How Can Turnitin Detect AI Writing: A Complete Guide

Turnitin can flag qualifying prose that its model predicts may be AI-written or AI-altered, but the result does not prove misconduct. For students facing a question about a paper, that distinction matters: a score can prompt a conversation, but it cannot tell the whole story of how the work was made.

The AI Writing Score and Similarity Score measure different things. Institutions set their own rules for AI use and review, so educators must consider the assignment and writing process alongside the report. This guide explains how to read a result without treating a detector as a verdict.

A student’s assignment rules and writing process matter more than any detector score alone.

Key Takeaways

• Turnitin’s AI Writing Score is a screening signal, not proof of misconduct.
• The Similarity Score measures text matches, not AI use.
• Institutions decide how their academic integrity policies apply.
• Human review matters when a report raises a concern.
• Keep your drafts and follow the AI rules for each assignment.

How Can Turnitin Detect AI Writing?

Turnitin evaluates patterns in qualifying prose rather than searching a database for a copy of the submitted paper. Qualifying prose means complete sentences in paragraphs that Turnitin’s AI writing model can evaluate.

The model looks for statistical and linguistic patterns associated with generated text. Turnitin describes its AI writing detection architecture as a transformer deep-learning architecture. It makes a prediction about text, not an identification of who wrote it or which application was used.

Turnitin’s model predicts whether qualifying prose resembles AI-generated writing; it does not identify an author or prove intent.

How the model evaluates qualifying prose

To make that prediction, Turnitin divides a submission into overlapping text segments for evaluation. Overlap gives the model neighboring language to assess rather than treating every sentence as an isolated unit. The system analyzes those passages for patterns, then presents its prediction in a report for an educator to review.

This is a different task from finding a passage copied from a saved paper. A database match can point to text that appears elsewhere, while AI writing detection estimates whether prose fits patterns associated with generated writing. The two processes can produce separate results for the same submission.

What the report highlights

The AI Writing Report uses colored highlights to show which passages contributed to its prediction. Likely AI-generated passages appear in cyan. Passages classified as likely AI-generated and AI-paraphrased appear in purple.

Those colors help an educator locate text for closer review, but they do not reveal the exact writing process behind it. A student may have used a writing tool, revised generated text, or written the passage without AI. A highlight identifies a model prediction about a passage, not the tool that produced it or the student’s intent.

Which Submissions Qualify for a Turnitin AI Report?

A report requires at least 300 words of qualifying prose in an eligible file and language. Only qualifying prose is assessed, so a submission’s visible word count does not necessarily equal the amount the model evaluates.

Turnitin requires 300 words of long-form prose, a file under 100 MB, and no more than 30,000 words. The supported report formats are .docx, .pdf, .txt, and .rtf. These practical checks determine whether a submission can receive an AI Writing Report.

Check	Requirement	What it means
Minimum prose length	300 words	Long-form qualifying prose is needed
Maximum word count	30,000 words	A longer submission exceeds the limit
File size	Less than 100 MB	A file at or above 100 MB is too large
File types and languages	.docx, .pdf, .txt, .rtf; English, Spanish, Japanese	Report generation depends on file type and language

Turnitin’s 2026 general AI writing support described here covers English, Spanish, and Japanese. A separate March 2026 language list includes English, Spanish, Portuguese, French, German, Italian, and Japanese. Those differing 2026 lists should not be treated as one confirmed support list.

Turnitin’s 2026 description also limits AI-paraphrasing detection to English. Check which languages the institution supports for the assignment, especially if the paper contains more than one language. A language listed for general AI writing does not necessarily mean every report feature applies to it.

The model does not reliably evaluate poetry, scripts, computer code, bullet-point lists, tables, or annotated bibliographies. A mixed-format submission may contain enough prose for a report, while other elements remain outside its reliable assessment.

A paper built mostly from tables, code, or bullet points may not receive a reliable assessment of all its content.

What Does a Turnitin AI Score Mean?

A Turnitin AI Writing Score estimates the share of qualifying prose the model classifies as likely AI-written or AI-altered. The percentage describes text, not the chance that a student committed misconduct.

A 25%, 30%, or 40% result is not a universal misconduct threshold. The report’s display changes at 20%, but that threshold describes how Turnitin presents a result. It does not set a school’s disciplinary rule or establish what happened during drafting.

Reading 0% and the asterisk result

A 0% display means the model did not classify qualifying prose as likely AI-written or AI-altered. It does not prove that no AI was used. AI-assisted text may not be detected, and a student’s process cannot be reconstructed from a percentage alone.

Display	What appears	How to interpret it
0%	Zero detected qualifying prose	Not proof that AI was never used
*% below 20%	An asterisk instead of an exact score	Positive result with no precise percentage shown
20% or higher	Exact AI Writing Score and contributing highlights	Percentage estimate that can prompt review

When a detected result is above 0% but below 20%, Turnitin displays an asterisk instead of the exact score. The asterisk hides the precise number because false positives are more common below 20%. It signals a low-range result, not a settled conclusion.

Understanding scores of 20% and above

At 20% or higher, the AI Writing Report displays an exact AI Writing Score and highlights passages that contributed to it. The percentage estimates the share of qualifying prose classified as likely AI-written or AI-altered. It is not a probability of misconduct.

So, what is a red flag on Turnitin? A displayed score can flag passages for review, but it cannot establish misconduct by itself. A 25%, 30%, or 40% result does not carry the same meaning in every course, because the assignment policy and writing context matter.

Treat the displayed percentage as a reason to review the highlighted text, not as a verdict or a safe-use cutoff.

How Is the AI Writing Score Different From the Similarity Score?

The AI Writing Score estimates AI-like qualifying prose, while the Similarity Score measures matches with existing material. They answer separate questions, even when both appear in reports about the same paper.

A high Similarity Score does not automatically mean plagiarism, and an AI Writing Score does not prove unauthorized AI use. One measure estimates patterns in qualifying prose. The other finds text that matches material in Turnitin’s comparison databases.

Measure	What it compares or estimates	What it cannot establish
AI Writing Score	Share of qualifying prose classified as likely AI-written or AI-altered	Authorship, intent, or misconduct by itself
Similarity Score	Text matches with material in comparison databases	Plagiarism or whether a match is improperly used

A properly cited quotation can still match a source in the database. References, common phrases, or correctly quoted material may contribute to the Similarity Score, so a match needs context. SUNY Empire’s Similarity Score bands range from blue at 0%-20% through red at 81%-100%.

AI-generated wording may not closely match any saved source, so it can receive a low Similarity Score while still drawing attention in an AI Writing Report. The reverse is also possible: a paper written without AI may contain many properly cited matches. Neither percentage answers both questions.

Read each report on its own terms, since neither percentage alone establishes plagiarism or misconduct.

How Accurate Is Turnitin, and Can It Be Wrong?

Yes, Turnitin can be wrong in both directions: it can flag human writing or fail to flag AI-written passages. Accuracy depends on context, including the text being assessed and the conditions used to test a detector.

A false positive is a result that flags human writing as likely AI-generated. A false negative occurs when the detector misses AI-written text. Neither error can be ruled out by a score alone. Educators should consider the passages, assignment, and writing process before drawing conclusions.

Why a human passage can be flagged

Turnitin reports a false-positive rate below 1% for documents containing more than 20% AI-generated content under its specified testing conditions. That statement applies to that test setup, not every paper, language, or use case. It should not be turned into a universal accuracy claim.

A 2023 Stanford HAI study found that AI detectors misclassified over 61% of TOEFL essays by non-native English speakers as AI-generated. That finding is a reason for caution when reviewing multilingual students’ work, not a prediction about any individual student.

Language patterns can vary across writers and genres, and short or formulaic passages may give the model less context. The model’s performance can also change depending on whether text is original, heavily edited, or mixed with generated passages.

Why AI-written passages can be missed

False negatives can occur when AI-written text does not fit the patterns a model recognizes. Human revision, mixed authorship, and different kinds of AI assistance can affect a result. One university guide cites a study reporting that detectors can miss roughly 15% of AI-generated text in a document.

A 2026 study of 192 human, AI-generated, and hybrid texts reported Turnitin accuracy of 0.61. Hybrid writing was particularly difficult to classify. This result describes one study’s sample and methods, rather than a fixed rate for all submissions.

Language, genre, document length, and the kind of AI assistance can all change detection results. A model may assess an essay differently from code or a short list, while a long paper can contain both human and generated passages. Each result needs a careful review of what was actually submitted.

A detector score is strongest as one piece of context, not as a stand-alone measure of authorship.

What Changes When AI Writing Is Paraphrased?

Turnitin can flag some ChatGPT-like text, but no result guarantees that every output from a chatbot will be detected. Detection is a prediction, and changing the wording can change the patterns a model sees.

Turnitin cannot promise detection of every chatbot output or every paraphrased passage. A result does not name a specific application, nor does it show whether a student used a tool in a way the assignment allowed.

AI-generated text and specific chatbot outputs

Can Turnitin detect ChatGPT? It can flag text predicted to resemble generative AI output, including some ChatGPT-like writing. That prediction does not confirm ChatGPT wrote the passage. Other chatbots may produce similar language, and human writing can sometimes resemble patterns associated with generated text.

Human revision can change the passage enough to affect the model’s prediction. A submission can also mix student-written sentences with generated material, making the writing process harder to classify. A detected or undetected passage does not reliably describe who contributed each sentence.

AI paraphrasing and bypasser labels

Turnitin introduced AI-paraphrasing detection in 2024 and added detection aimed at AI-bypasser tools in 2025. These labels describe model predictions about text. They do not identify a particular application or prove that a student broke a course rule.

Ordinary editing, such as correcting a typo or tightening a sentence, is not the same as a model label for AI paraphrasing or bypasser use. Still, a passage that has been substantially rewritten with AI can affect the result. The report cannot determine what the student intended from the label alone.

Follow the assignment’s AI-use rules rather than treating a detector’s ability to recognize a passage as permission to use it.

How Can Students See AI Detection on Turnitin?

Students usually need access through their school, because Turnitin is generally not sold as a standalone consumer detector. AI report access depends on the institution’s settings and the assignment configuration.

A student-facing Similarity Report does not automatically include the AI Writing Report. If a school provides access, sign in through its Turnitin login or learning management system, then check which reports appear for that assignment.

Where student access comes from

Turnitin is generally an institutional product, so students usually access it through a college, school, or learning management system. The course instructor or institution decides which Turnitin tools students can use. A login may allow students to submit work without giving them access to every report.

SUNY Empire’s Fall Term 2026 Student Turnitin Portal lets students submit drafts to review Similarity Scores and reports before final assignment submission. That draft review is about similarity. It does not mean students can see the AI Writing Report in the portal.

What the student portal may show

When a student can open a report, the portal’s options show what that account can view. Look for a separate AI Writing Report rather than assuming an AI indicator sits inside the Similarity Report. If no AI report appears, the setting may not be enabled for students.

• Check the LMS assignment for its Turnitin link and report options.
• Open the institution’s Turnitin link using your school login.
• Review the report options shown for that submission.
• Ask the instructor or writing-support office if the AI report is not visible.

Confirm with the instructor whether the report is visible to students before assuming a Similarity Report includes an AI indicator.

What Should You Do If Turnitin Flags Your Paper?

A high AI score is a reason to ask for a review, not proof that you have been found responsible for misconduct. Start with the specific concern, then ask what course rule and review process apply.

Turnitin says its AI Writing Report should not be the sole basis for adverse action against a student. You can ask to review the report and the passages that prompted concern. Keep the discussion calm, factual, and focused on the assignment.

Ask what the concern is based on

Ask the instructor which passages raised a concern and what the report says about them. Then ask which assignment rule applies, including any directions about AI disclosure or permitted support. A specific passage and rule give you more useful information than a percentage alone.

You can also ask what process the institution follows for reviewing an AI concern. A course conversation may be informal at first, but the school’s published procedure explains how a formal assessment or appeal works. Make a note of deadlines and any documents requested.

Keep records of your writing process

Writing records can help you explain how your paper developed. Collect materials you already have, and describe your process honestly. Do not create drafts or records after the fact to make your work appear different.

• Assignment instructions, including rules for AI tools.
• Notes and outlines created while planning the paper.
• Source and citation records for research used.
• Drafts or version history that show changes over time.
• Any required AI-use disclosure or related notes.

If an informal discussion does not resolve the concern, use the institution’s formal review or appeal procedure. Explain what tools you used, what changes you made, and which parts you wrote yourself. Honest detail is more useful than trying to guess what a detector expects.

Keep the conversation factual and use the institution’s published procedure if an informal discussion does not resolve the concern.

Can You Use a Free AI Detector Instead?

A free third-party AI detector cannot reproduce a Turnitin report simply by giving a percentage. Each tool makes its own prediction, using its own model, access, and testing methods.

Students generally cannot run a standalone Turnitin AI check unless their institution provides access. A third-party AI detector result is not a Turnitin result. Free access also says nothing by itself about accuracy or privacy protections.

How Turnitin differs from consumer-facing tools

Turnitin is generally accessed through institutions, while tools such as Winston AI, Originality.ai, and Walter Writes AI are separate products. The language support and benchmarks described for those products do not make their scores interchangeable. Each detector may treat a paper differently.

A May 2026 peer-reviewed comparison reported standardized accuracy of 99% for Winston AI and 98% for Originality.ai on the study’s selected material. Those numbers describe that study’s test set and conditions. They are not a universal real-world ranking, and they do not directly establish how Turnitin compares.

Winston AI lists support for English, French, Spanish, Portuguese, German, Dutch, Polish, Italian, Indonesian, Romanian, and Simplified Chinese. A 2026 review reports that Walter Writes AI supports more than 80 languages but has no peer-reviewed benchmark. Language coverage alone does not show how a detector performs on your kind of paper.

What to check before trusting another score

Before uploading a draft to another detection tool, look beyond its headline percentage. A service may test text unlike a student essay, or handle submitted data in ways that matter to you. Check the product’s stated terms before sharing unpublished work.

• Supported language: confirm the tool covers the language of your text.
• Tested text type: check whether its evaluation includes essays like yours.
• Benchmark conditions: note the sample and method behind accuracy claims.
• Data handling: review what happens to text you upload.

Treat any alternative score as a separate screening result, not a prediction of what Turnitin or an instructor will decide.

How Can Students Use AI Within Course Rules?

The right way to use an AI tool depends on what the specific assignment permits. Course rules come first, because instructors may allow one kind of support while restricting another.

Grammar support and generated prose are different kinds of assistance. An assignment might allow grammar correction but prohibit AI-written paragraphs. Other policies may treat brainstorming, outlining, translation, paraphrasing, code, or data analysis separately.

ChatGPT is one example of a tool whose permitted use depends on the assignment. Students use AI responsibly by checking the specific instructions rather than assuming one course’s rules apply to another. If the policy is unclear, ask the instructor before drafting or making major changes.

• Identify which AI tasks the assignment allows.
• Check whether you must disclose or cite AI assistance.
• Verify facts and citations instead of trusting generated answers automatically.
• Keep a record of the assistance you used and how you used it.

Use AI only for tasks the assignment permits, and keep a clear record of that assistance.

If an instructor’s rule is unclear, ask before using AI to draft or substantially rewrite submitted work.

Conclusion

Turnitin can help identify passages for review, but the report cannot settle authorship or misconduct by itself. Its AI Writing Score estimates AI-like qualifying prose, so educators need to consider the passage and the assignment rather than relying on a number alone.

Students should follow the AI policy for each assignment, even when a detector does not flag their work. Before you begin, check the instructions and save drafts or version history as you write. Those records can help you explain how your work developed if questions arise.

Before your next submission, confirm the AI-use rules and keep a record of how your work developed.

FAQ

Can Turnitin detect AI-generated work?

Yes, Turnitin can flag qualifying prose that its model predicts may be AI-generated. The report marks passages for educator review, but it does not prove who wrote them or whether an assignment rule was broken. Human revision, mixed writing, and the text’s language or genre can affect the result. A flagged passage needs context, not an automatic conclusion.

Is 25% on Turnitin too high?

No, 25% is not automatically too high or a universal disciplinary cutoff. It is above Turnitin’s 20% display threshold, so the report shows an exact score and highlights contributing passages. The percentage estimates the share of qualifying prose classified as likely AI-written or AI-altered. Whether the use complied with course rules depends on the assignment and review.

Can Turnitin detect ChatGPT?

Yes, Turnitin can flag some writing that resembles generative AI output, including some ChatGPT-like text. It cannot guarantee that every ChatGPT passage will be detected, and a result does not identify ChatGPT as the specific tool. Human revision and mixed authorship can affect the model’s prediction. Course rules, not a detector’s success, determine whether use was allowed.

Is 30% on Turnitin ok?

A 30% AI Writing Score is not automatically acceptable or unacceptable. It estimates the share of qualifying prose the model classifies as likely AI-written or AI-altered, but it does not show whether that use complied with course rules. Review the highlighted passages alongside the assignment instructions. A score alone is not a finding of misconduct or a measure of a student’s intent.

Is 40% a bad Turnitin score?

A 40% result is a reason to review the highlighted passages, not a probability of guilt. The percentage estimates qualifying prose classified as likely AI-written or AI-altered. It does not establish a universal misconduct threshold, and institutions apply their own procedures. An educator should consider the paper’s context and the applicable course rule before reaching a conclusion.

Can Turnitin be wrong?

Yes, Turnitin can produce false positives and false negatives. A 2026 study of 192 human, AI-generated, and hybrid texts reported an accuracy of 0.61 for Turnitin, with hybrid writing particularly difficult to classify. That result applies to the study’s sample, not every assignment. A detector’s prediction should be considered alongside the actual writing and other relevant context.

Can Turnitin detect copy and paste?

Similarity checking can identify pasted text when it matches material in Turnitin’s comparison databases. Text with no database match may not be identified as copied, and a match alone does not prove plagiarism. Properly cited quotations can also match existing sources. Educators need to review the matched text, its citation, and how it is used before deciding whether there is a problem.