Turnitin's AI Score Explained: What Each Percentage Actually Means

How Turnitin's AI Detection Actually Works

Turnitin's AI detection model analyses your writing at the sentence level. It doesn't just look at your whole essay and give a thumbs up or down -- it breaks your submission into individual sentences and scores each one based on how likely it was generated by an AI language model. The final percentage you see is the proportion of sentences Turnitin flagged as AI-generated.

This is important to understand because a 30% score doesn't mean "30% of your essay is copied from ChatGPT." It means that 30% of your sentences triggered the model's threshold for AI-like writing patterns. These patterns include things like predictable word choices, uniform sentence length, lack of personal voice, and overly structured paragraphs.

graph LR
    A["0-20%"] -->|"Low risk"| B["Likely human-written"]
    C["20-50%"] -->|"Medium risk"| D["Review flagged sentences"]
    E["50-100%"] -->|"High risk"| F["Expect investigation"]
    style A fill:#18181B,stroke:#FFAA00,color:#F5F5F7
    style C fill:#18181B,stroke:#FFAA00,color:#F5F5F7
    style E fill:#18181B,stroke:#FF4444,color:#F5F5F7
    style B fill:#18181B,stroke:#636366,color:#8E8E93
    style D fill:#18181B,stroke:#636366,color:#8E8E93
    style F fill:#18181B,stroke:#636366,color:#8E8E93

Tip: A score under 20% is generally safe. Most universities only investigate scores above 40-50%. But check your institution's specific policy.

What 0-20% Means

A score between 0% and 20% is generally considered low risk. Most universities won't flag your work at this level. However, it's worth noting that even fully human-written work can sometimes land in this range -- especially if you write in a formal, structured style or if English isn't your first language. Turnitin themselves say scores below 20% shouldn't be used as sole evidence of AI use. If your lecturer raises it, you can usually explain your writing process and resolve it.

What 20-50% Means

This is the grey zone. A score between 20% and 50% will almost certainly get flagged by your institution's academic integrity team. At this level, Turnitin is saying a meaningful chunk of your sentences look AI-generated. But here's the thing -- this doesn't necessarily mean you cheated. Students who heavily edit AI outputs, use Grammarly extensively, or write in a very formulaic way can land here. The problem is that the burden of proof often shifts to you. You'll likely need to show drafts, explain your process, or demonstrate knowledge of the topic in a viva.

What 50%+ Means

Above 50%, you're in serious trouble territory at most institutions. At this level, it's very hard to argue the work is entirely original. Turnitin is flagging the majority of your sentences as AI-generated, and most academic boards will treat this as a strong indicator of misconduct. Some universities have automatic referral processes at this threshold. If you're here and you did use AI, the best approach is honesty -- many institutions have lighter penalties for first-time admissions compared to denials that are later proven.

What You Should Actually Do

First, understand your university's specific policy. Some allow AI as a brainstorming tool, others ban it entirely. Second, if you do use AI tools in your workflow, make sure you're rewriting substantially in your own voice -- not just swapping synonyms. Third, keep your drafts. Google Docs version history, handwritten notes, and research bookmarks can all serve as evidence of your process. Finally, run your own checks before submitting. Turnitin's draft coaching feature (where available) or similar tools can give you an indication before your lecturer sees the score.

The reality is that AI detection is imperfect, and both false positives and false negatives happen. The best protection is genuinely understanding your topic and writing in your own voice -- no detector can flag authentic expertise.