Questions, answered

Yes. The free plugin is GPL-licensed and free forever. It runs complete A/B and A/B/n tests on its own, up to six variants and as many tests at once as you like; Pro adds statistics, automation and AI tools.

No. The point-and-click editor means you can click “A/B Tests” in the admin bar, click the element on your page, type your variants, and start - no builder field and no code. If you prefer working in your builder, you can instead tag variants with the CSS class the plugin generates. Either way, a step-by-step guide is built in.

Yes. Variants are matched by a CSS class, so it works with Gutenberg, Elementor, Bricks, Etch, Divi, Beaver Builder, Oxygen, Breakdance, Kadence, GeneratePress, Astra, and any classic theme.

WordPress 6.0 or newer and PHP 7.4 or newer, tested up to WordPress 7.1. It works on multisite, one installation per site. The WooCommerce goals have no minimum version and work with high-performance order storage. Language packs come from WordPress.org. Install and run your first test.

Headlines, buttons, images, layouts, and whole pages via URL/redirect tests - URL tests even work across two sites or subdomains. A whole section opens as one field per text, with images swapped from the Media Library. Goals run from a page visit to scroll depth, time on page, active time and WooCommerce purchases with revenue. In Pro, each variant can also carry its own auto-scoped custom CSS, and AI Copy and Style (using your own OpenAI, Anthropic, or Gemini key) help with wording and looks.

No. The visitor script is about 14 KB gzipped and loads from the page head, deferred on WordPress 6.3 and later, so it does not hold up the page. It comes from your own server with no external requests, and the tested element stays hidden until the assigned variant is applied, so the original does not flash first; if the script never runs, a two-second failsafe shows the original. Content a page adds after it has loaded, such as a popup, is the exception: its original can show for a moment. With JavaScript off, visitors simply see the original.

Yes. Variant assignment happens in the visitor’s browser, so the same cached HTML serves everyone and no cache exclusions are needed. When a test starts or stops, the plugin purges the popular page caches for you (WP Rocket, LiteSpeed Cache, W3 Total Cache, WP Super Cache, Cache Enabler, the Cloudflare plugin and others) so the small head snippet refreshes. How caching compatibility works.

Not with the lists they ship with. uBlock Origin, AdGuard, Brave and similar blockers work from public lists of ad and tracker addresses, and cloud testing tools that load from their own servers are on those lists. ImproveAB’s script and its tracking address are on your own site, and before each release they are checked against EasyList, EasyPrivacy and the uBlock Origin and AdGuard lists. A visitor whose browser runs no scripts sees the original version and is not counted.

Rarely. Assignment is stored in the browser’s local storage, so it is per browser and per device: one person on a phone and a laptop counts as two visitors, and clearing site data or using a private window starts over. Safari and iOS also delete script-written storage after seven days without a visit to your site, so a visitor who returns later than that on Safari may be assigned again. This is inherent to cookieless testing and rare enough not to move a result; run tests for a week or two and read the final numbers.

No cookies, and nothing that identifies a person. A random identifier generated in the visitor’s browser is kept in local storage so they stay on one variant; it travels with each impression or conversion to your own site, so every visitor is counted once. Your database stores that identifier, the variant, the event, a device class (desktop, mobile or tablet), the page path for conversions, the order value on a WooCommerce purchase conversion, and a timestamp. It never stores IP addresses, user agents, referrers or user accounts, and once a test is stopped, raw rows older than the retention window (90 days by default, adjustable) are rolled up into daily totals. Nothing is sent to us or to any third party. Exactly what the plugin stores.

The AI features are optional, Pro, admin-side and bring-your-own-key: nothing is sent unless you, the site owner, click a button. Copy ideas send the selected text, your direction if you typed one, plus the page URL, title, description and the element’s tag name; a rewrite of every field sends only the list of texts. Style sends the selected field’s markup, its current variant CSS, your instruction, the page title, the tag name and a summary of the page’s colours and fonts. Requests go straight from your server to the provider you configured (OpenAI, Anthropic or Google Gemini) under your own key and their terms; nothing touches improveab.com. The key is stored in your WordPress options, readable only by administrators and never sent to the front end, and no AI usage is logged. Visitor data is never part of any request, and Style cannot add JavaScript to your page.

It is built to keep that decision simple. For visitors it sets no cookies and stores no personal data: no IP address (the IP is used only as a hashed key to count requests per minute, and those counts are deleted by the plugin’s nightly clean-up at the latest), no user agent, no account. The one thing it keeps is a random identifier in local storage. Whether storing that identifier for testing needs consent is a legal question for you as the site owner: the EDPB reads ePrivacy Article 5(3) as covering local storage as well as cookies, and the plugin does not make that call for you. The optional AI features send admin-selected text to the provider you configure, never visitor data. You remain the data controller for your own site. A/B testing without cookies, explained.

No. By default, variant pages point their canonical at the original - Google’s own recommendation for A/B tests - and it also points your SEO plugin’s own canonical at the original through its filter, so Yoast, Rank Math, All in One SEO and SEOPress agree with it. In Settings you can switch to noindex instead, or turn the handling off and leave SEO entirely to your plugin. SEO settings for URL tests.

Free runs A/B and A/B/n tests with up to six variants, as many at once as you like, with every goal from a page visit to a WooCommerce purchase with revenue, page targeting, a device split, a day-by-day chart and CSV export - plus point-and-click creation with whole sections as fields, image swaps and undo, keep-showing-the-winner, a sample-size planner and clear results. Pro adds statistical significance and Bayesian probability, automation with scheduling and alerts, webhooks, role controls, audience targeting with traffic weights, per-variant custom CSS, and AI Copy and Style with your own key. The full Free vs Pro comparison.

Yes, everything stays. Pro is a complete version of the plugin that lives in the same plugin folder: under Plugins › Add Plugin › Upload Plugin, upload the Pro zip and choose “Replace current with uploaded”. Every test, result and setting carries over and running tests keep running; then enter the licence key. Going back to the free plugin the same way deletes nothing either: what only Pro uses stays stored for when Pro comes back.

In Pro, yes: role controls let editors, shop managers or any role you choose create and run tests, while settings, API keys and the licence stay administrator-only. In Free, tests are managed by administrators. Either way, staff who can edit posts are excluded from tests by default, so your own team does not skew the numbers.

Single covers 1 site, Professional covers 10, and Agency is unlimited. A licence is tied to each WordPress install you activate it on; you can move it by deactivating it on the old site. Licence and updates guide.

Yes - that’s what Professional and Agency are for. Use Professional for up to ten client sites, or Agency for an unlimited number.

Through Dodo Payments, acting as Merchant of Record. They take payment securely and handle applicable taxes. The plugin ships no secret keys and stores no card details.

Yes. Every Pro plan comes with a 14-day money-back guarantee on its first purchase: email [email protected] within 14 days and we will refund it. Refunds are processed by Dodo Payments, our Merchant of Record.

Updates and the features that need the licence stop, but nothing is deleted and nothing switches off mid-test: Pro stays installed, and tests that are already running keep serving variants and recording results, AI styling and custom CSS included. You can still create, run and copy tests without limits, because the plugin has none: a copy keeps its source's split and audience, and a brand-new test gets an even split and the default audience. Custom CSS stays available too. Scheduled starts, alerts, webhooks, the AI tools, new role grants and setting an audience or an uneven split stop, and the advanced analysis is hidden until you renew; your Pro settings stay stored for when you do. (Once the update server is live, Pro updates appear in your normal WordPress Updates screen while your licence is active; the free plugin updates from wordpress.org as usual.) What lapsing changes, in detail.

Planned next: exporting and importing tests as JSON (Free 1.1), revenue analysis and verdicts per device (Pro 1.1), shareable preview links and scheduled reports (Pro 1.2), test templates (Free 1.2), WP-CLI commands (Free 1.3) and AI test ideas from the page (Pro 1.3). Later: a learnings archive, multi-step funnel goals, personalization rules, optional GA4 event forwarding and AI Style for whole sections. Each item carries the release it is planned for, which is a plan, not a ship date. Multivariate testing is being researched with no commitment. Everything else described on this site is already in the plugin. The roadmap.