What is the Difference Between the Turnitin AI Score and the Similarity Score?
Table of Contents
- How Does Turnitin Calculate the AI Writing Score?
- What Does Turnitin's Similarity Score Indicate?
- Why Do Some Documents Have a High Similarity Score But a Low AI Score, or Vice Versa?
- Frequently Asked Questions
- Sources
- Related articles
Direct Answer – The Turnitin AI score and the similarity score are two independent metrics on a Turnitin originality report. The AI score indicates what percentage of a document's prose text was likely generated by an artificial intelligence tool such as ChatGPT, Claude, or Gemini. The similarity score, by contrast, measures how much of the document matches content in Turnitin's vast databases, including web pages, previously submitted student papers, and academic publications [1]. A submission can score high on one and low on the other—or high or low on both—because the two scores evaluate entirely different dimensions of academic writing integrity.
How Does Turnitin Calculate the AI Writing Score?
Turnitin's AI detection model works at the sentence level, analyzing linguistic patterns, sentence structure, and statistical probabilities to determine whether segments of prose were machine-generated [2]. Instead of searching for external matches (as the similarity score does), the AI detector examines the provenance of the text itself. It is trained on a broad corpus of academic writing and AI-generated content derived from large language models, enabling it to flag text that exhibits characteristics common to AI output.
The AI score is reported as a percentage of the document's prose that the model predicts was AI-generated. For example, an AI score of 60% means 60% of the prose in the submission is flagged as likely produced by an AI tool like GPT-4 or Claude [2]. Crucially, non-prose elements—bullet points, tables, mathematical equations, code snippets, and lists—are excluded from the AI evaluation, so only full-sentence writing is assessed. Turnitin also displays a *% label for any score below 20%, meaning the detector found insufficient or inconclusive AI-generated text to report a numeric result [1]. The system maintains a reported false positive rate of less than 1% for documents with more than 20% AI writing, based on Turnitin's internal validation studies [2].
What Does Turnitin's Similarity Score Indicate?
The similarity score works on an entirely different principle. Rather than evaluating how the text was written, it checks where the text appears elsewhere. Turnitin compares every submission against an enormous database that includes billions of current and archived web pages, a repository of student papers submitted through Turnitin globally, and a collection of academic journals, conference proceedings, and published books [3]. The result is a percentage representing the portion of the document that matches one or more sources in that database.
Scores are color-coded for quick interpretation: blue (0% similarity), green (1–24%), yellow (25–49%), orange (50–74%), and red (75–100%) [3]. Instructors can drill into the similarity report to see highlighted matches alongside their original sources, making it possible to distinguish between properly referenced citations, direct quotes, and unoriginal text that may need attribution. Unlike the AI score—which analyzes only prose—the similarity score examines every character type, including quotes, references, headings, and even boilerplate institutional disclaimers. Turnitin explicitly notes that the similarity score is not a plagiarism verdict; it is an originality indicator that requires human judgment to interpret contextually [3].
Why Do Some Documents Have a High Similarity Score But a Low AI Score, or Vice Versa?
This apparent contradiction arises because the two scores measure fundamentally non-overlapping attributes. The AI score concerns authorship attribution—whether the text was composed by a human or an AI system. The similarity score concerns textual overlap—whether the text matches pre-existing content in Turnitin's databases [4]. A document can easily exhibit one without the other.
Consider a student who writes an original research paper entirely by hand, with no AI assistance, but includes properly cited block quotes from several published journal articles. That submission will likely receive a low or *% AI score (since the prose is genuinely human-written) but a moderate or high similarity score (since the quoted passages match published sources in the database) [4]. Conversely, a student might ask ChatGPT to generate an analysis of a novel, producing entirely original sentences that do not match any web content or existing papers. That submission would receive a high AI score (potentially 80–100%) but a low or even 0% similarity score, because the AI-generated text is original in phrasing—it just wasn't written by the student [4].
Turnitin recommends reviewing both reports side by side to get a complete integrity picture. A high AI score with low similarity may indicate AI-generated original text; low AI with high similarity may indicate well-sourced human writing; and high scores on both warrant a closer look at both the authorship and the originality of the content [1]. Neither score alone tells the full story, which is why Turnitin displays them together on the report cover sheet.
Once you understand the difference between these two scores, the next logical step is to check your own submission before your instructor does. Turnitin0 gives you access to the exact same Turnitin AI writing report and similarity report that professors see in their institutional systems—so you can review both scores side by side, identify flagged passages, and take action before the official submission deadline. No subscription is required; pay per check and see your results in minutes.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
Frequently Asked Questions
1. Can a document have a 0% similarity score but a 100% AI score?
Yes. If a student asks an AI to generate original text on a unique topic, the AI-generated sentences will not match any existing source in Turnitin's databases (0% similarity), but the prose itself will be flagged as AI-generated (100% AI score). This is the most direct illustration of why the two scores measure separate things [1][4].
2. Does Turnitin combine the similarity score and AI score into one overall number?
No. Turnitin always reports them separately on the report cover. The similarity score appears as a colored percentage block, and the AI score appears in a separate tile next to it. There is no combined or aggregate score because they evaluate distinct dimensions of academic integrity [1][3].
3. What should I focus on if my assignment has a high AI score but a low similarity score?
Focus on the AI score first. A high AI score indicates that Turnitin's detector found substantial portions of prose that appear to be AI-generated. Even if the text is original and doesn't match existing sources, your institution may still apply AI use policies that require human-authored work. Consider the context of your assignment's rules about AI tools [2][4].
4. Can the similarity score be affected by AI-generated text that includes copied content?
Yes, but this is not the norm. Most AI models generate original phrasing rather than copying verbatim. However, if an AI tool reproduced existing web content in its output, that portion would appear in both the similarity score (as a match) and the AI score (as AI-generated). In practice, this combination is rare because modern LLMs are designed to avoid exact reproduction [1][3].
5. Does a low AI score guarantee that my similarity score is also low?
Not at all. A human-written paper that incorporates multiple direct quotes, references a large bibliography, or includes boilerplate text (such as university disclaimers) can have a low AI score but a moderate or even high similarity score. This is perfectly normal for research papers that cite sources extensively [3][4].
Sources
- Turnitin AI Writing Detection Frequently Asked Questions — https://guides.turnitin.com/hc/en-us/articles/28477544839821-AI-Writing-Detection-Frequently-Asked-Questions
- Using the AI Writing Report — https://guides.turnitin.com/hc/en-us/articles/22774058814093-Using-the-AI-Writing-Report
- Understanding the Similarity Score — https://guides.turnitin.com/hc/en-us/articles/23004110826253-Understanding-the-Similarity-Score
- Academic Integrity and AI Writing: Understanding the Difference Between Similarity and Authorship — https://www.turnitin.com/blog/academic-integrity-and-ai-writing-understanding-the-difference-between-similarity-and-authorship
Related articles
- Pre‑Defense Turnitin Checks: What Committees Look for Beyond the Percentage
- Turnitin Scan
- Chatgpt Turnitin Detection and Turnitin AI Detection: What Students Should Know Before Upload
- pay per scan Turnitin checker no subscription
- Chatgpt Turnitin Detection and Turnitin Chatgpt Detection: What the Report Actually Shows?