I Read the Fine Print on 9 WordPress AI Search Plugins So You Don't Have To

Full disclosure before anything else: I build one of these. Queryra is mine. That is exactly why I went and read what the other eight actually say about themselves, rather than what their names imply.

I did not run a benchmark this time. I read plugin pages. All of them, line by line, the way nobody does before installing something. It took an afternoon and it was more revealing than any test I have run, because almost every meaningful difference between these plugins is sitting in plain sight, in their own words, one scroll below the headline.

If you want the verdict on which plugins genuinely use vector embeddings, I wrote that up separately in Which WordPress AI Search Plugins Actually Use Semantic Search?. That post answers which. This one answers who pays for it, and how you check any of it yourself.

"AI search" is not a protected term. A plugin can carry *AI*, *semantic* or *vector* in its name while running the same word-counting algorithm that WordPress search has used for twenty years.

I am not accusing anyone of lying. The opposite, actually: the plugins below are honest in their own documentation. The gap is between what the name promises and what the description says, and the description is usually two paragraphs down.

So this post has one rule. Every claim below is quoted from the plugin's own page and you can check it yourself in about thirty seconds. No insider knowledge, no test rig, no trust required.

Start here, because this is the control group and it makes the rest legible.

FiboSearch describes an *"Ultra-fast search engine based on the inverted index"* and *"Fuzzy search, works even with minor typos"*. Nowhere on its page does the word *semantic* or *embedding* appear. It is a keyword search plugin and it says so.

Relevanssi goes further and publishes its actual method: *"Fuzzy matching: match partial words, if complete words don't match"* and *"tf * idf weighing"*. That is term frequency against inverse document frequency, a technique from the 1970s that still works extremely well. Relevanssi tells you this on the plugin page. Remember that sentence, it comes back later and it is the most useful thing in this post.

SearchWP offers a *"Customizable Algorithm"* and a *"Smart Search Algorithm"* that *"auto-improve[s] based on user behavior"*. No AI, no vectors, no embeddings claimed anywhere.

These three are the honest baseline. If your shoppers search the way people searched in 2015, typing one or two nouns, these plugins are genuinely good and you can stop reading. The reason to look further is if your shoppers have started typing sentences, which is a separate question I will not pretend to settle here.

Softlinkster Semantic Search is the clearest example I found, and I want to be fair to it: they do not hide anything. The plugin is called *Semantic Search*. Its own feature list says *"BM25 Ranking, Industry-standard text relevance scoring with TF-IDF weighting"*.

BM25 and TF-IDF are keyword ranking. Good keyword ranking, genuinely, and the plugin is free, self-hosted, needs no external service, and handles SKUs, attributes and Woo variations. If that is what you want, it is a reasonable pick.

But put it next to Relevanssi for a second:

Relevanssi                     method: tf * idf        name says: nothing about AI
Softlinkster Semantic Search   method: BM25 + TF-IDF   name says: Semantic

Same family of algorithm. Opposite naming.

AI Vector Search Semantic takes this further, and this is the finding I would most want a store owner to know. The name contains *AI*, *Vector* and *Semantic*. Its default mode, the one you get when you install it, is called Lite, and Lite runs on TF-IDF. Not vectors. Vector search is the paid path, and I will come back to what that path costs in a moment.

Three plugins. Same underlying technique. One of them names itself after it, one names itself around it, one names itself after the thing it is not doing by default.

This is the group where the fine print costs actual money, and it is the least visible difference of all, because it never shows up in a pricing table.

Several plugins do use real vector embeddings. What their pricing pages do not always foreground is who pays for generating those embeddings.

Mori AI Search (v2.1.1) is self-hosted and requires your own OpenAI API key. So the real monthly cost is the subscription plus whatever OpenAI bills you, which depends on your catalogue size and how often you resync. Its focus is knowledge bases and documentation, with WooCommerce on the roadmap rather than shipped, and the interface is built around chat.

FindAstra Pro wants either your OpenAI key or a Hugging Face token. Pricing for Pro is not published on the page.

AI Vector Search Semantic, to finish the thought from above, reaches actual vector search through a paid self-hosted Supabase instance plus your own OpenAI key. Two external accounts, two bills, before the first search runs.

And then there is the one that surprised me most.

FindAstra's free tier avoids all of that with what it calls a Local engine. No external API, no accounts, nothing leaves your server. Here is how they describe it, in their own words:

> "The first time the Local engine runs (when you index products in wp-admin, or when a shopper performs a search), the browser downloads a roughly 33 MB open-source AI model from the Hugging Face model hub." "After the download, all search runs entirely in the browser."

Note *when a shopper performs a search*. That is not just your admin machine. That is a legitimate design and the privacy story is real. It is also a cost that has been moved rather than removed, and it has been moved onto your customers, including the ones on mobile data. Nobody would think to ask about this, which is exactly why it belongs in a post about fine print.

wp-search.ai claims *"Semantic Search, not only Keyword Matching"* and lists *"unlimited embeddings"* as a future premium feature. It is currently free, in its own words *"because we want real users to help shape the product"*. The page does not state where embeddings are computed or whether you supply a key, so I cannot place it in either group honestly. Early product, worth watching, not yet possible to read the fine print because there is not much of it yet.

Queryra is mine, so treat this section as the disclosure it is.

The whole reason I ended up reading everyone's fine print is that "bring your own API key" kept showing up, and I wanted to know whether that was the category norm or a choice. Having read all of it: four of these plugins pass the embedding cost or the infrastructure to the customer, by four different routes. Your OpenAI key. Your Hugging Face token. Your Supabase instance. Your visitors' bandwidth.

Queryra is one price, all in. No OpenAI account, no API key, no second bill that scales with your catalogue. Current plans are on the pricing page and I am deliberately not quoting numbers here, because prices in blog posts go stale and then get quoted back at you for years.

That is the honest pitch and it is also the honest limitation: a hosted service is a hosted service. If your requirement is that nothing leaves your server, Softlinkster or FindAstra's local mode answer that and Queryra does not.

Everything above is a plugin you install from a catalogue. There is a separate shelf above it, hosted search platforms like Algolia, Klevu, Searchspring and Hawksearch, and I left them out on purpose.

They are a different purchase: you are buying a search platform and integrating it, not installing a plugin and configuring it. Their pricing is quoted, often negotiated, and the semantic layer is frequently an add-on rather than the base product. If you are at the scale where that shelf makes sense, the fine print you need to read is a contract, not a plugin page. Different post.

You do not need this post. You need the habit it describes. Before installing any search plugin that mentions AI:

1. Open the plugin page and search it for "TF-IDF", "BM25" and "inverted index". If those appear, it is keyword search. That may be fine. It is just not what the name suggests.
2. Search the same page for "API key", "OpenAI", "Supabase" and "Hugging Face". Each hit is a second bill or a second account that the pricing table did not mention.
3. Look for what the free tier does differently from the paid one. "Lite mode" and "local engine" are where the real trade lives, and it is almost never in the comparison chart.

Three searches on one page. That is the entire method behind this post.

I did not benchmark these plugins against each other this time and I am not going to pretend otherwise. Everything above is what these tools say about themselves, which is a different and in some ways more reliable thing, because a vendor's own documentation is the one place they have to be accurate.

Some of these are very small and very new. I am not writing about them as rivals, because they are not. I am writing about them because they happen to illustrate the difference better than the big names do, and because a store owner comparing options will meet exactly this spectrum.

If you spot something I read wrong, tell me and I will correct it here. The whole value of this format is that it is checkable.

*I build Queryra, an AI search plugin for WooCommerce and WordPress. Every claim above about another plugin is quoted from that plugin's own public page and was verified on 3 August 2026. Plugin pages change, so check them yourself before you buy.*