If the assistant produces something wrong, unsafe or otherwise off, you can report it from the message itself.
How to report
Click or tap the flag under the AI message. A Report response menu opens: pick the reason that fits, harmful or unsafe, offensive or hateful, sexual or explicit, inaccurate or misleading, or something else.
What happens next
The report goes to us with the response attached, and a person reviews every report we get. Reports directly shape model choices and the guardrails around the agent.
For wrong answers about your own notes
If the assistant misread your notes rather than misbehaved, reporting still helps, but also try tightening the ask: name the note or folder, add the date range, and check that indexing has finished so search is seeing your whole library.
The general caution
The phone says it plainly under the chat: AI can make mistakes, check important information. It holds on desktop too. Answers point back at your notes so you can verify the source in one click.