Rebuilding Trust in Comet Browser's AI Permission Experience
Research on why users hesitate to switch on Comet Browser's AI features, and why permissions are the blocker
✳The problem they went after
Comet, the browser from Perplexity AI, automates work like job applications, email sorting and form filling, which means asking for extensive permissions during setup. Research found that ask lands badly. The scope of access makes people uncomfortable, and the specific fear they voice is that one security breach would expose their entire digital life at once.
✳What they actually did
A second finding covers reliability. Comet handles simple repetitive tasks like LinkedIn Easy Apply very well, then fails silently on long multi-step workflows such as full applications on company career portals, stopping partway without flagging the problem or asking for help. Between the two, trust was prioritised, because the permission wall sits at setup and causes drop-offs before anyone reaches the features. The goal is a user trust score of 8 out of 10 within six months, via clearer permission explanations, granular controls and transparent data access communication.
✳What you can take from it
Worth stealing
- Two friction points ranked by position in the funnel
- Silent failure treated as distinct from plain failure
- Trust score of 8 out of 10 set as goal
- Granular controls replace one broad setup-time ask
Read this if
Designing permission or consent flows for an AI product
The transferable bitOne problem sits earlier in the funnel and gates the other, so improving reliability for users who never turned the feature on buys nothing. Both are real; only one is worth doing first. Silent failure is also called out as its own category of harm, distinct from a feature simply not working.
✳The submission itself
Open the PDF →Reading on a phone? The embedded viewer is cramped — open the PDF in a new tab instead.
