Should platforms offer a temporary research mode?

I’d like to see a temporary “research mode” for recommendation systems: before clicking, searching, or following, users could mark the activity as curiosity, work or school research, or a one-off rabbit hole. The system could still use it to serve the current session, but not fold it into long-term interests, inferred identity, or advertising categories.

YouTube already offers a partial precedent by letting people pause watch or search history, including for researching a subject, but that is a blunt switch and easy to forget. Intent labels could be more precise, with a visible mode indicator, automatic expiry, and an undo/review screen. The trade-offs are real: accidental misuse, weaker personalization, and the need to retain some short-lived operational logs. The harder question is whether platforms would genuinely separate temporary use from permanent profiling. Would intent labels give us meaningful control, or just another interface promise platforms can ignore?

The expiry and boundary matter more than the label. If “research mode” still emits the same event stream into every downstream warehouse, the UI is theater. I’d want a documented data path: session-serving signals allowed, profile and ad pipelines excluded, with deletion after a stated window.

I like the idea, but three labels may be too much friction for a behavior people want to be instant. A single temporary toggle with an optional reason could work better. The product risk is that users forget it is on and then complain that recommendations went strangely generic.

“Not retained” needs careful wording. Security, abuse prevention, billing, and service delivery can still require short-lived records. That is compatible with the feature, but only if the retention purpose and duration are explicit rather than hidden behind a broad privacy promise.

The label could also change the social meaning of browsing: curiosity becomes something the interface asks us to declare. Still, an obvious temporary mode is better than making people clean up an inferred identity after every unusual search. I’d test the toggle before adding a taxonomy.