Definition
An exposure event is an analytics event sent when a visitor actually sees an arm of a test, carrying two values: which test it was and which variant the visitor saw. The Google Analytics tutorial for reading A/B test results sends it from every arm, the control included, and registers both values so they can be used in reports and explorations. Each exposure then records a single fact: this person saw this version at this moment.
With that record in place, analytics can supply what the testing tool's own report leaves out: an independent count of each arm, what each arm did before reaching the goal, and the test set beside every other event the business already collects.
When it fires and what it carries, and why assignment is the wrong moment
The moment it fires matters. A visitor can be assigned to an arm before reaching the part of the site the test changes, or without ever reaching it. An event sent at assignment counts those people as exposed although they saw nothing, and the analysis tutorial warns that this dilutes the difference between the arms. The event belongs at the point where the tested page or step is actually shown.
It is sent in both arms. If only the variant sends it, the control has no count of its own, and there is nothing to divide by when the two are compared. The test and the version travel as short, controlled identifiers agreed in the tracking plan, never as free text, and the same pair of parameters can be reused for every later test.
Analytics will not match the testing tool exactly. Consent choices and blockers stop some tags, so analytics sees fewer people than the tool assigned. A shortfall of similar size in both arms leaves the comparison intact; one that differs between the arms means an arm is losing events.
In practice: server-side tests, guardrails, shipping and results reports
Where a test runs changes where the event is sent from, not whether it is needed. The guide to choosing client-side or server-side testing notes that a server-side test assigns before the page is built, so analytics has to be told when the visitor reaches the tested page; when visitors are redirected between two separately built pages, the event comes from the page they land on. A test moved to the server loses its flicker but none of this measurement work.
The A/B testing guide uses the event so that the main metric and every guardrail can be compared arm by arm in analytics, not only in the testing tool's report, because a win on the primary metric alongside a breached guardrail calls for a business decision rather than an automatic rollout.
When the test ends, so does its event. The tutorial on shipping a winning variant removes the test's code, its exposure event included, so nothing from a finished test lingers on the site, while the analysis tutorial keeps the two registered parameters for the next test. Before a figure goes into an experiment results report, the report's validity checks confirm that the event and its variant value reached analytics from both arms.
An online florist moves its test record to the right moment
Glarvoor Flowers, an invented online florist, tested a shorter delivery-date step in its checkout. Its first exposure event fired when the testing tool assigned each visitor on arrival, so analytics counted 18,400 visitors per arm as exposed, although only 4,900 per arm ever reached the delivery-date step. Checkout completion across everyone counted read 6.1% in the control and 6.4% in the variant, a gap the team could not read with any confidence. The analyst moved the event to the moment the step appeared, in both arms, and restarted the test. On the corrected count, completion among visitors who saw the step read 22.8% against 24.1%, and the daily split check stayed within the agreed tolerance throughout.
Related terms
An A/B TestControlled comparison of two variants. compares two versions, and the exposure event is how analytics learns which one each person saw. Counting exposures per arm every day is how a Sample Ratio MismatchA persistent gap between the share of visitors an A/B test was meant to assign to each variant and the share each actually received. comes to light, and a gap that persists is found and fixed before anything is read. Each identifier travels as an Event ParameterA named value sent with an analytics event that describes it, such as a page location, form name or item price., registered so that reports can use it. Every Guardrail MetricA metric an experiment must not worsen beyond an agreed bound, watched beside the primary metric that decides the test. is read per arm, among the people the event marks as exposed. FlickerThe brief flash of the original version before a browser-side test script applies the variant, which turns a result about the change into one about the fault. is a fault of browser-side tests that moving the test to the server removes; the exposure event is needed in either case.
Common confusions
- It is an ad impression. Exposure here means seeing a test arm on the business's own site, not seeing an advert. Media exposure and impressions are counted elsewhere.
- Firing it on assignment is close enough. People assigned but never shown the change wash out the difference between the arms.
- The variant alone is enough. Without exposures from the control there is no count to compare the variant against.
- Analytics and the testing tool should agree exactly. They count different things. What matters is whether the shortfall is similar in both arms.
- Server-side tests can skip the tracking. Running a test on the server ends flicker, but the exposure event and the daily split check still apply.
