{"id":16,"date":"2025-11-23T17:43:59","date_gmt":"2025-11-23T17:43:59","guid":{"rendered":"https:\/\/pilago-ai.io-dev.se\/?page_id=16"},"modified":"2026-01-15T14:17:22","modified_gmt":"2026-01-15T14:17:22","slug":"technical-foundations-behind-io-semantic-search","status":"publish","type":"page","link":"https:\/\/pilago-ai.io-dev.se\/?page_id=16","title":{"rendered":"Technical foundations"},"content":{"rendered":"
\n\t
\n\n\t\t
\n\n

Technical foundations<\/h2>\n\n\n\n
\"\"<\/figure>\n\n\n\n

iO Smart Search is not a single algorithm, but a collection of carefully combined building blocks that together create a modern search experience for WordPress. Each part is designed to solve a specific problem \u2013 from understanding meaning, to ranking trustworthy content, to keeping the system observable and safe in production.<\/p>\n\n\n\n

Hybrid search: keywords + semantic vectors<\/h3>\n\n\n\n

At the core of the plugin is a hybrid ranking model that blends classic keyword search with semantic vector search. Keywords are still great for exact matches and short, \u201cknown item\u201d queries, while vectors capture meaning even when users phrase things differently from the content. For every query, iO Semantic Search runs both layers and combines them into a single score using a tunable weight: you can decide how dominant the semantic part should be compared to raw keyword matches.<\/p>\n\n\n\n

Embeddings and AI providers: OpenAI and Bonzai<\/h3>\n\n\n\n

To understand meaning, the plugin converts text into vector embeddings. These embeddings are generated through either OpenAI\u2019s API or iO\u2019s own Bonzai platform, which acts as a universal gateway to multiple large language models. The admin can choose which provider and model to use for embeddings and summaries and optionally enable fallback: if the primary provider fails, the plugin automatically tries the other one before falling back to classic WordPress search. This gives you the benefits of AI without tying the site to a single external service.<\/p>\n\n\n\n

Vector backends<\/h3>\n\n\n\n

MySQL and QdrantEmbeddings are stored in a dedicated vector backend. For small and medium sites, iO Smart Search uses a custom MySQL table in the same database as WordPress \u2013 easy to back up, easy to host, perfect for local MAMP setups and simple environments. For larger installations, the plugin can switch to Qdrant, a dedicated vector database accessed via REST. The rest of the code does not care which backend you use; the choice simply changes performance and scaling characteristics, not the public API.<\/p>\n\n\n\n

Content authority and per\u2011post control<\/h3>\n\n\n\n

Not all content is equally important. To reflect that, the plugin lets you define authority levels (for example \u201cofficial\u201d, \u201cnormal\u201d, \u201copinion\u201d) with different weights. Editors assign these levels per post via a metabox, and the final ranking score is multiplied by the authority weight. This ensures that official pages and policy documents naturally rise above opinion pieces and blog posts, even when they talk about similar topics.<\/p>\n\n\n\n

Fuzzy search, synonyms and autosuggest<\/h3>\n\n\n\n

Real users misspell things and use their own terminology. iO Smart Search tackles this with a combination of fuzzy\u2011friendly keyword logic, editor\u2011defined synonyms and live autosuggest. Synonyms and aliases (such as SUV \u2192 XC90, XC60) are injected into both keyword and semantic layers so the system understands domain language. Autosuggest is powered by a lightweight REST endpoint that uses the same hybrid ranking as full search, meaning that suggestions and final results stay in sync and still benefit from authority and synonyms.<\/p>\n\n\n\n

Headless\u2011ready APIs and classic theme integration<\/h3>\n\n\n\n

All of the above is exposed both to classic WordPress themes and to headless frontends. The plugin integrates with the native search query to override results when enabled, but it also ships a dedicated REST API endpoint that returns ranked results as JSON. This allows the same search logic to drive Gutenberg themes, custom React\/Next.js frontends and even external tools without duplicating ranking or embedding code.<\/p>\n\n\n\n

AI\u2011generated summaries of results<\/h3>\n\n\n\n

On top of the ranked list, the plugin can generate a short AI summary of what the search results collectively say about the user\u2019s question. It does this by bundling the top hits (titles, types, URLs and excerpts) and sending them to the configured AI provider with an editor\u2011defined system prompt. The summary is always based on content that has already been retrieved and respects per\u2011post settings, authority levels and languages. Title and disclaimer text are configurable so you can clearly communicate that the summary is AI\u2011generated.<\/p>\n\n\n\n

Observability and debug tools<\/h3>\n\n\n\n

To make all these layers understandable and trustworthy, iO Smart Search exposes rich debug information for administrators. For each query you can see the highest combined score across results, which provider and model were used and whether a fallback kicked in. On the result level you can inspect keyword scores, semantic scores, authority multipliers and content type. This makes it much easier to explain \u201cwhy did this page appear where it did?\u201d and to tune thresholds and weights with confidence.<\/p>\n\n\n\n

Taken together, these technical components turn WordPress search from a simple text filter into a controlled, observable search pipeline: one that understands meaning, respects editorial priorities, tolerates real\u2011world user behaviour and remains portable across themes, environments and AI providers.<\/p>\n\n<\/div>\n\n\t<\/div>\n<\/section>","protected":false},"excerpt":{"rendered":"","protected":false},"author":1,"featured_media":0,"parent":0,"menu_order":0,"comment_status":"closed","ping_status":"closed","template":"","meta":{"_acf_changed":false,"footnotes":""},"class_list":["post-16","page","type-page","status-publish","hentry"],"acf":[],"_links":{"self":[{"href":"https:\/\/pilago-ai.io-dev.se\/index.php?rest_route=\/wp\/v2\/pages\/16","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/pilago-ai.io-dev.se\/index.php?rest_route=\/wp\/v2\/pages"}],"about":[{"href":"https:\/\/pilago-ai.io-dev.se\/index.php?rest_route=\/wp\/v2\/types\/page"}],"author":[{"embeddable":true,"href":"https:\/\/pilago-ai.io-dev.se\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/pilago-ai.io-dev.se\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=16"}],"version-history":[{"count":3,"href":"https:\/\/pilago-ai.io-dev.se\/index.php?rest_route=\/wp\/v2\/pages\/16\/revisions"}],"predecessor-version":[{"id":103,"href":"https:\/\/pilago-ai.io-dev.se\/index.php?rest_route=\/wp\/v2\/pages\/16\/revisions\/103"}],"wp:attachment":[{"href":"https:\/\/pilago-ai.io-dev.se\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=16"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}