Detection guide
Turnitin AI Detection Guide: How It Works and How to Read Your Report
A practical overview based on Turnitin's publicly available documentation — written for students and researchers who want to understand results before they submit.
Turnitin is widely used by universities to support academic integrity. Its AI writing detection and similarity checking tools are related but measure different things: one estimates how much qualifying text may have been produced or modified by generative AI tools; the other shows how much of your submission matches text in Turnitin's databases and other sources.
This guide explains how Turnitin describes its detection process, what the numbers and colors in your reports mean, and where common misunderstandings arise. TurnitChecker is an independent service that provides access to AI writing and similarity report-checking services. TurnitChecker is not affiliated with or endorsed by Turnitin, LLC.
How Turnitin AI Writing Detection Works
According to Turnitin's public materials, submitted documents are analyzed in segments of text rather than as one undifferentiated block. Segments overlap so that sentences are considered in context.
Each segment is evaluated by Turnitin's AI writing detection model, which estimates how likely the text was written by a human versus generated by AI. Using qualifying text across the document, Turnitin produces an overall estimate of how much of the submission may be AI-generated.
Turnitin emphasizes that this output is an estimate — not proof of misconduct. Educators are advised to use it as one signal within a broader review process that includes their institution's academic integrity policies and human judgment. Segmentation rules, scoring methods, and thresholds may change as Turnitin updates its models.
What text qualifies for AI detection?
Turnitin's AI writing indicator applies to qualifying text: prose sentences contained in a long-form writing format (such as essays and papers). Headers, bullet lists, tables, and other non-prose elements may not contribute to the AI percentage in the same way.
Supported languages
AI Writing Reports are currently available for qualifying prose in English, Spanish, and Japanese. Documents written primarily in other languages may not generate an AI Writing score. Similarity checking may support a broader range of languages.
AI Report Categories and Highlights
In current Turnitin AI writing reports, detected text may be broken down into interactive categories that show how AI tools may have been used:
- AI-generated only — qualifying text likely produced by a large language model (LLM), sometimes further modified by an AI bypasser. Highlighted in cyan in the report.
- AI-generated text that was AI-paraphrased — text likely produced by AI and then revised using an AI paraphrasing or word-spinner tool (e.g., Quillbot). Highlighted in purple in the report.
Understanding the AI Writing Percentage
The overall AI percentage reflects the proportion of qualifying prose that Turnitin's model estimates could be AI-generated, AI-paraphrased, or modified using AI bypass tools. It is not a measure of textual overlap and does not indicate how much of your work is "original" in a citation sense.
Since July 2024, Turnitin adjusted how low-range scores are displayed to reduce the risk of false positives:
- 0% — displayed as 0%
- 1%–19% — displayed as *% (an asterisk percentage; no specific number is shown)
- 20% and above — the exact percentage is displayed
Understanding the Similarity Report
The Similarity Report is separate from the AI writing report. It shows what percentage of your submission matches text found in Turnitin's databases, publications, previously submitted student papers (depending on repository settings), and other indexed sources.
Turnitin calculates similarity by dividing the number of matching words by the total words in the document. Matched passages are highlighted in the submission. Properly quoted and cited material may still appear as matches — the score is a starting point for review, not an automatic verdict.
Similarity score color ranges (classic view)
In many classic Turnitin integrations, icon colors correspond to similarity ranges. Turnitin notes that colors may vary by LMS or product integration — always rely on the numeric percentage and highlighted matches, not the color alone.
- Blue — no matching text (0% in classic assignments view)
- Green — 1% to 24% matching text
- Yellow — 25% to 49% matching text
- Orange — 50% to 74% matching text
- Red — 75% to 100% matching text
Match groups in newer reports
Newer Similarity Report experiences group matches by citation and quotation status — for example, text that is not cited or quoted, missing quotations, missing citations, or properly cited and quoted. This helps distinguish formatting issues from potential integrity concerns.
Supported Formats and Submission Requirements
On TurnitChecker, you can check your work by pasting text or uploading a file. To ensure accurate processing, follow these TurnitChecker upload requirements:
- Accepted file formats: .pdf and .docx
- Word count: 400 to 28,000 words per submission
- Maximum file size: 10 MB
- Document type: essay or thesis-style long-form writing. Presentations, scanned image-only PDFs, and non-prose formats may not process reliably
- Languages: English, Spanish, and Japanese are supported for AI detection on this service; English is recommended for the most consistent results
Common Misconceptions and Limitations
- AI detection is not 100% accurate. Turnitin explicitly states the model may misidentify human-written, AI-generated, and AI-paraphrased text.
- A high similarity score does not automatically mean academic misconduct — it may reflect correctly quoted material, bibliographies, common phrases, or institutional template text.
- A low or 0% AI score does not guarantee a paper was written without AI assistance — heavily edited, paraphrased, or hybrid human-AI workflows can reduce detection signals.
- Turnitin recommends that AI indicators should not be used as the sole basis for adverse actions against a student. Human review and institutional policy always apply.
- Detection models evolve as new AI tools emerge. Scores may change over time as Turnitin updates its models.
- Non-native English writers and highly formal academic prose have been associated with higher false-positive rates in industry discussions; Turnitin's sub-20% masking policy partly addresses this.
Practical Tips Before You Submit
- Run a check before your final institutional submission — especially if you used any writing assistance, translation tools, or heavy editing software.
- Review highlighted AI passages in context. Ask whether flagged sections match your actual writing process.
- For similarity reports, check whether matches are properly quoted and cited according to your style guide (APA, MLA, Chicago, etc.).
- Keep drafts and revision notes. If questioned, documentation of your writing process is as important as the score itself.
- Follow your instructor's or institution's stated thresholds. There is no universal "safe" percentage across all schools.
- Use TurnitChecker's private checking workflow to understand your results before your institutional submission.
Ready to check your document?
Upload your paper on TurnitChecker and review AI and similarity insights before your final submission.
Start a check →