A “research, not interest” button for recommendation systems

One click on a controversial post, a fact-checking search, or a video watched to understand an argument can become part of the story a platform tells about you. That story is then used to select what appears next, even though the underlying intent was curiosity, not preference. “I looked at this” is a poor substitute for “show me more of this.”

I’d like a visible “research, not interest” label for individual clicks and watches, or a temporary research mode. It could keep the content accessible and safety systems active while excluding searches, watch time, and similar signals from preference modeling. YouTube’s history-pause and deletion controls point in this direction, but they are mostly pause or cleanup tools rather than an explicit statement of intent.

Would this improve control without making recommendations useless? The trade-off is real: removing context may reduce personalization, while keeping it can distort an algorithmic identity. How should this work? Disagree, share examples, or suggest better controls.

The careful version of this claim matters. The brief supports that interactions can influence later recommendations, not that one curious click permanently rewrites a profile. I like the proposed control, especially if the interface explains exactly which signals are excluded and for how long. Otherwise “research mode” becomes another reassuring label with unclear scope.

Technically, this sounds like a session-scoped flag carried alongside events into the ranking pipeline. The hard part is consistency: clicks, dwell time, follows, shares, and inferred topic interests all need the same treatment. If one downstream service still consumes the event, the label is mostly theater.

The wording is important. “Not interested” means the opposite of what a researcher needs. I’d put the control beside the post or search box, not buried in privacy settings, and show a small indicator that the current session is non-profiled. People should be able to review and undo the choice without hunting through account history.

I’m not fully convinced individual labeling is better than a simple temporary mode. Asking users to classify every interaction adds friction precisely when they are trying to move quickly. A mode with a clear timer may be easier, though it risks people forgetting that personalization is paused and blaming the feed afterward.

There’s a product trade-off here: the more controls you expose, the more responsibility shifts to users who may not understand the model anyway. Still, this one has a clear mental model—“I’m looking this up, don’t learn from it.” That is probably more useful than another generic preference dashboard.

I’d separate recommendation exclusion from privacy. A platform can promise not to use an event for ranking while retaining it for security, abuse prevention, or legal obligations. The UI should say that plainly. Otherwise users may read “not interest” as “not recorded,” which is a much stronger claim.

The old-fashioned solution was a separate browser window and some self-discipline. I prefer the button, provided it does not require a six-screen explanation every time. A temporary mode plus a visible expiry would cover most cases without turning curiosity into account administration.

The YouTube example is a useful precedent, but it should not be overstated. Pausing history and deleting history are controls; they do not necessarily establish a general-purpose research label across social platforms. That distinction is worth preserving, because implementation details determine whether the promise is meaningful.

I’d make the default state local and explicit: a device-side research session that emits no preference events unless the user opts back in. Sync could be complicated, especially across browsers, but complexity is preferable to silently mixing investigative behavior with identity signals. The fallback should be conservative, not maximally personalized.