Skip to content

How targeting works

How a visitor becomes an audience and a page becomes arms: signals, rules, scores, variant containers, experiments, consent and the opt-outs.

Targeting is two questions answered in sequence: which audience is this visitor, and which arm of the page does that audience get — both settled before anything reaches the browser, and both overridable, by the visitor themselves or by consent.

What the server knows before it renders anything

visitor signal

Any value a rule condition can test — UTM parameters, referrer, entry page, self-ID answers, and more. Every signal but Pages Visited is resolved on the server on the first request. See What each signal can see, and when.

visitor id

The persistent id, in a personalization-category cookie, that ties a visitor's requests together — what makes A/B stickiness possible. Without consent a fresh one is minted every request, so a split re-draws on every page view. See What consent does to your targeting.

From signals to one audience

audience

A named segment defined by condition groups and, optionally, a score threshold. Everything else here exists to decide which audience a visitor is in, or to serve a page against that decision. See How a visitor ends up in one audience.

condition group

One AND-ed set of conditions inside an audience's rule; groups are OR-ed against each other, so an audience matches the moment any one group matches in full. See What your rule will actually match.

instant match vs. score contribution

The two ways a condition can count: instant is pass/fail and gates its group's AND directly; score never gates anything, it only adds points toward the audience's threshold. See Score-based audiences.

score threshold

The points total an audience's scoring conditions must clear. With rule conditions also present a visitor must pass both; with none, the threshold alone decides. See Score-based audiences.

audience priority

An audience's position in the admin's list, and the order every audience is checked in. Every audience is still evaluated for every visitor — only the first match in that order is served. See How a visitor ends up in one audience.

When the visitor decides instead of the rules

self-identification

A visitor answering directly — what brought them here — instead of being inferred from behavior; their answer feeds the same audience resolution any other signal does. See Letting visitors tell you who they are.

self-ID field

One configured question on a self-identification widget; each field's answer becomes a signal a rule can read, the same way a UTM value is. See Letting visitors tell you who they are.

audience pin

An id stored on the visitor — by a self-ID submission or the switcher — that forces one specific audience ahead of ordinary resolution. A pin naming an audience inactive for this site is simply never honored, with nothing announced. See The transparency panel and audience switcher.

public audience

An audience deliberately marked so its name and label — not its rules — are exposed through a public listing route, narrower than what the client runtime ships to every visitor when it's on. See What is public about a personalized page.

From an audience to a page

variant container

A section, or a module inside one, wrapped to hold more than one arm and resolve to exactly one per visitor — by audience, or by A/B split. See Personalizing or A/B testing a section in place.

page variant

The page-level counterpart to a variant container: a whole page, not one section, carrying audience-targeted variants, resolved by the same precedence. See Which version of a page a visitor gets.

arm

One alternative inside a variant container or page variant. Exactly one is served to a given visitor; the rest are pruned before the page reaches the browser. See Which version of a page a visitor gets.

default arm

The arm served when nothing stronger claims the visitor — resolved like any other arm, not a placeholder, and not the same as an experiment's control arm unless you set them up that way. See The default arm is the real page.

A/B experiment

A named record with its own arms and split, wired to a container by an experiment id and matching arm keys. Switching a section or module to A/B mode in the admin mints an id for it, but that id names no experiment yet: point the container at one you actually run, or nothing splits and every visitor keeps the Default arm. See Running an A/B experiment.

traffic split

The whole-number percentages, summing to exactly 100, assigned to an experiment's arms — the authoritative division, distinct from a container's own advisory arm weights. See Running an A/B experiment.

deterministic bucketing

Assignment computed as a hash of the experiment id and the visitor id, with no row stored — the same two ids always produce the same arm. Only whether an exposure gets recorded is a separate decision. See Which version of a page a visitor gets.

exposure

One visitor bucketed into an arm and counted for it. Consent gates only the exposure write, not the bucketing itself. See Reading an experiment readout.

conversion goal

The event an experiment measures against exposures to compute each arm's conversion rate. See Reading an experiment readout.

statistical significance

A two-proportion test between an arm and control at 95% confidence, which doesn't run until both arms clear a minimum exposure floor — below that floor is a different claim from "not significant." See Reading an experiment readout.

locked winner

An arm forced onto a container ahead of its split, set on the container itself — distinct from declaring a winner on the experiment record, which retires the experiment and reverts to Default instead. See Which version of a page a visitor gets.

What suspends or overrides all of it

preview token

A signed, revocable link scoped to one audience that forces it for whoever opens it, briefly — distinct from the rule tester, which only proves logic against values you type in. See Checking a rule before you ship it.

A visitor's stored tracking decision, and the category — such as personalization — each signal is classified under. An undecided visitor is treated as not consented; there's no third state. See What consent does to your targeting.

personalization opt-out

A visitor's signal to suppress personalization. It forces audience resolution to none, so audience-driven arms fall back to Default. That has one consequence worth stating plainly: an arm bound to a matched audience is exactly what keeps a visitor out of that container's split, so clearing the audience hands them to the split instead. Opting out never takes a visitor out of a running A/B experiment, and it can put them into one they were being kept out of. See The default arm is the real page.

transparency panel

The persistent, visitor-facing control showing whether a page was personalized, naming the audience, with a switcher, reset and the opt-out above. Collapsing it isn't dismissing it — the pill always reopens it. See The transparency panel and audience switcher.