Page builder text extraction

Elementor, Oxygen, Breakdance, Beaver

Real extractors for each page builder’s data blobs. Whitelisted text keys plus a shared noise filter that skips hex colours, CSS units, URLs and dates.

code: class-queryra-postmeta.php:98-214

WordPress 7.0 APIs

Connectors + Abilities registration

Registers as a native search_connector and exposes queryra/semantic-search via the Abilities API. Feature-detected, so it stays quiet on WP < 7.0. AI Search plans to ship this in 1.41.0.

code: class-queryra-connector.php, class-queryra-abilities.php

Reliability patterns

Escalating backoff + PII redaction

API-down backoff climbs 1min → 5min → 30min → 6h. Outbound error events scrub IPs, hostnames, ports, internal paths before shipping. Careful patterns worth adopting.

code: class-queryra-search.php:519, class-queryra-analytics.php:138

Frontend UX

Two Gutenberg blocks

AI Search Box (live dropdown) and AI Search Button (full-screen overlay). Real blocks in the inserter, styled by your theme, built on the Interactivity API. Queryra ships zero frontend JS: search is a POST to their backend, results render in your theme’s default search template.

code: blocks/search-box/, blocks/search-button/

Trust model

Auditable ranking, local vectors

Every ranking rule (SKU pinning, keyword tier, cosine similarity, term boost) is in the plugin source. Embeddings live in your own postmeta. If our service ever disappears, your existing search keeps working from cached vectors. Queryra stores everything on queryra.com and ranks server-side.

code: ai-search.php: filter_search_results, calculate_similarity, merge_tiers

Compositional search

Keyword results always kept

AI Search hooks posts_results: WordPress finds candidates, we re-rank and merge (exact → keyword → semantic). Products your visitors already find keep finding. Queryra hooks pre_get_posts and replaces IDs entirely – deactivate the plugin and the results your customers were seeing change.

code: ai-search.php: filter_search_results (posts_results filter)

Cross-lingual

Query in any language, catalogue in any language

Real cross-lingual matching via the underlying multilingual embedding model. A shopper searching “sneakers” finds products titled “tênis” or “zapatillas”. Queryra’s readme.txt lists WPML / Polylang / TranslatePress / Weglot as “trusted compatibility”; the code has zero references to any of them.

See cross-lingual demo

Provider choice

Bring your own OpenAI key

Switch AI Search from the free service to your own OpenAI account whenever you want – unlimited indexing, your bill, your data. Queryra ties you to queryra.com: no other provider, no self-hosting option in the plugin.

code: includes/class-ai-search-service.php

Pick AI Search if

• You use Gutenberg block themes or Full-Site Editing
• You want a live search dropdown as a native block
• Your store is multilingual and shoppers write in different languages
• You want the option to bring your own OpenAI key
• You want a free tier that supports real search volume
• You want ranking logic you can read and audit
• You value not being locked into a hosted backend

Pick Queryra if

• Most of your content lives inside Elementor, Oxygen, Breakdance, or Beaver Builder
• You want to register as a WordPress 7.0 native Connector today, not next release
• You have a B2BKing wholesale setup and want a first-party integration
• You prefer the pre_get_posts model where ranking happens server-side
• Your search volume is very low (under 100 searches/month) and the free tier fits