Turnitin Similarity Score vs AI Score: What Each One Means

Turnitin can show you two numbers on the same submission, and they answer two different questions. The similarity score tells you how much of your text matches sources already in Turnitin's database. The AI writing score is a separate estimate of how much of your qualifying prose a model thinks may have been generated by AI. Neither number is a verdict of plagiarism or misconduct; both are prompts for a human to look closer.

Most panic about Turnitin comes from mixing the two up. This guide takes them one at a time, then shows how to read them side by side.

The similarity score: what it actually counts

Turnitin's own guide is blunt about this: the similarity score is simply the percentage of text in a submission that matches other sources. Those sources include web pages, previously submitted student papers, and published journals and periodicals. The same guide says Turnitin does not check for plagiarism. It finds matches. Deciding whether a match is a problem is the instructor's job.

That distinction matters because plenty of matching text is perfectly fine:

• Direct quotes. Even with quotation marks and a reference, quoted text still shows as a match unless the instructor has chosen to exclude quotes.
• Your bibliography. Reference lists are full of standard titles and journal names, so they match other papers that cited the same work.
• Common phrases and technical terms. Method names, standard definitions and set phrases in your field will match other papers.
• Your own earlier drafts. If a draft was submitted to the same system, the final version can match it. Instructors can exclude previous submissions.

Turnitin's guide gives an example that sums it up: one paper scored 20% because of text copied from the web, while another scored 22% because of properly referenced quotes. The higher number was the honest paper.

Is there a "safe" similarity percentage?

Turnitin does not publish one. Some departments set their own thresholds, and they vary by course, level and subject. A literature-heavy dissertation with many quotes will naturally score higher than a reflective essay. If your course has a limit, it will be in your handbook or assignment brief; ask your supervisor if you cannot find it.

The AI writing score: a different model, a different question

The AI writing report runs a separate detection model. According to Turnitin's documentation on AI writing detection, the percentage is calculated only on qualifying text: prose sentences in long-form writing, such as the paragraphs of an essay or article. It is not necessarily a percentage of your whole document. The model is not built to assess bullet points, tables, poetry, code, scripts or annotated bibliographies.

Three details students often miss:

• An asterisk instead of a number. When the detected share is between 1% and 19%, Turnitin shows *% with no number and no highlights, because results in that range are less reliable.
• It is not a similarity match. The AI score does not mean your text was found anywhere. It is a prediction about how the text was produced.
• It is not proof. Turnitin states that the AI detection result should not be used as the sole basis for adverse action against a student, and its review guidance describes the score as one data point, not a definitive answer.

We have a full explainer on reading that number in Turnitin AI score: what the percentage actually means. Here we focus on how it sits next to the similarity score.

Similarity score vs AI score, side by side

• What it asks: similarity asks "does this text match something already written?"; AI asks "does this prose look machine-generated?"
• How it decides: similarity compares your text against a database of sources; AI uses a trained model to make a prediction, with no source to point to.
• What gets counted: similarity can include quotes, references and common phrases unless filtered; AI counts only qualifying long-form prose.
• What you can check: similarity shows you the matching source, so you can see exactly why a sentence was flagged; AI highlights sentences but cannot show you a source.
• What fixes a bad result: for similarity, correct quoting, citing and real paraphrasing; for AI, there is no reliable "fix", only evidence of your own writing process.

A worked example: reading both numbers together

Imagine a 3,000-word assignment comes back with similarity at 31% and AI at *%.

The wrong reading: "31% is high, I copied too much, and the asterisk means it caught some AI."

A better reading: open the similarity report and look at the source list. Suppose 12% comes from the reference list, 9% from three long block quotes from the same textbook, and the rest is short matches of terms like "semi-structured interviews" and "thematic analysis". None of that is copying. The asterisk means the AI share, if any, fell in the 1-19% range where Turnitin itself does not show a number.

Now the problem looks different. The real weakness is that three long quotes are doing work your own analysis should do. Cutting one quote and explaining the other two in your own words would make it a better assignment, and the similarity score would drop as a side effect.

Here is what that change looks like in one paragraph:

Before: a 70-word block quote defining thematic analysis, followed by "This study used this method."

After: one sentence naming the method and citing its source, then two sentences of your own on why it suited your interview data and how you coded the transcripts.

What to do if either number worries you

If the similarity score is high

1. Open the source breakdown and group matches into quotes, references, common terms, and genuinely unattributed text.
2. Fix only the last group: add citations, use quotation marks, or rewrite the idea fully in your own words with a citation.
3. Check whether your instructor excludes quotes and bibliography. If not, explain that in a short note if your score is inflated by them.

If the AI score worries you

1. Keep your drafts, notes, outline and version history. Process evidence is what actually supports you in a conversation with your instructor.
2. Follow your course's AI policy, and disclose and cite any AI use the way it asks. Our guide on AI disclosure and citation for students covers the basics.
3. Do not chase a number with tools that promise to "beat" a detector. No one can promise that honestly, and rewriting text to dodge a model rather than to say what you mean usually makes the writing worse.

FAQs

Is a Turnitin similarity score the same as a plagiarism score?

No. Turnitin describes it as a measure of matching text, not plagiarism. A high score can come entirely from properly cited quotes and references.

Can my AI score be high while my similarity score is low?

Yes. The two use different methods. Freshly written text, by a person or by AI, may match nothing in the database and still be given a high AI prediction.

Why does my AI report show *% instead of a number?

Turnitin shows an asterisk when the detected share falls between 1% and 19%. It hides the number and highlights in that range because results there are less reliable.

Does citing my sources lower the similarity score?

Not by itself. Cited quotes still match unless your instructor excludes quotes. Citing shows the match is legitimate; it does not hide it.

Can I see both reports myself?

It depends on how your institution has set up Turnitin. If you cannot see a report, ask your instructor what it showed and how they read it.

Want a human to check your draft?

If your similarity report shows weak paraphrasing, or your writing just reads stiffly, a careful edit helps more than any score-chasing. HumanizeBot gives your text an automatic first pass, then a human editor checks meaning, grammar and tone, usually within 15-20 minutes. We do not promise any detector score; we make the writing clearer and more yours. Students in India can pay by UPI, see HumanizeBot for India, and you can compare options on our pricing page. Start with the free 300-word trial, no card needed.