Relevanssi already does the hard work: fuzzy matching, custom field weighting, partial words, taxonomy boosts, and per-post-type ordering. But the output is still a list of ten blue links. Visitors who can't translate a list into an answer bounce, and the wp_relevanssi_log table fills up with the same unresolved queries every week.

SleekAI maps the Relevanssi result set into a variable that the chatbot reads on every turn. Titles, excerpts, custom field snippets, and ACF values from wp_postmeta all land in the system message, and the model summarises the top three matches with a direct link to each. Per-post-type rules from Relevanssi survive because SleekAI re-uses the same query, so weighting for product pages versus docs still applies.

Most generic search bots index your site separately and drift out of sync within a week. SleekAI doesn't. It calls Relevanssi at request time, so a freshly published post is searchable the moment Relevanssi indexes it, and a removed page disappears from answers the moment Relevanssi drops it. The wp_relevanssi_log data feeds the same chatbot so you can spot the queries that never resolved and write a page for them.

Relevanssi already solves the ranking problem most WordPress sites have with the core search. It indexes custom fields, applies per-post-type weighting, handles synonyms, and logs every query so the editorial team can see what real visitors typed. But ranking is not the same as answering.

A visitor on a long documentation site or a deep editorial archive still has to read three or four results, scan headings, and decide which page actually addresses their question. Many will not. They will bounce or open a ticket.

Bolting a generic chatbot on top makes this worse, because it indexes the site separately, ignores Relevanssi's weights, and quietly hallucinates when fresh content has not made it into its own embedding store yet. SleekAI takes a different path. It calls Relevanssi at request time, inherits every weight and synonym and post-type rule already configured, and asks the language model to write one paragraph from the top three results with the canonical URL attached.

The visitor reads an answer, follows the link if they want detail, and the relevanssi log keeps recording the long tail of natural language queries the team can use to plan next quarter's content.