Overview
How might we design inline comment thread so people can trust and act on AI output?
When to use
- Perfect for review flows on AI-generated PRDs, emails, code, and policies where stakeholders need precise, asynchronous discussion on fragments.
When to skip
- Ephemeral answers users will never revise.
- Surfaces with no selection model.
- Mobile views where comment chrome destroys readability (offer a compact mode).
Rules
Comments that detach when the AI regenerates the paragraph.
No resolve state, leaving eternal open threads.
AI replies in comments with no human attribution.
Commenting on content the commenter cannot see due to ACL bugs.
Evidence
| Product | Implementation |
|---|---|
| Notion | Inline comments on blocks and AI drafts. |
| GitHub | Line comments on code and Copilot-suggested changes. |
| Confluence | Inline page comments for collaborative review. |
| Google Docs | Selection comments and suggestion threads. |