# What consent does to your targeting

URL: https://kywi-docs-production.up.railway.app/targeting/what-consent-does-to-your-targeting
Type: docPage

An undecided visitor loses every stored signal, but a campaign link, a live A/B split, and a hand-picked audience switch all still work.

A visitor who hasn't answered the consent banner is treated as **not** consented — there's no third, in-between state a page can resolve against.

## What that actually costs

Every stored signal in the personalization category reads as absent: UTM attribution, the audience the runtime previously resolved and remembered, known-identity flags, the session's entry page, self-ID answers. Nothing fails — the page still renders — it just resolves on every request as though this were a first-ever, signal-less visit, because none of that history is there to read back.

## What still works anyway

Signals that belong to the request itself, rather than tracking carried into it, are never gated: this request's own query string (including `utm_*` on the link that was just clicked), the referrer, and the path a visitor landed on. A campaign link still personalizes the page it points at on the very first click, before any consent decision has been made — it just doesn't survive to the next page unless consent follows.

## A switch you made by hand

The audience pin written when a visitor picks one from the [transparency panel's switcher](/targeting/the-transparency-panel-and-audience-switcher/), or submits a [self-ID](/targeting/letting-visitors-tell-you-who-they-are/) answer, is treated as essential rather than as personalization tracking — a control the visitor operated deliberately is honored regardless of the consent decision, even though the underlying bag of self-ID answers behind it is still gated like any other stored signal.

## A live split still buckets you

A running [A/B experiment](/targeting/running-an-ab-experiment/) isn't gated by consent at all: an undecided visitor is still assigned a real arm so the page renders something coherent. Only the exposure write is skipped — recording it would insert a fresh, uncounted row on every single page view for a visitor with no persistent id to collapse them into. That missing id costs the visitor the split's stickiness too: the arm is a hash of the experiment id and the visitor id, so a throwaway id re-draws it on every request rather than keeping them on one arm.

## Opting out doesn’t exempt a visitor from a running split either

The opt-out flag isn't checked by experiment assignment any more than consent is — an opted-out visitor lands in a running split exactly like an undecided one does, for the same reason: the page has to render *something*. Opting out stops stored personalization signals, not which arm of a live test someone sees.

## Turning the gate off entirely

A site whose consent is owned by external tooling can turn this whole gate off with one config flag, so every category reads as allowed regardless of any visitor's decision. It's a single switch for the whole site, not something set per audience or per signal — the site takes over responsibility for compliance the moment it's flipped.
