Turning personalization off must land on an understandable default ranking

What it is

After personalization is off, the screen still has an order. If that order cannot be stated — not time, not popularity, not the filter the user just set — then off is only a swap into another opaque weighting. An intelligible default after opt-out is a ranking rule that can be said in one ordinary sentence, and that sentence matches what is on screen.

Off is not “guess some other way.” Off is “switch to a rule the user can check.”

Why it happens

“Off” is a state flip in the user’s head: from “according to me” to “not according to me.” If the list still looks personally aimed after the flip, people cannot confirm the switch worked; they either toggle it again or judge it fake. An intelligible default gives the flip an observable landing: by time, by global popularity, by an editorial slate, by the filter just ticked — each can be checked by eye.

A default is still an arrangement, not a law of nature. What is required here is not “objective neutrality” but a rule that can be stated and checked. The editorial character of that arrangement is a later layer. This layer only requires that off land on a speakable rule, not on another implicit model.

Studying it

Have people turn personalization on, then off. Ask “what is it ranked by now?” and have them use the rule to forecast the next row. Independent variables: type of post-off rule (time / global popularity / editorial / unexplained “discover”), whether the rule is written beside the switch. Dependent variables: correctness of the restated rule, forecast hits, belief that the switch has taken effect.

If off still uses the same model with the user vector replaced by a population mean, the surface barely changes and restatement fails. That is an unintelligible default, even if engineering did “turn off the personal vector.”

Where it stops holding

Some products have no speakable non-personalised order (matching, dynamic pricing). Then off should shut the whole recommendation channel rather than hand over a fake default. Live auctions and stock constraints can make even “by time” impossible; declare the actual rule in advance. This entry only requires that off land on a speakable order. It does not argue that an off switch must exist, and it does not treat undeclared inference from context after off.

Applying it

• Write the post-off rule beside the switch: “when off, ranked by publish time, newest first.” Do not write only “turn off personalization.”
• Change the rail title with the switch: from “For you” to “Latest” or “Site-wide popular,” so the title matches the rule.
• Check: after off, ask “what will the next row be, and why.” If people cannot name the rule, or the rule disagrees with the screen, the default is not intelligible. Then sample ten rows against the written rule.

Related

• Same group: L6.05.1 Users must be able to inspect and reset their preference profile · L6.05.3 Personalization that cannot be turned off undermines autonomy
• Nearby: L6.10 Turning Personalization Off and Resetting It · L6.12 Conflict Between Personalization and Predictability · L1.02 Expressing Capability Boundaries
• Search terms: intelligible default after opt-out · non-personalized fallback · chronological feed