Overview
How might we design context mentions so people can trust and act on AI output?
When to use
- Perfect for AI coding assistants, enterprise knowledge management tools, and collaborative platforms where users need to reference specific files or documents in conversations.
When to skip
- Simple single-context tools where there is only ever one obvious source to reference.
- Interfaces with no indexed file or entity list to autocomplete against, where "@" would return nothing useful.
- Voice-first or no-keyboard surfaces where a typed trigger character is not a natural interaction.
Rules
Autocomplete lists with no search or filtering, forcing users to scroll through everything.
Mentioned items with no visible chip or confirmation that the reference was actually attached.
Referencing a file but sending stale or truncated content without indicating what was cut.
No way to remove a mention after inserting it without deleting surrounding text.
Evidence
| Product | Implementation |
|---|---|
| Cursor | @ mentions files, folders, docs, and past chats to ground code generation in real context. |
| Slack | @ mentions people and channels, with autocomplete over the workspace directory. |
| Notion AI | @ references pages and people inline, feeding page content into AI actions. |
| Linear | @ mentions issues and teammates directly inside comments and AI-assisted fields. |