Turnitin AI Score vs Similarity Score: What's the Difference?

Similarity: 'has this text appeared before?'

A matching problem. Turnitin compares strings of your text against its index and reports how much overlaps. Fully original writing scores near zero. Correctly quoted and cited material still counts toward the number unless filters are applied.

AI score: 'does this text read as machine-written?'

A prediction problem. No source is matched. Turnitin estimates from sentence structure and word predictability whether each passage was generated. Text can be 100% original and still score 90% AI.

The four combinations

Where you land tells you what to fix.

• Low similarity, low AI — clean, submit
• High similarity, low AI — citation and quoting problem; fix references and paraphrase properly
• Low similarity, high AI — generated or heavily tool-rewritten prose; needs a genuine rewrite
• High similarity, high AI — usually AI output that itself reproduced known sources

Why paraphrasing tools make things worse

A spinner keeps the machine-like sentence rhythm the AI detector looks for while introducing awkward phrasing that can still match its source. You often end up with both numbers unchanged and worse writing.

Fixing a high AI score properly

Rewrite from your own notes rather than editing the generated text. Vary sentence length deliberately. Add concrete specifics — your data, your reading, your examples. Then re-check to confirm.