Can Turnitin Detect AI? How Its Detection and Report Work

Yes, Turnitin can identify long-form prose that its model considers likely AI-generated or AI-generated and later paraphrased. The AI Writing Report is separate from the Similarity Report and is not proof of misconduct. Turnitin says the result requires further scrutiny, human judgment, and the institution's academic policy.

Turnitin AI Detection at a Glance

The most authoritative description is Turnitin's Using the AI Writing Report guide, updated in March 2026.

How Does Turnitin Detect AI?

Turnitin says it applies an AI writing detection model to qualifying text: prose sentences in long-form work such as an essay, article, or dissertation. The model classifies portions of that prose as likely AI-generated, likely AI-generated and then AI-paraphrased, or not identified as AI. The report then shows the share of qualifying text assigned to the detected categories.

This is classification, not source matching. Turnitin does not describe the AI percentage as a search for copied ChatGPT passages, and the percentage is independent of the Similarity Score. Its current documentation does not publish enough detail to reproduce the proprietary model or justify claims that it relies on one simple metric such as “perplexity” or “burstiness.”

The documented workflow is:

1. Check eligibility. The file must meet Turnitin's type, size, language, and long-form prose requirements.
2. Identify qualifying prose. The report focuses on prose sentences rather than treating every word, table, bullet, quotation, or code block as equivalent.
3. Run the classification model. Turnitin evaluates qualifying text for patterns its model associates with AI generation and, for the English detector, AI paraphrasing or bypasser modification.
4. Calculate the document-level result. The displayed percentage represents the portion of qualifying text identified by the model, not the probability that the author used AI.
5. Highlight review areas. For surfaced results, the Submission Breakdown distinguishes likely AI-generated text from likely AI-generated and AI-paraphrased text.
6. Require human review. Turnitin instructs educators to treat the output as one data point alongside drafts, context, policy, and discussion with the writer.

Turnitin's model release notes also show that detection changes over time. Recent releases mention changes to recall, false-positive controls, language models, paraphrasing detection, bypasser detection, and segment boundaries. A report is therefore a result from a particular model version and submission—not a permanent verdict about a passage.

What Turnitin Does Not Publicly Disclose

Turnitin publishes report behavior, eligibility rules, detection categories, and selected model updates. It does not publish the complete production architecture, feature weights, training data, sentence thresholds, or a formula that outsiders can use to calculate the same score.

That distinction matters because many explanations online present guessed mechanisms as established facts. Statements such as “Turnitin flags any sentence below a specific perplexity value” or “a fixed percentage of short sentences will pass” are not supported by the current official guide. The defensible answer to how Turnitin detects AI is therefore high-level: a changing proprietary classifier evaluates qualifying prose and returns review signals with known limitations.

What Turnitin's AI Writing Percentage Means

Turnitin defines the percentage as the amount of qualifying text in a submission that its model considers likely AI-generated or likely AI-generated and then modified with an AI paraphrasing tool. “Qualifying text” means prose sentences in a long-form format such as an essay, article, or dissertation.

The result does not mean the same thing as text similarity. Similarity compares submitted writing with source material; AI detection classifies patterns in qualifying prose. A paper can therefore have a low Similarity Score and a separate AI indicator—or the reverse.

The Submission Breakdown currently distinguishes AI-generated text, including text that may have been modified by a bypasser, from AI-generated text likely changed by an AI paraphraser. Reviewers should inspect the highlighted passages and surrounding argument rather than relying on the headline percentage alone.

Which Files Can Generate a Turnitin AI Report?

Turnitin's current English-language documentation says a submission must:

• be smaller than 100 MB;
• contain at least 300 words of prose in a long-form writing format;
• contain no more than 30,000 words;
• use a supported language: English, Spanish, or Japanese;
• use an accepted file type: .docx, .pdf, .txt, or .rtf.

Poetry, scripts, code, bullet points, tables, and annotated bibliographies are examples of non-prose or unconventional writing that the model does not reliably detect. A document containing several writing formats may therefore show a percentage that does not map neatly to the entire file.

Always verify requirements in the live Turnitin AI Writing Report guide, because supported models, languages, and limits can change.

Why Turnitin Shows an Asterisk Below 20%

Turnitin says its testing found a higher incidence of false positives between 0% and 19%. For current submissions, a result in that range is displayed as an asterisk instead of an exact percentage, and the report does not show highlights. This is intended to reduce misinterpretation of a less reliable result.

That design is an important reminder: a detector threshold is not a statement of authorship. Turnitin explicitly acknowledges that its model can misidentify human, AI-generated, and AI-paraphrased text.

How to Review a Turnitin AI Writing Report

1. Confirm the report applies to the document. Check the file type, language, length, and whether the highlighted material is qualifying long-form prose.
2. Separate AI detection from similarity. Do not describe the AI percentage as a plagiarism score.
3. Inspect highlighted passages in context. Look at the surrounding argument, quotations, citations, and assignment instructions.
4. Review the writing process. Examine outlines, research notes, version history, earlier drafts, and instructor feedback.
5. Discuss the work with the student. Ask them to explain the sources, reasoning, and revision decisions.
6. Apply the actual policy. Determine which forms of brainstorming, drafting, grammar support, or AI assistance were permitted.
7. Document a human conclusion. Turnitin says the report should not be the sole basis for an adverse action.

Turnitin's separate report-review guidance describes the score as one data point rather than a definitive response.

Does Turnitin Detect ChatGPT and AI Paraphrasing?

Turnitin says its AI Writing Report can identify text likely generated by large language models and can categorize likely AI-paraphrased text. Its release notes also describe model updates for newer language models and the inclusion of likely AI-bypasser modifications in the report.

That does not establish a fixed detection rate for every ChatGPT version or document. Model releases, editing, genre, language, and qualifying-text requirements affect the context. Avoid converting provider statements into a guaranteed percentage for an individual paper.

See Turnitin's AI writing detection model release notes for current updates.

What to Do If Human Writing Is Flagged

Preserve the original document before revising it. Save the report, outlines, source notes, earlier drafts, comments, and version history. Ask the reviewer to consider that evidence and the applicable institution's process.

Do not deliberately add errors or repeatedly transform authentic writing to chase a lower number. That can weaken the paper and erase evidence showing how it developed. Use the AI detector false-positive response guide for a practical evidence checklist.

Where Humanizer PRO Fits

Humanizer PRO helps revise permitted AI-assisted drafts for clearer and more natural expression. It cannot guarantee a Turnitin result, and it should not be used to conceal prohibited assistance.

For academic work, check the institution's AI policy first, compare the revision with the source, verify every quotation and citation, and keep your own reasoning central. The responsible essay humanizer guide explains that workflow in detail.

Frequently Asked Questions

Can Turnitin detect AI writing?

Yes. Turnitin provides an AI Writing Report for qualifying long-form prose and highlights text its model considers likely AI-generated or AI-paraphrased.

How does Turnitin detect ChatGPT?

Turnitin applies its AI writing classifier to qualifying prose and marks text the model considers likely generated by a large language model. It does not publish a complete ChatGPT-specific formula, and a highlight does not prove which tool produced the text.

Does Turnitin use perplexity and burstiness?

Turnitin's public report guide does not define its production model as a simple perplexity-and-burstiness test. Those concepts are common in general explanations of AI detection, but they should not be presented as Turnitin's disclosed scoring formula.

Is the Turnitin AI score the same as plagiarism?

No. The AI writing percentage is separate from and independent of the Similarity Score.

Why does Turnitin show an asterisk?

Current scores between 1% and 19% are shown as an asterisk because Turnitin says that range has a higher incidence of false positives and is less reliable.

Can a Turnitin report prove academic misconduct?

No. Turnitin says its model may make mistakes and should not be the sole basis for adverse action. Human judgment and institutional policy are required.

Can Humanizer PRO guarantee a low Turnitin score?

No. No humanizer can guarantee a result across model updates, languages, document types, and institutional workflows.

Sources

• Turnitin: Using the AI Writing Report
• Turnitin: How should I review the AI Writing report?
• Turnitin: AI writing detection model updates
• Turnitin: What should I do if the AI Writing score is high?