin AI writing report.

The percentage is also not a percentage of the entire submission. It applies to the text that qualifies for analysis, and what qualifying text is in Turnitin AI detection explains where that boundary is drawn. When parts of the document are in a format the model does not reliably detect, those parts sit outside what the number is describing.

The help article lists three formats that are especially relevant to creative writing and research courses.

Poetry is non-prose by definition. Line breaks, fragmentary syntax, and compressed phrasing do not look like standard grammatical sentences, which is what the model analyzes. The article says AI-generated text in the form of poetry is not reliably detected.

Scripts are written as speaker names, stage directions, and dialogue exchanges. That structure is closer to a series of short fragments than to continuous prose. The same coverage gap applies.

Annotated bibliographies mix citation entries with brief commentary. The citation lines themselves are short-form and conventionally formatted, and the help article names annotated bibliographies as a format the model does not reliably detect.

None of this is a claim that these formats are always missed. The article's wording is "does not reliably detect." The point is that a score derived from a text that is mostly poetry, script-based, or built from annotated bibliography entries should be read with that limitation in mind.

If your assignment is a poem, a scene from a play, or an annotated bibliography, the AI percentage on your report has a limited amount to say about your work.

The detection model is built around prose sentences. When your submission is largely non-prose, the analyzed text is a small portion of what you actually wrote, or the model may struggle with the format altogether. A low or high percentage in that situation does not translate directly into "this writing is or is not AI-generated."

For instructors, the FAQ offers guidance about text that is hard to judge. It says: "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."

That sentence is about text with features that can trigger unreliable readings, and it is a reasonable frame for non-prose formats too. When the indicator shows a higher amount of AI writing in a poetry or script submission, taking the format into consideration is the documented approach.

The takeaway for students is simple: do not treat a creative writing AI score as a strong signal about your work. The reliable part of the report is the highlighted prose. The percentage is best read as a number tied to the analyzed text, not to the whole document.

If your poetry or annotated bibliography submission was flagged and you want to raise it, the documentation gives you concrete words to anchor the conversation.

You can reference the specific help article that names these formats. The article says the model does not reliably detect AI-generated text in the form of poetry, scripts, or code, and does not detect short-form or unconventional writing such as bullet points, tables, or annotated bibliographies.

You can also reference the second sentence of the same article, which says a document containing several different writing types will result in a disparity between the percentage and the highlights. That sentence is a documented explanation for the exact symptom you may be looking at.

It is worth being precise about the source. The list of formats comes from one Turnitin help article. The FAQ summarizes the coverage in a shorter form, but this specific list is what that article says. Keeping the attribution to the article keeps you accurate when discussing it.

Finally, the indicator was never meant to be read alone. Turnitin's FAQ says the percentage on the AI writing indicator should not be used as the sole basis for action. A creative writing score that came from an unreliable-format analysis is a weaker basis than most. The vendor's own instructions for that conversation are collected in what your instructor is told to do when your AI percentage is high.

If you want to see how your own prose reads under AI detection before it becomes a topic in class, you can copy a paragraph of your writing into a check. We offer a free trial for exactly that.