Release history
AI Search changelog.
Every release, what shipped, and why. Newest first.
Inline notes live on the plugin page at wordpress.org.
Version 1.49.0
- Add: weekly email digest for the site admin. A Monday-morning summary of the past seven days — total searches handled, the top queries visitors asked, and the ones that returned nothing — lands in the
admin_emailinbox. Opt-in from the Insights tab (off by default); a “send me a test” button confirms delivery before the real cron fires. Free sites get a short “try Standard” footer; paid plans do not. - Add: WordPress dashboard widget. A compact panel on the
wp-admindashboard shows search volume for the week, the top three queries, and the number of no-results searches, each linking to the full Insights tab. Visible to any user withedit_posts, so editors see the content signal without needing options access. - Add: no-results search report. Queries that returned zero results now appear as their own panel on the Insights tab, treated as a content-gap signal rather than being discarded. The same report feeds the weekly email and the dashboard widget.
- Change: the Upgrade page is now rendered natively in
wp-admininstead of embedding/pricing/in an iframe. Standard and Pro are shown as side-by-side cards built from the plugin’s own styling, the admin’s email is forwarded to Stripe automatically so Stripe can send its own abandoned-checkout follow-up, and the full origin context (origin=wp_admin, site URL, install UUID, plugin version) is attached to the Stripe Session metadata — which makes it finally possible to tell which activations came from in-plugin upgrades versus the marketing site. - Change: admin menu cleanup. “Support Forum” is now “Ask for help! Support”, and the duplicate “Contact the author” entry has been removed.
Version 1.48.4
- Fix: installs that happened to share the same
get_site_url()as another install (two developers running onhttp://localhost:10008, a staging/production URL collision, a misconfigured multisite) were silently collapsed onto the same service record — same API token, same quota counters, same Fast Search cloud index. Each install now generates a per-install UUID on activation and sends it asX-Site-UUIDon every service call. The service treats(origin, site_uuid)as the composite identity, so colliding installs each get their own isolated record, token and cloud index. Legacy installs (any version before 1.48.4) keep working exactly as before. - Fix: sites whose initial registration with the AI Search Service was blocked by an outbound firewall, a security plugin or a transient network timeout are now recovered automatically. The plugin reattempts the registration in the background on the next three
admin_inithits (spaced 15 minutes apart), so a merchant whose activation call was swallowed by Wordfence or a cold DNS no longer silently stays unconnected. After three failures a persistent notice on the plugin’s own admin screens prompts a manual reconnect.
Version 1.48.3
- Fix: Fast Search now engages on the frontend search path for Standard and Pro merchants. A guard that resolved the plan via the quota manager class was skipping the semantic cloud path on every frontend request because the class was only loaded in wp-admin, silently falling back to the local cosine loop. Load the quota manager universally so the plan gate resolves correctly everywhere a search runs.
- Fix: nested fallback queries no longer re-enter the tier pipeline. When all three tiers returned nothing above the threshold, the enhanced fallback fired
get_posts({'s' => ...})which triggered another full semantic pass that often emptied again, looping until the request timed out. Inner queries now carry anai_search_skipflag that short-circuits the filter on re-entry.
Version 1.48.2
- Copy tweaks on upgrade and trial callouts to reflect the current 14-day free trial on annual plans.
Version 1.48.1
- Service requests now route through a global CDN for improved reliability and latency across hosts worldwide. No merchant action required.
Version 1.48.0
- New: dual-vector embeddings — on by default for every site. Each post is now indexed as two embeddings in addition to the existing one: a title vector (title + categories + tags + Woo attributes) and a body vector (description + custom + Woo fields). Semantic ranking blends the two cosines with a weighted-sum score, defaulting to 60% body / 40% title. Measurably better on intent queries (“something warm for winter”, “gift for someone who reads”) and cross-lingual search (“sapatilhas” matching sneakers) where the single-vector pipeline sometimes loses the right post in the noise. Exact-match queries (“mechanical keyboard”, “soy candle”) are unaffected — the two paths agree in those cases.
- New: a background backfill cron walks the catalogue 20 posts at a time until every post that already had a single-vector embedding has its dual pair stored. New publishes dual-write automatically. The Search Configuration tab shows the backfill progress and a Run Next Batch button for merchants who want to speed it along without waiting for the daily tick.
- New: body-weight slider on Search Configuration (30%–80%). Default 60% suits product catalogues; lower it toward title for sites where titles carry most of the identifying signal, raise it toward body for long-form content stores.
- Compatibility: the single-vector postmeta (
_ai_search_embedding_v2) stays as the fallback for every post that hasn’t been backfilled yet. The hot path mixes the two cleanly — posts with a dual pair are scored by the combined cosine, posts without one are scored by the single vector exactly as before. Fast Search merchants on Standard+ keep sub-30ms candidate selection from the cloud, and the plugin re-ranks those candidates locally with the dual pair when it exists. Zero merchant action required; no change to any existing setting; disabling dual (viawp option set ai_search_dual_embeddings_enabled 0) reverts the hot path to pure single-vector with no data loss.
Version 1.47.0
- New: on Standard and Pro, Sync with AI now also pushes each item to our Fast Search cloud index in the same batch. One click, both layers covered — merchants upgrading from Free get sub-30ms semantic results across their existing catalogue without having to run a second button. New publishes keep syncing to the cloud automatically, as before.
- Fixed: entering any AI Search admin screen now invalidates the plan cache automatically. A fresh plan read runs on the next pageview, so claiming a key — or any out-of-band plan change — is reflected immediately instead of up to 6 hours later. The merchant no longer has to press “Refresh quota” for Fast Search, Smart Matching and the upgrade banners to see the new plan.
- Fixed: the
/wp-json/ai-search/v1/refresh-quotaendpoint introduced in 1.46.2 returned HTTP 500 on REST calls becauseAI_Search_Quota_Managerlives under/admin/and only auto-loads on admin page requests. Added a lazyrequire_onceso the endpoint works from any REST client (including application-password callers). - Copy: references to the old “Indexing” tab across setup wizard, WooCommerce, custom fields and upgrade views now point at “Sync with AI” (the tab’s current label). References to a stand-alone “Quota tab” (merged into General in 1.29.0) now point at General. pt_PT / es_ES / fr_FR catalogs updated.
Version 1.46.2
- Fixed: after upgrading plan (Free → Standard, Standard → Pro), the Fast Search panel could stay greyed-out for up to 6 hours while the plan cache expired. Added a
POST /wp-json/ai-search/v1/refresh-quotaREST endpoint (requiresmanage_options) that drops the cache and refetches immediately; support can also use it to unstick a site without the merchant touching the admin.
Version 1.46.1
- Fixed: bulk sync could stop with “cURL error 28: Operation timed out after 40001 milliseconds” on larger batches. The HTTP timeout for batched embedding calls is now a flat 120 seconds (up from a
count × 2scaling formula that capped at 60s), which covers OpenAI slow days and Smart Matching enrichment without starving legitimate sync requests. Filterable viaai_search_embeddings_batch_timeoutfor hosts with shortmax_execution_time.
Version 1.46.0
- Fast Search is now available on Standard. Semantic ranking runs on our Postgres + pgvector cluster instead of your WordPress host, so search stays fast on catalogues of any size. The keyword tier and WordPress visibility rules still run in the plugin — nothing about the result quality changes. Previously limited to Pro.
- Default-on for qualifying plans. Standard and Pro sites get Fast Search without having to flip a toggle. The Search Configuration tab still carries the checkbox for merchants who want to turn it off (for example to compare against local ranking on their own hardware).
- Automatic fallback to local search. If the service is ever unreachable mid-request, the plugin silently falls back to the on-site cosine pipeline. No visitor sees a broken search.
- Backfill button on the Indexing tab pushes an existing catalogue to the cloud index in batches of 100 posts.
- Updated UI copy on the Search Configuration tab: the Fast Search panel now reads “STANDARD+” instead of “PRO” and points new merchants at the Standard plan when they upgrade.
Version 1.45.0
- Up to 7× faster search on large catalogues. Search vectors are now stored in a packed binary format at a smaller dimension size, which cuts the per-request cosine CPU from ~700 ms to ~90 ms on a 1 000-product catalogue, and 4× less postmeta storage. The AI ranking is faithful to the previous format (verified top-10 overlap = 10/10 on real embeddings across five diverse queries). No merchant action needed.
- Automatic index migration in the background. On update to 1.45.0, a daily WP-Cron job converts the existing index from the legacy JSON format to the new packed binary. The migration is fully off the critical path — zero wp-admin or frontend impact — and typically completes within a day on any traffic level. A notice on the admin screens confirms completion.
- No risk of data loss. The legacy index is kept intact alongside the new one until the migration finalises, and even then stays as a safety net (dual-write for new posts). A merchant who ever needs to roll back can do it with a single option change, with no re-sync required.
- Manual override card on the General tab for admins who want to force immediate completion instead of waiting for the cron.
- Smaller memory footprint. Peak memory per search dropped from ~25-65 MB to ~12-20 MB on 1 000 products — large catalogues on shared hosts (which were running close to
memory_limitandmax_execution_time) now have real headroom. - Developer: new helper class
AI_Search_Vector_Migrationwithpack_vector(),unpack_vector(),truncate_and_normalize(),dot()static helpers. The hot search path (process_posts_in_batches,compute_max_term_similarity,find_posts_by_term_match) dispatches on theai_search_vector_format_versionoption; both legacy JSON and new packed-binary postmeta coexist until finalisation.
Version 1.44.2
- Fixed: the “Contact the Author” submenu link pointed at a page that no longer exists on the author’s personal site, returning a 404 when clicked. It now points at the current contact page (
samuelsilva.pt/work-with-me).
Version 1.44.1
- [High] Removed the legacy
complete_setupadmin-post handler, which skipped CSRF verification by design and acceptedapi_keystraight from$_POST. A cross-site forged POST while an admin was logged in could silently swap the site’s OpenAI key for an attacker-controlled one. The handler had no live UI caller since v1.40.0 (the setup wizard writes settings through nonce-checked AJAX). - [High] OpenAI API key is now stored with autoload disabled, displayed masked, and swapped only on explicit “Change key” click. A one-shot migration flips existing installs’
ai_search_api_keyandai_search_service_tokenfromautoload=yestoautoload=noso these secret-bearing options are no longer loaded into the sharedalloptionscache on every request. The Advanced tab renders ask-•••••••1234preview instead of the raw key; theistype=passwordand the form only persists a new value when the admin clicks “Change key” and types one. - [High]
/wp-json/ai-search/v1/suggest?mode=fullnow has a site-wide daily ceiling of 500 calls (filter:ai_search_suggest_full_daily_cap), after which it transparently servesmode=keyword. The per-IP minute throttle was no defence against a distributed attacker varying the query string on every request to burn the site’s embedding quota. The dropdown still shows keyword results past the cap, so visitors see no degradation. - [High]
/wp-json/ai-search/v1/settingsno longer leaks environment fingerprint and schema. Removedwordpress_version,php_version,custom_fields,woocommerce_fields,total_embeddings, andpost_typesfrom the public response. The endpoint now returns only boolean / low-entropy flags useful for debugging (has_api_key,has_service_token,plugin_version,provider,woocommerce_active,openai_model,similarity_threshold,badge_public,setup_completed). Also removed the per-requestSELECT COUNT(*)onwp_postmetathat was a cheap DoS amplifier on large sites. - [High] OpenAI embedding call now honours the same timeout contract as the AI Search Service path (3 s on the search path, 20 s on the publish path, filterable via
ai_search_query_embedding_timeout/ai_search_publish_embedding_timeout). The OpenAI branch previously fell back to WP’s 5 s default; on a slow OpenAI day each synchronous search tied up a PHP worker for the whole 5 s. - [High] Removed the hardcoded placeholder
Authorization: Bearer your-secret-tokenfrom the service registration call. A dev-time string that shipped to production, visible in every build. The service identifies clients by theirOriginheader; the header is now gone.
Version 1.44.0
- Renamed “Smart embeddings” to “Smart Matching” across the admin UI. Same feature, less jargon — merchants don’t need to know what an embedding is to understand that we match queries to products more intelligently.
- New: Smart Matching spotlight on the General tab. First thing a merchant sees when they open the plugin — gradient hero card with the “up to 60% better semantic results” headline and a one-click CTA (7-day trial for Free, deep-link to the toggle for Standard+). Hidden automatically once Smart Matching is active for the site.
- Updated the Standard upgrade banner copy to lead with Smart Matching (“cheap sofa” → “affordable couch” example) instead of the generic “better model” pitch. Snooze reduced from 60 to 30 days — the banner comes back twice as often while still respecting the permanent “Don’t show again” opt-out.
Version 1.43.3
- Search never hangs on a slow service. The query-embedding HTTP call used to allow 20 seconds before falling back — on a bad day for OpenAI or our Railway edge, every visitor on every site watched a spinner that whole time. The search path now gives up after 3 seconds and transparently falls through to the exact + keyword tiers so the shopper still sees results. Publish-side embedding (which runs async in cron) keeps the generous 20-second budget. Both values are filterable via
ai_search_query_embedding_timeoutandai_search_publish_embedding_timeoutfor stores on slower networks or stricter latency budgets.
Version 1.43.2
- “Upgrade: 7-day trial” sidebar item painted green so the Free-plan CTA pops in the wp-admin menu. Hover state uses a brighter green on top.
Version 1.43.1
- Smart embeddings panel moved to the “Sync with AI” tab. The toggle used to sit under Advanced; it now sits above the Sync button, where it belongs — the admin reads the “boost this sync” option in the same flow as pressing Sync itself.
- New: Smart embeddings trial card inside the setup wizard. Free-plan users see a one-click “Start 7-day free trial” offer before their first sync, so brand-new installs can convert without ever leaving the wizard. Standard+ users don’t see the card.
- Translations across 11 locales for the new Smart embeddings copy — pt_PT, pt_BR, es_ES, fr_FR, de_DE, it_IT, nl_NL, ja, sv_SE (Swedish), pl_PL (Polish) and zh_CN (Simplified Chinese) all ship with Smart embeddings UI, trial card and upgrade badge translated. The 3 new locales are seeded with just the paid-feature strings for now; the broader catalogue will land via wp.org GlotPress community contributions.
Version 1.43.0
- Faster saves: publish hook no longer blocks on the embedding call. When a post or product is published or updated, the embedding is queued via
wp_schedule_single_eventand runs on the next WP cron tick (typically within seconds) instead of making the admin wait for the service round-trip. On shared hosts where the HTTP call added 500ms-2s to every save, the admin UI now returns instantly. Semantic search coverage for the just-published item backfills immediately. Developers who want the old synchronous behaviour can return false from the newai_search_async_publish_embeddingfilter. - Faster search: lexical scoring N+1 fix. The keyword-match scorer used to run one
wp_get_post_termsand oneget_post_metaper taxonomy per result inside the search loop — about 100 queries for a 20-result WooCommerce search. The filter now primes the object cache for every lexical candidate before the loop, cutting that to two batched queries. - Faster admin: Sync with AI tab caches its index-status queries. The “Indexed / Needs AI sync” table at the top of the tab used to run two JOIN+GROUP BY queries on
wp_posts × wp_postmetaon every page load — hundreds of milliseconds on large catalogues. Results are now cached for 1 hour and busted automatically when a post is published, updated, or deleted. - Faster search on large sites: early exit for the semantic loop.
process_posts_in_batchesnow stops scanning as soon as it has collected 100 above-threshold candidates. The visible result set never changes (WordPress only renders ~10-30 per page), but the worst-case cosine loop on 2000-post sites shrinks by ~50-150ms. - Faster term retrieval: capped at 500 terms. The phase-two taxonomy-term lookup only keeps the top 25 matched terms anyway; loading unbounded rows + JSON-decoding every embedding was pure waste on stores with thousands of attributes and categories.
- Service-side perf: OpenAI client now uses explicit 15s timeout and a keep-alive HTTP agent (~50-150ms saved per call, bounded worst case). MySQL pool bumped to 20 connections for bulk-sync concurrency.
- New: Smart embeddings (Standard plan). When enabled, each product or post is enriched by our LLM into 3-5 alternative phrasings a customer might search for (“Hoodie Oversized” → “chunky sweatshirt · baggy pullover · streetwear hoodie…”) before the vector is generated. Queries like “cheap sofa” start matching products titled “affordable couch” without any search-time cost — the enrichment is cached per-text and runs once at publish. Zero latency added to visitor’s search. Setting is under Advanced; Free users see it greyed out with an upgrade badge.
- New: paid-tier feature gating infrastructure. Reusable
AI_Search::plan_meets()andAI_Search::feature_lock_badge()helpers let any settings view expose a Standard-or-Pro feature with a greyed-out control and an “Available on Standard — Upgrade” badge. Powers Smart embeddings today, Fast Search’s Pro gate, and anything added in future releases.
Version 1.42.0
- Fixed: Indexing tab no longer times out on large catalogues. The “Sync content with AI” button on the Indexing tab used to POST the entire run in a single request, calling the embeddings provider hundreds of times inside one PHP process. Sites with 500+ posts on shared hosting routinely hit the 30-second PHP timeout, the LB timeout, or plain white-screened halfway through. The button now drives the same batched AJAX loop the setup wizard uses: 20 posts per request, live progress bar, same options (post type, taxonomy filter, “regenerate everything” checkbox). Limit dropdown reaches 5000 now instead of stopping at 1000 — the network cost that made higher values dangerous is gone.
- Refresh-safe syncs. Every batch persists the run state in a 30-minute transient. Close the tab or reload the page mid-sync and the Indexing view will pick up exactly where it stopped — same CPT, same offset, same accumulated summary — with no user action needed. The transient auto-clears on done or fatal error.
- Faster syncs: batch embeddings. Every 20-post AJAX batch used to fire 20 HTTP calls to the AI Search Service (one per post). Now it fires exactly one, with all 20 texts inside a single request body. The service was already batch-capable; the plugin was just calling it in a loop. Result: on the AI Search Service provider, catalogue-wide syncs are ~10-15× faster, provider rate limits are hit far less often, and OpenAI costs are unchanged (same tokens either way). The singular publish hook is untouched; only bulk/backfill paths use the batched call.
- New: Fast Search cloud backend (Pro plan preview). Pro-tier sites can now flip a “Fast Search” toggle in Search Configuration that swaps the local PHP similarity loop for a hybrid lexical+vector query on our Postgres+pgvector backend. Same plugin, same tier merge, same badge look — the local pipeline still owns Tier 0 exact matches, Tier 1 keyword matches, and Tier 2b term-taxonomy boost. The toggle is greyed out on Free and Standard plans; on Pro it opts in per-site. Falls back to the local pipeline transparently on any service error.
- Fixed: Fast Search dropped lexical hits from non-English sites. The cloud tsvector column is indexed under each site’s locale (
portuguese,spanish,french, etc.). Search queries were sent without a language hint, so the service defaulted tosimple— which does no stemming and misses inflected words the index had already stored as stems. Queries now carrylanguage, matching what the index was built with.
Version 1.41.1
- Fixed: Test Search tab was returning the same results on both sides. The AI Search filter has an
is_admin()short-circuit that keeps semantic ranking off the wp-admin search screens (Posts list, Media library, etc.). Because the Test Search tab runs its comparison from inside a wp-admin AJAX call, that short-circuit fired for the AI query too, so both columns showed the raw WordPress results and merchants saw no difference. The filter now honours anai_search_force_runWP_Query var that the Test Search tab sets on its AI query, unblocking the semantic pipeline for that specific call. Other admin-side searches are unaffected.
Version 1.41.0
- Signed bypass mechanism for support diagnostics. A search request carrying
?ai_search_bypass=1plus anAuthorization: Bearerheader now skips the AI pipeline entirely and returns native WordPress results. Only the site’s own service token grants access, so nobody else can silently disable the plugin. Used by the AI Search Service sweep to measure “same site with AI on vs off” back-to-back, quantifying the plugin’s real per-host overhead. - Smaller embedding storage. Each embedding dimension is now rounded to 4 decimal places at write time. Resulting JSON is ~30% smaller, so MySQL data transfer and PHP
json_decode()on the hot search path both drop proportionally. Cosine similarity difference vs 6-decimal values is <0.0001, far below any real ranking threshold. Backward compatible: existing JSON reads exactly as before, no migration required, storage shrinks naturally as posts are re-indexed on future publishes.
Version 1.40.1
- Listing rewrite: pain-first description opening with concrete shopper queries, dedicated “Works with any WooCommerce store” section covering 8 verticals, integration name-drops (ACF, Meta Box, Pods, JetEngine, WPML, Polylang, Elementor, Divi, Beaver Builder, Bricks), and four new comparison FAQs answering “how is this different from Relevanssi / SearchWP / FiboSearch / Elasticsearch”. Copy pass to drop em-dashes and tighten clauses. No code change.
Version 1.40.0
- New: one-click “Sync content” setup. The five-step setup wizard is gone. In its place, a single screen that auto-detects everything (post types, custom fields, product taxonomies) and shows one button, “Sync content”. Custom fields default to all public meta_keys, term embeddings auto-enable on WooCommerce stores, and progress is live across every content type. The old “Advanced settings” link opens the existing Content tab for anyone who wants to prune. Existing users’ pruning is respected, auto-populate only fills post types that were never configured.
- Brand-styled sync experience: single card with circular progress ring (purple→pink gradient), on-brand typography, pill cards per content type that light up as each one finishes, and a green check mark when done. Matches the look of wp-search.ai rather than the wp-admin form idiom.
- Persistent “Sync content” nudge in wp-admin. As long as the site has content but zero embeddings, every wp-admin page shows a small brand-styled notice with a one-click CTA to the sync page. No dismiss button, the nudge disappears automatically the moment the very first embedding lands. Editors and lower users never see it. Data-driven: two indexed LIMIT-1 lookups per page load, cached per-request.
- Translations for the 34 new msgids across
pt_PT,pt_BR,es_ES,fr_FR,de_DE,it_IT,nl_NL,ja(each catalog now at 663 translated messages, 0 fuzzy).
Version 1.39.0
- New: “Before you leave” survey on plugin deactivation. When you deactivate AI Search from the Plugins screen, a small modal asks the single most useful question, why. Six one-click options (Too complicated, Poor results, Something broke, Switched to another plugin, Just testing, Other) plus an optional comment. The deactivation itself is never blocked; the modal is a one-screen ask, dismissable in a click. Every answer flows straight to us and shapes the next release.
- Translations for the 14 new msgids across
pt_PT,pt_BR,es_ES,fr_FR,de_DE,it_IT,nl_NL,ja(each catalog now at 629 translated messages, 0 fuzzy).
Version 1.38.1
- Free plan quota raised: 10,000 embeddings/month (was 2,000), no more daily cap. The service reset was already monthly-based since day one; the label just hadn’t been updated to say so and the numbers were tight enough to feel restrictive during first-day setup. Existing free-tier customers already have the new limits applied; new installs get them out of the box. Standard and Pro tiers are unchanged.
- Copy correction across the Quota tab and Setup Wizard: “Lifetime total” and “Total Embeddings” now read “This month”. The wording was misleading, the service was resetting
used_totalback to zero on the anniversary of the origin’s registration for as long as the plugin existed, but the UI called it lifetime. Now it says what it does. - Translations for the 3 new msgids across
pt_PT,pt_BR,es_ES,fr_FR,de_DE,it_IT,nl_NL,ja(each catalog now at 615 translated messages, 0 fuzzy).
Version 1.38.0
- New: “Test search” tab. A dedicated screen where you can run any query against your own content and see WordPress default results side by side with AI Search results. The AI-added items get highlighted so you can feel the plugin’s contribution at a glance. Includes a live threshold slider, a sync-status table telling you what’s ready to be tested against, and copy that explains the plugin works in the background, you don’t have to change your search UI.
- Removed: the frontend “AI Search Threshold” widget that used to appear on search results pages for logged-in admins. It was scary UX, admins could reasonably assume visitors saw it too. The Test search tab in wp-admin covers the same use case, safely.
- Tab consolidation: Tools merged into Advanced. Both were “settings a merchant rarely touches”; keeping them apart added navigation friction. Old
?tab=toolsand?tab=cachelinks continue to resolve. - Mini badge redesigned into a small green dot. Purple filled with the letters “AI” was overpowering serif titles; now it’s a subtle friendly green dot that scales with the parent text. Same tooltip on hover (“This search result was generated by an AI system.”). Same visibility gate, always visible to visitors by default (from 1.37.0), toggleable in the Appearance tab.
- Translations: 27 new msgids across
pt_PT,pt_BR,es_ES,fr_FR,de_DE,it_IT,nl_NL,ja(each catalog now at 612 translated messages, 0 fuzzy).
Version 1.37.0
- New: dedicated “Upgrade” (or “Subscription”) tab in the plugin settings. The old “Activate a paid plan” call-to-action that used to sit in big letters at the top of the General tab was giving new merchants the impression that the plugin required an upgrade to work, it doesn’t. Free is fully functional. Now activation and subscription management live in their own tab, renamed dynamically to “Subscription” once a plan is active. General stays quiet: connection status only.
- New: dedicated “Appearance” tab hosting the AI badge visibility toggle and two color pickers (background + text) for the tooltip. Live preview in the settings updates as you drag the color sliders.
- New: subtle round “AI” badge on genuine AI-tier search results (semantic matches the keyword search would have missed). Exact SKU/title matches and native keyword results no longer show any badge, those look like normal results because they *are* normal. Old rectangular Exact/Keyword/Fallback badges removed.
- New: custom tooltip on the badge in the plugin’s own visual style (dark purple background matching the wp-search.ai brand, small centered arrow, smooth fade-in) instead of the browser’s OS tooltip. Text: *”This search result was generated by an AI system.”* Configurable colors per-store from the Appearance tab.
- Default changed: badge is now visible to all visitors (previously editor-only). Merchants who want unbranded search results can turn it off in Appearance. The rationale: transparency about AI-assisted results is worth defaulting to.
- Fixed: the persistent “AI Search is ready!” success notice would not stay dismissed on some installs, the AJAX handler was hooked inside
admin_enqueue_scriptswhich does not fire onadmin-ajax.phprequests, so the dismiss click reached an unregistered action. - Fixed: welcome notice now respects the site locale. The first admin notice the merchant sees (“Welcome to AI Search! Let’s set up your intelligent search system…”) was hard-coded English in every previous release; every non-English locale saw it untranslated. Now wrapped in
__()and covered by the 8 shipped translation catalogs. - Renamed: “Keyword Match” copy and the whole notification stack around fallback results was pruned. If a result isn’t semantically AI-tier, no badge appears, the plugin no longer explains what it *didn’t* do.
- Translations: 20 new msgids added and translated to
pt_PT,pt_BR,es_ES,fr_FR,de_DE,it_IT,nl_NL,ja(each catalog now at 585 translated messages, 0 fuzzy).
Version 1.36.0
- New: plugin now talks about “syncing your content with AI” instead of “generating embeddings” throughout the admin. The vocabulary was accurate but reads like documentation, not like something you actually want to click. Every user-facing button, heading and status line was renamed: “Generate Embeddings” → “Sync content with AI”, “Regenerate Embedding” → “Re-sync with AI”, “Generate Term Embeddings” → “Sync taxonomy terms with AI”, and so on. Internal action names, meta keys and developer hooks are unchanged, existing integrations keep working.
- New: dedicated Advanced tab for the provider chooser and OpenAI API key + model. New installs no longer see the choice at all, the AI Search Service is set up silently in the background and that is that. The controls are still there for anyone who wants their own OpenAI key, just no longer in the way of the 95% who do not.
- Setup Wizard step 2 is now silent. The old “Choose Your AI Provider” step with two big cards is gone; the wizard now auto-registers on the AI Search Service and shows a short “Connected & Working” confirmation. A small link points to the Advanced tab for anyone using their own OpenAI key. One fewer decision for a first-time user.
- Fixed: the “AI Search is ready!” success notice would not stay dismissed. Clicking the ✕ made it disappear, then it came right back on the next page load, the AJAX handler that persists the dismissal was registered on a hook that does not fire during admin-ajax.php requests, so the click reached an unrecognised action and the option was never saved. Handler is now registered globally.
- Five more languages, five out of the box:
de_DE(German),it_IT(Italian),nl_NL(Dutch),pt_BR(Brazilian Portuguese) andja(Japanese) join the existingpt_PT/es_ES/fr_FRcatalogs. All 8 locales carry the full admin UI at 565/565 strings translated. Portuguese speakers finally get pt_BR-specific vocabulary (usuário, salvar, configurações, conexão) rather than the pt_PT catalog fallback. - New listing title: AI Search – Semantic Search & Live Dropdown for WordPress and Woo. The previous “Smarter WordPress Search” title read broadly but hid what the plugin actually does. New title leads with the two headline features, mentions both audiences.
- Changelog condensed: the readme was pushing wp.org’s file-size warning. Detailed entries kept for 1.30.0 onwards; older releases summarised into decade-style groupings. History still readable, no functional change.
Version 1.35.0
- New: “Regenerate posts that already have an embedding” checkbox on the Indexing tab. Upgrading from Free to Standard or Pro switches every new embedding to the larger, more precise model, but old vectors stayed on the small one until you cleared them in Tools and then re-indexed from scratch. Now it is one step: tick the box, pick the post type, generate. Existing embeddings get overwritten on the current model, no separate clear-then-regenerate ritual.
- New: “By model” column in the index-status table. For each content type, shows how many posts are on
text-embedding-3-large, how many are ontext-embedding-3-small, and how many were indexed before this version started recording the model (they show up as “unknown” and get resolved the moment you regenerate them). - Auto-dismiss for the “regenerate your embeddings” reminder that appears after activating a paid plan. Once every indexed post is confirmed on
text-embedding-3-large, the reminder card removes itself, no more “Remind me later” click when you have already done the work. - Auto-dismiss for the “search as you type” blocks announcement on the General tab. Once any post, page, template or pattern on the site contains an AI Search Box or AI Search Button block, the card removes itself, because the site already found the feature.
- “Don’t show again” links on both the Pro upgrade banner and the quota-warning banner, next to the existing “Not now” snoozes. If you have decided you are staying on Free, one click silences the banner permanently instead of the 60-day (Pro banner) or 3-day (quota banner) reminder cycle.
- Longer TTL on the “Welcome to AI Search Standard/Pro” celebration shown after activating a paid plan. Was 60 seconds; if you got sidetracked between clicking Activate and landing on the General tab you would miss it entirely. Now 24 hours, first view still consumes it, but coming back later today still gets you the hello.
- New: model + indexed-at line on every post’s “AI Search Status” meta box. Shows the exact model that produced the embedding (
text-embedding-3-smallortext-embedding-3-large) plus a “indexed 3 days ago”-style relative timestamp, inferred from the vector dimensions so it works retroactively on posts embedded by older plugin versions. - New: quota-warning banner on the plugin’s own admin screens. Two states: an amber warning at ≥80% of the monthly cap (“You are running low”), and a red block at 100% (“New content is no longer being indexed”). Only shown on Free-plan sites, dismissible with a short 3-day snooze because the underlying state changes fast. Suppresses the general Pro upgrade nudge while active so the user sees one urgent message, not two.
Version 1.34.9
- Outbound links to wp-search.ai (from the readme description/FAQ, the “Manage subscription” and pricing links on the General tab, the Pro banner, and the Upgrade menu iframe) now carry
utm_sourceandutm_mediumparameters so referrer traffic can be measured. Nothing user-facing changed.
Version 1.34.8
- Small label tweak: the Upgrade submenu now reads “Upgrade: 7-day trial” instead of “Upgrade, 7-day trial”.
Version 1.34.7
- Fixed: the new “Upgrade” screen shipped in 1.34.6 was blank. The plugin’s embedded pricing iframe used
?embed=1, which turns out to be a query variable WordPress core reserves for its oEmbed handler, so wp-search.ai returned the minimal oEmbed preview (a title and site name) instead of the actual pricing page. Switched to?minimal=1.
Version 1.34.6
- New: “Upgrade, 7-day trial” entry in the AI Search menu. Free-plan sites get a one-click way to compare Standard and Pro without leaving the WordPress dashboard: the pricing page opens right inside wp-admin as an embed. Only shown to Free-plan sites on the managed AI service, paid customers and BYO-OpenAI-key users don’t see it.
Version 1.34.5
- Setup Wizard stops nagging you once it’s done. After you finish the wizard, the persistent “Welcome to AI Search” bar at the top of every admin page is replaced by a one-shot success notice: *”AI Search is ready, check the Indexing tab if you still have content left to embed.”* Dismissible, doesn’t come back. The “Setup Wizard” submenu also drops out of the sidebar once setup is completed, the plugin isn’t asking to be set up any more.
- The wizard completion screen was also tightened: one clear congratulation and a direct link to the Indexing tab, replacing the old three-card grid that pointed at tab slugs the plugin renamed back in 1.29.
Version 1.34.4
- Fixed: setup wizard could die with “The link you followed has expired.” The General Settings tab guarded its save handler by checking for
provider+api_keyin the POST body, which collided with the same hidden fields the wizard’s step 4 and step 5 forms carry as state pass-through. When wizard route delegation missed, the tab would then run WordPress’s referer check against the wrong nonce and blow up. The save handler now requires an explicit form marker instead. New installs can complete the wizard end-to-end again.
Version 1.34.3
- Service endpoint moved to service.wp-search.ai. New installations point there for embeddings, quota checks, and billing. Existing installs continue to work against the previous Railway URL – no action needed. Both endpoints serve the same backend.
Version 1.34.2
- De-narrowed the copy: rewrote the listing title, description and FAQ so blog owners, documentation sites and directories can tell at a glance that AI Search is for them too – not just WooCommerce stores. The plugin has always indexed posts, pages, custom post types and custom fields; only the framing was store-flavoured. New title: “Smarter WordPress Search That Understands Meaning”.
- Translations (pt_PT, es_ES, fr_FR) updated to match.
Version 1.34.1
- Small in-plugin nudge for Free users: a compact card on the plugin’s own admin pages introduces the Standard plan (larger embedding model, higher monthly cap, 7-day free trial). Never shown outside the plugin’s own screens, never to customers already on Standard or Pro, and always dismissible – the “Not now” link snoozes it for two months.
- Fixed: after releasing the current site from a paid subscription, the “One last step: regenerate your embeddings” reminder no longer stays behind. It only appears when this site is actually on a paid plan.
Version 1.34.0
- New: Paid plans. Activate Standard or Pro directly from *AI Search → General* by pasting the activation key that arrives by email after checkout at wp-search.ai/pricing. Standard and Pro upgrade the site to
text-embedding-3-largeand raise the monthly quota. - Manage subscription in-place: the General tab now shows your current plan, trial status, and next renewal date. One click opens the billing portal to update your card, change plan, or cancel. Another click releases the site so you can move the subscription to a different WordPress install.
- Guidance after upgrading: once you activate a paid plan the plugin walks you through regenerating your embeddings on the new, more precise model — content indexed on the Free plan uses a smaller vector shape and benefits from being refreshed.
- Cleaner settings header: the redundant “AI Search Settings” heading was removed. The tab bar carries the same information without competing with page content for attention.
Version 1.33.1
- Fixed: the search box and search button blocks showed every matching result instead of the number configured on the block. A dropdown set to six results could return the entire catalogue.
- Fixed: product titles containing quotes or dashes displayed raw HTML codes in the dropdown (for example a 34″ monitor showing as 34″).
Version 1.33.0
- New: AI Search Box block: a search field with live results in a dropdown. Keyword and SKU matches appear as you type; AI suggestions are added underneath when you pause typing.
- New: AI Search Button block: a search icon or button that opens a full-screen search overlay with the same live results. Ideal for headers and menus.
- Typing stays free: every keystroke runs a keyword search, which costs no embeddings at all. Only one AI request is made once you stop typing, so a ten-character search uses one embedding, not ten.
- Results show a thumbnail, the product title and the price on a single row, with a “See all results” link to the full search page.
- Full keyboard and screen-reader support: arrow keys move through results, Enter opens one, Escape closes the list, and the whole dropdown follows the ARIA combobox pattern.
- Styling comes from your theme. The blocks read your theme’s own colour, typography and radius settings, so they look native on light and dark themes without any configuration.
- Frontend text ships translated in Portuguese, Spanish and French.
- Tier badges (“Exact”, “Keyword”, “AI”) remain visible to editors only, unless you have chosen to show badges to all visitors.
- Both blocks are entirely optional: if you do not place one, nothing on your site changes and no extra scripts are loaded.
Version 1.32.0
- WooCommerce has its own tab again: product field settings moved out of Content into a dedicated WooCommerce tab, shown only on stores.
- New store status panel: at a glance, how many published products are indexed, how many have SKUs, how many variations exist, how many products are hidden from search by catalog visibility, and how many categories and tags are indexed – each with a link to fix what is missing.
- Actionable warnings instead of silence: if products have SKUs but the SKU field is not being indexed, or if nothing is indexed at all, the panel says so and points you to the right screen.
- Field pickers are readable again: the custom field and WooCommerce field checkboxes are now a responsive card grid instead of CSS columns that flowed items vertically – easier to scan, and they collapse to a single column on small screens.
- Better WooCommerce defaults: product description, short description, SKU, categories and tags are now included out of the box. Fixes a mismatch where the settings screen showed defaults the indexer did not actually apply on stores that never opened the tab – most importantly the SKU, which powers exact-match search.
Version 1.31.0
- New Insights tab: see what users searched for and which results they were shown. Only searches where the AI surfaced results that exact and keyword matching had missed are recorded – so the list is a running record of the plugin earning its keep.
- Private by design: no IP addresses, no user accounts, no cookies – just the query, the result counts per tier, and the top result IDs. Nothing that needs a consent banner.
- Logging can be switched off, records are kept for a configurable window (30/90/180/365 days, default 90) and cleaned up automatically, and the whole log can be cleared with one button.
- Writes happen after the page has been sent to the visitor, so logging never slows a search down.
- Added
uninstall.php: removing the plugin now cleans up its settings and the log table. Post and term embeddings are intentionally left in place so an accidental uninstall does not cost you your whole index.
Version 1.30.0
- Products found by their categories and tags: search now also finds products whose *category or tag* matches what you typed, even when the product’s own text does not. A user searching “trainers” finds items tagged “Sneakers”; “running shoes” finds them too. Until now those products could stay invisible because their description never used your visitor’s word.
- Term-only matches are scored conservatively and can never outrank an exact SKU/title match or a strong content match – they are added to the results, never in front of them.
- New developer filter
ai_search_term_match_thresholdto tune (per taxonomy) how close a term must be to the query before it pulls in its products. - Requires term embeddings to be enabled (Indexing tab) – already the default on WooCommerce stores.
Version 1.29.0
- Rebuilt the settings screens with native WordPress components. Consolidated to five tabs (General, Search, Indexing, Content, Tools) with old tab links still working. Bundled full es_ES, pt_PT and fr_FR translations. Indexing tab shows per-content-type ratios (e.g. “14 of 18 products indexed”).
Version 1.28.1
- Fix: secondary WP_Query searches (blocks, widgets, AJAX handlers) now go through the tiered pipeline correctly. Previously they read the wrong
sfrom the global query.
Version 1.28.0
- Three-tier results (exact → keyword → semantic): WordPress’ own lexical results are never discarded, ordered by keyword relevance and shown before AI extras. Blue “Keyword Match” badge for editors. Themes’ “Showing all N results” counts now match reality. The old Hybrid Search toggle was removed (superseded by the always-on lexical tier).
Version 1.27.0
- Exact SKU + title pre-match, always on, pinned first with a green “Exact Match” badge. New
ai_search_exact_matchesfilter to add or remove pinned matches. Bulk generation now reports success/failure counts and reasons (bad token, exhausted quota, empty content). Products flaggedproduct_visibility=exclude-from-searchno longer appear anywhere. Site-wide search now actually runs the AI ranking (previously fell back to WordPress default).
Version 1.26.0
- Per-taxonomy-term embeddings with additive boost – a search for “trainers” now surfaces products tagged “Sneakers” even when no description mentions trainers. New Term Embeddings settings tab and a 0-50% Term Boost slider on the Search tab. Default ON for WooCommerce stores. New
ai_search_term_embedding_contentfilter.
Version 1.25.0
- Five new developer filters for customizing search behavior without forking:
ai_search_embedding_content,ai_search_should_generate_embedding,ai_search_query_before_embedding,ai_search_similarity_threshold,ai_search_results. See the Developer Hooks FAQ.
Version 1.24.0
- One-click “Reconnect with new token” on the General tab when the AI Search Service no longer recognises this site’s token. Quotas are preserved – only the credential rotates.
Version 1.23.0 – 1.23.1
- WooCommerce-first onboarding (product-priority defaults, “store detected” welcome banner). Documented multilingual support (100+ languages via semantic search). Fixed a PHP 8 DivisionByZeroError in cosine similarity. Added the first PHPUnit test suites. Fixed a “link expired” error on the setup wizard’s final step.
Version 1.22.0
- Domain mismatch detection (admin warning when site URL differs from the registered origin) plus a one-click reconnect for the current domain. Makes staging → production migrations painless.
Version 1.20.0 – 1.21.0
- Reusable CPT/taxonomy selector; clear embeddings by post type or specific taxonomy term; dynamic AJAX loading of taxonomies. Batched post processing (100/batch) for sites with thousands of items. 5-minute search-results cache. Default similarity threshold dropped from 65% to 30% for better out-of-box results.
Version 1.15.0 – 1.19.0
- Dedicated top-level admin menu with SVG icon (moved out of Settings). New Quota tab with real-time usage from the service. OpenAI model selector (
text-embedding-3-smalldefault,-3-largeavailable). REST diagnostics endpoint at/wp-json/ai-search/v1/settings. Threshold range extended to 10-100%. Setup wizard expanded to 4 steps with custom fields configuration. First Hybrid Search implementation (later replaced by the always-on lexical tier in 1.28.0). pt_BR readme translation.
Version 1.1 – 1.14.0
- Foundational releases: initial embedding-based search, per-post
custom_post_metastorage, transient caching, bulk generation UI,ai_search_searchable_post_typesfilter, i18n scaffolding, service registration and token validation, Cache Management tab, 4-tier fallback for zero-result searches, similarity CSS classes, percentage-based threshold display, automatic embedding on publish, Quota Management tab, service token controls.
Version 1.0
- Initial release, adding the settings page and the search filter.
Still running an older version?
Open AI Search inside WordPress and update to the latest. Every fix and every feature lands on Free first.
Download on wordpress.org →