Glossary

Plain-language definitions of the terms used throughout Accelerate — the statistics behind testing, the analytics metrics, the privacy concepts, and the AI surface.

A/B testing & experiments#

A/B test — Showing two or more versions of the same content to different visitors at the same time to learn which performs better against a chosen goal. In Accelerate you create tests in the block editor. See A/B Testing.

Variant — One version of the content in a test. The original is the control; each alternative is a challenger.

Conversion goal — The measurable action a test is judged on — an engagement, a click on any link, or a form submission. Themes can register custom goals. See Appendix → Goal types.

Conversion rate — For a block or variant, the share of unique visitors who saw it and took the goal action. It's counted per visitor, not per event: unique conversions ÷ unique blockviews.

Bayesian statistics — The statistical approach Accelerate uses to evaluate tests. Instead of a frequentist p-value, it reports the direct, business-ready answer: the probability that each variant is the best. No fixed sample size to set in advance.

Probability to win — The Bayesian probability that a given variant is the best of those in the test (also called probability-to-be-best). A variant reported at 96% has a 96% chance of being the winner. This is what the results screen and get-experiment-results report.

Credible interval — The Bayesian counterpart to a confidence interval: the range within which a variant's true conversion rate most plausibly falls.

Multi-armed bandit — A traffic-allocation method that shifts more visitors toward the better-performing variant while the test is still running, instead of holding a fixed split. New Accelerate tests run as bandits, so fewer visitors see the losing option. See A/B Testing → Traffic that follows the winner.

Thompson sampling — The algorithm behind the bandit: it allocates traffic in proportion to each variant's current probability of being best, balancing exploration with exploitation.

Burn-in phase — The opening phase of a bandit test, where traffic stays evenly split until enough conversions accumulate, so early noise can't hijack the allocation.

Confirmatory phase — The closing phase: once a variant crosses the probability threshold, the split returns to even for a clean, unbiased final reading before a winner is declared.

Personalization rule — A mapping from an audience to a content variation, so different visitors see different content. See Personalization.

Audience (segment) — A group of visitors defined by attributes such as referrer, geography, device, or behaviour, used for targeting and personalization. See Audiences.

Analytics#

Pageview / visitor / session — A pageview is one view of a page; a visitor is a unique browser identified by a pseudonymous UUID; a session is a single continuous visit.

Block-level analytics — Measuring individual blocks (not just pages), so you can see how a specific call-to-action or pattern performs wherever it appears. See Synced Patterns and Global Blocks.

Blockview — Recorded when a block actually scrolls into the visitor's viewport — real exposure, not just a page load.

Engaged time — Time that accrues only while a visitor is actively reading, scrolling, or interacting — attention, not a tab left open. Available as a dashboard breakdown and in the engagement metrics. See Analytics.

Bounce rate — The share of visits that left without a meaningful second interaction.

Referrer — The page or source a visitor arrived from.

UTM parameters — Standard utm_* tags on a link (source, medium, campaign) that Accelerate groups into campaign reporting with no extra setup. See Analytics → Campaign tracking.

Attribution — Assigning credit for a conversion to a traffic source. First-touch credits the first known source; last-touch credits the most recent before conversion. See Appendix → Attribution terms.

Outbound click tracking — An opt-in feature that records the external destination host visitors click to — never full URLs or personal data.

Privacy#

Cookieless analytics — Measurement that uses no tracking cookies. Accelerate recognizes returning visitors with a pseudonymous identifier in localStorage instead. See Privacy Overview.

Pseudonymous identifier — A random UUID that distinguishes a browser without revealing who the person is, and is never linked to a name or email by default.

WP Consent API — The WordPress standard for communicating consent status between plugins. Accelerate honours it so one consent signal governs collection. See WP Consent API.

Data controller / processor — Under GDPR, the controller (you, the site owner) decides why and how data is collected; the processor (Human Made, via Accelerate) acts on the controller's instructions. See Data Privacy.

AI & automation#

Abilities API — Accelerate's structured set of named capabilities (such as accelerate/get-performance-summary) that lets AI agents query data and take controlled actions. See Accelerate AI.

MCP (Model Context Protocol) — The open standard AI clients use to discover and call tools. Accelerate exposes its abilities over MCP. See Quick Start.

Application password — A WordPress credential, separate from a login password and independently revocable, used to authenticate AI clients and API integrations.

Autopilot — A standing approval to let an optimization loop run a block unattended — creating variants, testing, and applying winners without per-step confirmation, while the statistical discipline still holds. See The Autonomous Loop and Permissions.

Design & Evolve — The Toolkit workflows that generate on-brand content variants (Design) and improve a block round after round (Evolve). See Toolkit.