Similarity percent and AI writing percent are two different scores

Charles Li · HumanPen · similarity-vs-ai-two-scores

Two percentages often sit on the same LMS card. One is similarity. One is AI writing. Screenshots crop them into a single panic number. They are two different scores—do not treat one as proof of the other.

Distinct from similarity-vs-ai-writing-two-instruments (same literacy, longer instruments framing) and authorship-vs-ai. Here the short point is two scores, two evidence trails.

Better order

1. Name which dial you are talking about before forwarding a screenshot.
2. Open the similarity match groups when arguing overlap; open the AI writing highlights when arguing classifier flags.
3. Refuse a causal bridge from card adjacency alone.
4. Keep authorship off the table until that surface is open.
5. Scope any rewrite to the report you are actually answering.

Why this matters

Low similarity with high AI writing is possible. High similarity does not mechanically cause a high AI writing estimate. Collapsing the dials into folklore wastes the next rewrite pass.

What this is not

• Not a guarantee the next AI percentage will drop or clear.
• Not permission to wash the whole document or to chase “undetectable” output.
• Not a substitute for your course integrity process if one is opened.

Product: humanpen.net. Charles Li · HumanPen.