Turnitin Flagged Your Quoted Interview or Survey Answers as AI? Here Is What Is Actually Happening

Turnitin's AI detection does not skip quoted text. The official documentation describes a detector that extracts sentences from the submission and analyzes them with a language model — quotation marks are not an exclusion in any of the official pages.

A quoted participant answer that reads like fluent prose is analyzed like any other sentence in your paper. That is why a real person's words can show up as AI: the detector looks at sentence structure and word probability, not at who wrote the original.

Turnitin's FAQ on how the detection works is explicit about the unit of analysis:

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.

The unit is the sentence. The official page on the AI writing report adds the same point from the other direction: the percentage is based on prose sentences in a long-form writing format, and "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."

Nowhere in these descriptions is there a "quoted text" category. A sentence inside quotation marks is still a sentence to the detector.

Two properties of quoted participant data make it vulnerable:

The answer was given as speech, not written prose. Spoken answers, when transcribed, are often short, repetitive, or conversational. If you leave them in that shape, they may be treated as short-form or non-prose text — which the detector excludes. But most research papers integrate quotes into flowing prose: you introduce the participant, quote the answer, and comment on it. At that point the quote is a sentence inside a paragraph, and it gets analyzed like one.

Interview answers can be unusually consistent. A participant who answers a structured questionnaire tends to produce complete, evenly structured sentences. The official wording describes the detector as trained to distinguish "word probability sequences" of human and AI writers, and highly regular sentences are exactly the shape AI-generated text tends to take. The quote looks like the thing the model was trained to find.

The result is a mismatch: the words are a real person's, but their statistical shape resembles generated text.

The exclusion list in the official text is about text shape, not text origin:

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.

A short participant answer kept as a bullet or a one-line list item can fall outside the analysis. The same answer, rewritten into a flowing sentence in your paragraph, falls inside it. Quotation marks alone change nothing — what matters is whether the sentence reads as long-form prose.

There is no documented way to "protect" a quote from analysis other than changing its shape: keeping it as short-form text, or not including it at all.

If your similarity or AI report flags a quoted participant answer, the official guidance is the same as for any flagged sentence: treat the indicator as a signal, not a verdict. Turnitin's FAQ says the percentage "should not be used as the sole basis for action or a definitive grading measure."

Your options, in order:

• Check the highlighted sentence itself. Open the AI report and read the flagged text. If it is your participant's quote, you can verify it against your recording or transcript — that is your evidence.
• Decide whether the quote needs to be in its current shape. Only genuinely short-form text — a one-line answer kept as a list item — falls outside the analysis; a full sentence stays inside it whether or not you indent it as a block quote, because the model reads prose sentences regardless of formatting. So this only helps for answers that were short and list-shaped to begin with.
• Keep your records. Transcripts, recordings, and signed consent forms are the evidence that the words are not generated. The flagged quote plus your records is what a conversation with your instructor should rest on.
• Adjust only the surrounding text, not the quote. Rewriting the participant's words to change their shape alters the data. The defensible move is to preserve the quote as-is and reconsider its presentation, not its content.

Turnitin's AI detector analyzes sentences with a language model, and there is no quoted-text exemption in the official documentation. A participant's answer, when integrated as prose, is analyzed like any other sentence — and highly regular answers can statistically resemble generated text. The exclusions that do exist are about text shape (short-form, lists, non-prose), not about quotation marks. If a quote gets flagged, the indicator is a signal, not a verdict: verify the words against your records, reconsider the presentation, and keep the quote's content intact.