SearchCue: AI Site Search & Answers

SearchCue can answer a visitor's question using indexed content from your site and link to the sources used. AI answers can be incomplete or inaccurate, so visitors should use the cited sources to verify important information.

The plugin is free. The hosted service has a 15-day free trial with no credit card. Paid plans currently start at €9 per month. Current prices and limits are listed on the SearchCue pricing page.

The browser embed downloads without delaying the page's load event. If the site's Content Security Policy prevents this loading route, a native script request starts only after the page has loaded. SearchCue does not build or query a local search index. The plugin stores settings and pending updates in WordPress. Ordinary visitor and post-save hooks do not wait for SearchCue HTTP requests. Background updates and authenticated content exports still use PHP workers, CPU, and the database. WordPress cron normally runs in a separate loopback request, but spawning it, output buffering, and competing for a small worker pool can still delay visitors. Use an external cron runner if WP-Cron is disabled or if your host needs scheduled background work separated from page requests.

SearchCue is designed to answer questions from the content of your site. If your website does not support an answer, our system says as much rather than inventing an answer.

SearchCue extract text content from images and also does its best to to derive an understanding of what your images contain. This can make image content discoverable even when captions or alt texts are missing.

No. SearchCue includes a floating search launcher and can take over a standard search input or any link to /search. As with other frontend plugins, the theme must call the standard WordPress wp_head() function.

If WordPress could not add SearchCue, a Search on your website section appears in WordPress admin. It explains what happened and what to check next. Details for support includes the check result and a list of other plugins, their versions, and whether they are active. Use Copy diagnostics to share the report with support. The list provides candidates to check, not proof that a plugin caused the problem. The section disappears when the problem clears. There is no extra message during normal operation or first-run setup. SearchCue handles common changes made by performance plugins, including supported WP Rocket delayed loading. It respects cookie-consent rules and does not force blocked scripts to run. After changing a setting, clear your website's cache. Open your website and try a search. Then refresh the SearchCue settings page. A saved connection does not prove that search works for visitors, so always check the live website too.

Yes. Supported PDFs, Word documents, PowerPoint presentations, and spreadsheets can be indexed alongside your pages, so manuals, specifications, and price lists can appear in results.

Yes. When WooCommerce is active, SearchCue indexes published, catalog-visible products and visible variations as rich product results. Their prices, sale and stock status, categories, tags, and attributes can power faceted filters and refinement menus. Ordinary page crawling remains available if the catalog cannot be retrieved.

Yes, for active GTranslate Translation Delivery Network (TDN) installations that translate on the server and use language subdirectories on the same origin, such as /en/. Browser-only translation, language subdomains, and separate translation domains are not supported. SearchCue detects the configured source and enabled target languages. Product names, long and short descriptions, visible attribute names and values, category and tag labels, image alternative text, and public URLs are translated. Product and variation IDs, SKUs, prices, currency, and stock status stay tied to the current WooCommerce catalogue. Category hierarchies remain separate for each language. Proven product-page URL aliases are replaced by the corresponding rich product result without merging different languages or losing variations. Retained aliases are reconciled before applying partial-crawl page limits, so they do not displace ordinary pages. Editorial pages remain pages when they merely embed a product. Previously replaced product pages can fall back to retained page content when only the crawl request budget prevented a fresh fetch. Confirmed updates and removals still take precedence. The hosted crawler requests only bounded product selections through GTranslate, using a per-request translation header rather than changing global JSON translation settings. Cached text is reused when its source content is unchanged, including during price and stock updates, which apply to every cached language. Adding or removing another target language does not refresh valid cached translations for unchanged targets. On full crawls, cached translations become eligible for refresh after seven days to pick up manual translation edits; missing translations take priority. Incremental updates do not refresh otherwise valid text solely because it is old. The hosted crawler requests one parent product with its visible variations at a time, with at most 200,000 bytes of untranslated JSON per request. There is no cumulative byte allowance, product-count cap, or catalogue-wide translation deadline. WordPress enforces each request's size allowance, and the crawler retains individual network timeouts, rate limiting, and cancellation. Terminal page or document storage failures stop further product work rather than leaving translation requests running for a crawl that cannot be saved. Translation requests never follow redirects. Tags are translated as a list, and older cached translations that omitted them are refreshed. Local language detection checks substantial translated prose against the requested language, including whole paragraphs made up of short sentences, and checks for copied source passages. Short labels, proper-name lists, inconclusive detection, and languages outside the detector's coverage are handled conservatively. The same checks apply to cached text. Transient request failures and wrong-language prose get up to three attempts per parent-language pair. Individual failures, including families too large for one translation request, do not stop the remaining requests. Original products and successful translations are retained. A failed refresh can retain a complete, valid cached translation; otherwise eligible crawled or retained HTML provides a page fallback for the missing translation. Translation access, safety, and cancellation failures still stop product fetching. These checks cannot guarantee idiomatic machine translation or correct wording in every short label. Translation requests may still incur charges under your GTranslate plan. Crawl progress reports the language and batch being fetched, for example Fetching products (EN), distinguishes cached text from translation refreshes, and reports the number of products with translation failures. Completed crawl results show the distinct parent-product count, including failed refreshes that retained cached text. Failed product IDs, languages, and reasons are saved with the crawl's product payload.