Personalization that cannot be turned off undermines autonomy

What it is

Home, search, and notifications are all organised as “for you,” settings have no exit, and signing out only swaps in another identity that will be estimated the same way — personalization has stopped being a service and become the environment. Inescapable personalization means the user cannot switch the system into a mode that does not estimate “me,” and so cannot choose not to be defined by the model.

What is undermined is autonomy: not that the list is inaccurate, but that the person cannot decide “according to me, or not.” Inaccuracy can be fixed in the model. No off means there is not even standing to ask for a fix.

Why it happens

Autonomy depends on an alternative path. Once every path is rewritten as a personal estimate, the alternative disappears and the person can only act inside the world the model gives. The only refusal left is leaving the product, and the cost of leaving (social graph, purchases, work dependence) is often higher than staying. “Keeps using” cannot be read as consent.

No-off also reshapes the self: people start performing a preference to get usable results, or give up exploring anything outside the model. Behaviour stops being an expression of taste and becomes adaptation to an environment they cannot exit. Inspect-and-reset without an off still leaves a hole — the person can edit the profile, but cannot choose not to have one.

Studying it

Compare three product states: off available and on by default, off available and off by default, off unavailable. Tasks: demonstrate the product to someone else, open it on a public screen, explicitly try to spend a session “not being estimated.” Dependent variables: felt ability to choose, shift to non-personalised entry points such as search, use of throwaway accounts to evade. Independent variables: discoverability of the off control, lock-in of leaving (social or work binding).

Do not substitute overall satisfaction for autonomy. People can like an accurate list and still refuse “I cannot turn this off.” Ask control and “can I go without this estimate” as their own items.

Where it stops holding

Safety filters and legally required orders (a financial disclosure sequence) are not personalization; being unable to turn them off is not a forced profile. If a full product exists while logged out, logged-out already is the exit. In medicine and education, role-required adaptation (dose, year) is configuration, not taste estimation. This entry argues that being unable to exit “estimate me” undermines autonomy. It does not require the post-off default to be chronological, and it does not specify the fields of a profile pane.

Applying it

• Provide a real exit: after off, the main path must not eat a personal vector. “Show less like this” is not a substitute for exit.
• Logged-out or guest mode must be a complete, usable product, not a crippled splash, or exit is priced as loss of function.
• Check: without deleting the account, have someone put the product into “do not estimate me.” If they cannot in five minutes, or the main path still uses personal signal after they do, autonomy is still locked.

Related

• Same group: L6.05.1 Users must be able to inspect and reset their preference profile · L6.05.2 Turning personalization off must land on an understandable default ranking
• Nearby: L6.10 Turning Personalization Off and Resetting It · L6.12 Conflict Between Personalization and Predictability · L5.03 Trust Calibration
• Search terms: inescapable personalization · autonomy and opt-out · forced personalization