What AI Score Is Acceptable at UK Universities? The Honest Answer

Students search for a safe Turnitin AI percentage every day — is 20% acceptable, is 35% bad, is there an official limit? The honest, sourced answer: no UK university we have reviewed publishes an "acceptable" AI percentage, and the tools' own vendors say the score should never decide anything on its own.

That doesn't make the question pointless. There are real thresholds in how the tools behave and how universities respond — they just aren't the pass/fail line people expect.

The short answer

There is no acceptable percentage, because the number does not measure rule-breaking. Turnitin's AI percentage is the share of qualifying prose its model classifies as AI-like — quotations, formulaic sections and permitted grammar tools can all raise it, and the same essay scores differently on different tools. What matters at a UK university is whether your module allowed AI assistance, whether you disclosed what you used, and what your drafts show. A "low" score with no drafts is weaker protection than a "high" score with a full version history.

Why no university publishes a safe percentage

Three sourced reasons:

1. The vendors tell them not to treat it as proof. Turnitin's own guidance says AI scores should not be the sole basis for adverse action, and its report guide describes the result as an indicator for review.
2. UK complaints guidance requires evidence. The Office of the Independent Adjudicator's casework note on academic misconduct expects institutions to reach decisions on evidence a student can respond to — a bare percentage does not meet that bar. Our guide to detector scores as evidence covers this in depth.
3. Publishing a number would create a target. Any published "acceptable" percentage becomes an instruction to edit until under the line — which is why universities that address the question at all talk about process and permitted use, not thresholds.

The thresholds that do exist

These are behavioural thresholds in the tools, not acceptability lines:

• 300 words. Turnitin will not produce an AI report for documents under 300 words of qualifying prose (file requirements). Text length changes reliability.
• 1–19% shows as an asterisk. Turnitin suppresses low AI scores because false positives are more common in that band.
• 20%+ is where the number is displayed — and where some universities' staff workflows say a report becomes worth reading closely.
• Similarity is different. "Is 36% similarity bad?" is a different question about a different score; see plagiarism vs similarity.

What actually triggers a review

From UK universities' own published procedures (linked in our policy tracker), a review typically starts when a marker forms a genuine concern — and a detector score, where the university uses one at all, is only one possible prompt. Markers also act on sudden style shifts, hallucinated references, content that doesn't match the module, and answers that don't engage with the brief. Several UK universities have switched Turnitin's AI indicator off entirely and rely on those human signals; others confirm the report is in use as a starting point for a conversation.

The consistent shape across procedures: score or no score, a case needs evidence you can respond to — which is why the section below matters more than any percentage.

What different bands mean in practice

Assuming your university runs an AI indicator at all:

• Asterisk / under 20%: below the display floor; vendors treat this band as unreliable. Nothing to act on.
• 20–40%: worth understanding before submission. Read which passages carry the signal — quotations, definitions and formulaic sections land here innocently. See the worked 30% example.
• 40–60%: the mixed band where tools disagree most; if this is your own unaided writing, it is worth keeping your drafts organised in case questions come later.
• 60%+: the band vendor labels call "AI". If you did use AI beyond what the module permits, this is the moment to revisit the work, not to reach for a "humaniser" — rewriting tools change the evidence, not the underlying problem. If you didn't, your process evidence is your answer: here's how to prove it.

What protects you better than a low score

1. Know your module's AI rules — permitted use varies by course, and disclosure requirements are increasingly common.
2. Keep drafts, notes and version history. Document your process as you write; it is the single most persuasive evidence in a dispute.
3. Write in your own voice and check your own signals. A free Genutext scan shows which passages read as AI-like so you understand your own writing — it does not predict Turnitin's number, and nothing does.
4. If accused, follow the process calmly. Our step-by-step UK guide for accused students covers what to say and what to bring.

Frequently asked questions

What percentage of AI is acceptable on Turnitin?

None is officially "acceptable" — no UK university we reviewed publishes a safe percentage, and Turnitin says the score should not decide outcomes alone. Module rules about permitted AI use are what define "acceptable".

Is 35% on Turnitin bad?

Not by itself. It means roughly a third of the qualifying prose was flagged as AI-like — which can include quotations and formulaic writing. What matters is which passages were flagged and whether your module allowed AI assistance.

Is 20% AI on Turnitin safe?

20% is simply the first value Turnitin displays as a number. It is neither safe nor unsafe: universities do not publish thresholds, and reviews depend on evidence beyond the score.

What AI percentage is acceptable in university essays generally?

The question has no numeric answer at any UK university we reviewed. Courses define what AI assistance is permitted; the percentage measures pattern similarity, not compliance with those rules.

Do universities automatically investigate above a certain AI score?

The published procedures we reviewed describe marker judgement and evidence, not automatic score triggers. Some universities do not run an AI indicator at all.

Sources and further reading

• Turnitin: Using the AI Writing Report
• Turnitin: File requirements for an AI writing report
• OIA: Casework note — complaints relating to academic misconduct
• University of Melbourne: Advice for students regarding Turnitin and AI writing detection
• Genutext: UK University AI-Detection Policy Tracker
• Genutext: What an AI-detection percentage means

Turnitin is a trademark of its owner. Genutext is independent and not affiliated with any university or detection vendor named here. Statements reflect public documentation as of 23 August 2026.