Short answer: Relevanssi rebuilds WordPress’s keyword search with smarter scoring, fuzzy matching, and thorough indexing of custom fields, taxonomies and ACF data. AI Search adds a semantic layer that understands what visitors mean, not just what they typed. They approach the same problem from different angles — and for a lot of sites, the answer is Relevanssi.
What each plugin actually does
Relevanssi replaces WordPress’s built-in search with its own MySQL-backed index and scoring engine. It weights matches by where they appear (title vs body vs custom field vs comment), respects taxonomies, indexes ACF and WooCommerce fields, catches misspellings with fuzzy matching, supports exact-phrase queries with quotes, and lets you configure synonyms and stopwords per site. It has been in active development since 2010 with over 100,000 installs. In short: it makes WordPress’s keyword search do everything keyword search should do — and it does it well.
AI Search works on meaning instead of words. Every post, page, product or taxonomy term gets a numeric fingerprint (an embedding, from OpenAI’s text-embedding-3-small on Free or text-embedding-3-large on Standard/Pro). When someone searches, the query gets its own fingerprint and matching is done by geometric proximity in meaning. That is what lets it match “trainers” to “sneakers”, or a Portuguese query to an English product description, without you ever writing a synonym list. Exact and keyword matches always rank first; semantic matches are appended after them. Two optional Gutenberg blocks give you a live search-as-you-type dropdown.
Side by side
| Relevanssi | AI Search | |
|---|---|---|
| Core approach | Better keyword search (lexical) | Adds semantic on top of keyword |
| Understands “trainers” ≈ “sneakers” | Only with a synonym list (Premium) | Yes, out of the box |
| Cross-language matching | Per-language via WPML/Polylang | Native, especially on Standard/Pro model |
| Typo tolerance | Fuzzy matching, native | Partial (semantic overlap) |
| Custom fields / ACF | Deep, mature | Configurable via UI |
| PDF / media contents | Yes (Premium) | No |
| WooCommerce SKUs, attributes | Yes | Yes (SKUs pinned first, exact + prefix) |
| Live dropdown / search-as-you-type | Third-party integration | Built-in Gutenberg blocks |
| Configuration required | Some (weights, fields, synonyms) | Minimal (setup wizard, defaults work) |
| First released | 2010 | 2024 |
| Active installs | 100,000+ | ~500 |
| Free tier limit | Unlimited indexing | 2,000 embeddings |
| Paid tier | Premium from ~$99/year | Standard from €31.85/year |
| Runs offline | Yes, fully self-hosted | Needs an embedding service (built-in or your OpenAI key) |
When Relevanssi is the right choice
- Content-heavy blogs and documentation sites where readers search with terminology they already know. Relevanssi’s scoring is tuned for exactly this and has been for a decade.
- Custom fields and ACF are central to your content model. Relevanssi Premium indexes them thoroughly with fine-grained weight control.
- Typo tolerance matters — readers misspell author names, technical terms, product codes. Relevanssi’s fuzzy matching is native and configurable.
- You need to index PDF or Office documents in the media library. Relevanssi Premium does this; AI Search does not.
- You want zero recurring cost or offline operation. Relevanssi Free covers a lot of ground and there is no external service in the loop.
- Battle-tested integrations with WPML, Polylang, WooCommerce and Yoast are non-negotiable. Relevanssi has had 10+ years to polish these.
When AI Search is the right choice
- Visitors describe what they want in natural language. “Something warm but breathable for hiking”, “a book about grief that isn’t depressing”, “how do I stop my sidebar from collapsing on mobile.” Keyword search — even Relevanssi’s — needs the right words. Semantic search does not.
- Multilingual sites where queries and content mix languages. A Portuguese shopper searching “vestido de verão às flores” on an English product catalog. AI Search on the large model handles this natively; Relevanssi needs per-language dictionaries and synonym lists to approach it.
- E-commerce with attribute nuance. “USB-C hub with HDMI 4K 60Hz” vs a generic “USB hub”. Semantic distinguishes these without you writing product-attribute rules.
- You want it to just work. AI Search’s setup wizard picks sensible defaults, indexes your content, and semantic results start appearing immediately. No weight tuning, no synonym lists, no field-by-field configuration required to get useful results.
How they differ under the hood
Relevanssi builds and maintains its own search index in your WordPress database. When a post is saved, Relevanssi extracts every word (respecting weights and included fields), stores tokens, and scores queries against them with a customised BM25-ish algorithm. Everything happens locally — no external calls, no per-search cost.
AI Search converts each post into a high-dimensional vector using an OpenAI embedding model (managed by us on Free, Standard and Pro, or against your own OpenAI key). Those vectors live in post meta. At query time, the same model turns the visitor’s search into a vector and cosine similarity finds the semantically closest posts. It is more expensive per request than a MySQL lookup, but it catches meaning that no lexical algorithm can. The plugin also runs its own exact and keyword tiers before the semantic tier — so you don’t give up the good parts of keyword search to add semantic on top.
Real-world scenarios
A photography blog with 400 long-form articles
Relevanssi wins. Content is text-heavy, readers know photography terminology, and the value is in indexing that content well — not in semantic reinterpretation. AI Search would still work, but you would be paying for embeddings on content that keyword search handles perfectly.
A WooCommerce store with 2,500 products, selling in Portugal, Spain and France
AI Search wins. Cross-language matching between shopper query and product language is where semantic pulls ahead the most, and semantic understanding of variants (“trail runners” vs “walking sneakers” vs “lightweight hiking shoes”) is hard to replicate with synonym lists per language.
A documentation site for a software product
Either works. If your readers know the exact terminology (“how do I use the useEffect hook with dependencies”), Relevanssi’s keyword matching is enough. If they describe problems in natural language (“how do I make my component re-render when a value changes”), AI Search catches the intent that keyword search cannot. If you can afford it, running AI Search alone gives you both.
A membership site or forum with mixed content types
AI Search wins for the mixed-content-type search. Its 3-tier system handles the case where the exact word appears in one post type but the intent is really about another. Relevanssi can be configured to weight post types differently but doesn’t bridge intent gaps the way semantic does.
Can you run them together?
Not cleanly. Both plugins hook into WordPress’s posts_results and posts_search filters — whoever runs last wins, and the loser’s results are overwritten. In practice you pick one. If you want to switch between them to compare on your own site, both are one-click activate/deactivate; neither destroys the other’s data on deactivation.
Migration between the two
From Relevanssi to AI Search: install AI Search, run the setup wizard (five short steps, adds ~5 minutes on a typical site), then deactivate Relevanssi. Your existing search UI — the WordPress search block or your theme’s search — keeps working, only now it returns semantic-aware results. Reactivating Relevanssi later is one click.
From AI Search to Relevanssi: install Relevanssi, run its indexer, deactivate AI Search. Semantic results disappear, keyword recall improves for terminology-heavy searches. AI Search’s stored embeddings are left in place (post meta) and are ready if you reactivate.
Bottom line
Relevanssi is the mature, proven choice for content-heavy sites where keyword search — done well — is the answer. It has 10 years of polish, deep integrations, and a Free tier that covers a lot of real sites. AI Search is for cases where meaning matters more than the words used: multilingual stores, e-commerce with attribute nuance, sites where visitors describe what they want in natural language, and content that lives across languages or terminology.
If you’re not sure which fits your site: AI Search Free indexes 2,000 items on the small model, no card required, no configuration beyond the setup wizard. Turn it on, see if semantic results catch queries your current search misses, and decide from evidence rather than guesswork.
Last updated: September 2026. Relevanssi details based on its public documentation and plugin listing. If something is out of date, tell us and we’ll fix it.