Don't Collapse Similarity and AI Writing Into One Story

Two percentages on one card invite one panic narrative. Similarity and AI writing look like twin dials. Mechanically, they are not.

## Core point

**Similarity** measures overlap with indexed sources. **AI writing** estimates machine-like continuous prose under current classifier and display rules. Collapsing them into "the score" invents a bridge the systems do not document as automatic.

## Quick separations

- Low similarity can sit next to a high AI estimate (smooth prose, little indexed overlap).

- High similarity can sit next to a muted AI band (heavy source matches, different qualifying-prose slice).

- Authorship features, when present, are yet another surface—not a third percentage of guilt.

## Two-instrument reading script

"I am separating the similarity report from the AI writing report. Which dial are we opening first? Here are the match groups / here are the AI-highlighted passages."

Dull scripts beat merged folklore.

## Soft CTA only after AI scope is clear

When AI-flagged continuous prose must be rewritten on a matching Turnitin or iThenticate AI report, import that report, rewrite only marked passages, preserve citations and formatting. [HumanPen](https://humanpen.net) is built for that document-level job—without promising either dial's next percentage.