

Site search
AI site search: when the search box should answer, not list
Your site search assumes the visitor already knows your vocabulary. They usually do not: they know what they want, not what you call it. That gap is where most internal search fails, and it is measurable in your own logs today.

Open your analytics and find the internal site search report. Sort it by "no results".
That list is the most honest document your company owns: every line is a visitor who typed something specific and then left without it. A few of those searches are for things you genuinely do not sell. Most are for things you do sell, described in words you never used. That gap between their words and yours is exactly what AI site search closes.
Why traditional search fails on real questions
Conventional site search matches strings. It returns the pages that contain the words you typed, ranked by how often and where those words appear. That works when the visitor already knows your term.
Four patterns break it, and they cover most real searches:
| The visitor types | String matching gives | What they wanted |
|---|---|---|
| "something for a wet room" | Nothing (your site says "bathroom") | The bathroom range |
| "how much is delivery to Denver" | 27 pages containing "delivery" | One number |
| "difference between A and B" | Two product pages, side by side | The difference, stated |
| "who handles invoicing" | Nothing (it sits in a table on the contact page) | A name and an email |
In every one of those four cases the information is already on the site. Search did not fail to find a missing page. It failed to understand a question.
What AI site search changes
Two things, and the second one matters more.
1. It matches meaning rather than characters
"Wet room" finds the bathroom page, because the system compares meaning instead of spelling. That alone recovers a large share of your zero-result searches. You write no new content: only the retrieval method changes.
2. It answers instead of listing
This is the real shift. Instead of handing over ten links and leaving the reading to the visitor, it pulls the relevant passages and writes the answer, with a link to the source.
For "how much is delivery to Denver", twenty-seven pages is a worse outcome than one sentence. Your visitor did not set out to research your delivery policy. They wanted a number.
The one-line difference: traditional search tells you where the answer might be. AI search states the answer and shows its source. On a large site, that is the gap between finishing the task and closing the tab.
How this differs from a chat widget
They read the same content and often run on the same platform, but they are two products doing two jobs.
| Chatbot widget | In-page AI search | |
|---|---|---|
| Where it lives | Floating in the corner | Inside the page, often replacing the search box |
| When it is used | The visitor asks for help | The visitor is already looking for something |
| Intent | Mixed: support, sales, complaints | Navigational and informational |
| Best for | Any site with repeat questions | Large sites where finding things is the problem |
Plenty of sites run both. The distinction that matters: a chat widget waits to be opened, while in-page search sits exactly where someone has already decided to look.
Where it is worth it, and where it is not
Worth it
- Large catalogs. When products differ in ways only a specialist can name, the category tree fails the buyer who knows the use but not the word.
- Documentation and support sites. The answer is nearly always there and nearly always buried. This is the highest-yield case.
- Organisations serving several audiences. Municipalities, schools and associations: hundreds of pages, many audiences, and one question arriving in fifty phrasings.
Not worth it
- When the site is under about twenty pages. The menu is faster. Search of any kind solves a problem a small site does not have.
- When the content genuinely is not there. No retrieval system invents a delivery policy you never wrote. It will tell you the answer is missing, which is useful, but that is a content job rather than a search job.
- When search already works. If your zero-result list is short and few people leave straight after searching, spend the money elsewhere.
How to tell whether you need it, in five minutes
- Open your internal search report. Filter to the zero-result searches from the last month.
- Read twenty of them. Mark each one: content genuinely missing, or content that exists under a different name.
- Count the second pile. If it is more than half, meaning-based search recovers those visitors without you writing a word.
- Check the exit rate after a search. A high rate means people search, do not find, and leave. That is the clearest signal you will get.
The by-product is often worth more than the search upgrade: the "genuinely missing" pile from step 2 is a content backlog written by your customers and ranked by demand. Bot conversations do the same job. Within its first week of conversations, the e-bike retailer Ebike Center learned that a large share of its shoppers were looking for a bike under 2,000 euros. One shop, one first week, and the result varies with the range.
Frequently asked
What is AI site search?+
Search that matches the meaning of a query rather than its characters, and returns a written answer with a source link instead of a list of pages. It lets a visitor ask "something for a wet room" and reach the bathroom range even if that phrase appears nowhere on the site.
How is AI site search different from a chatbot?+
They usually share the same content and platform but occupy different places. A chat widget floats in the corner and waits to be opened, handling mixed intent including support and complaints. In-page AI search sits inside the page, often replacing the search box, and handles visitors who are already looking for something specific.
Does AI site search hurt SEO?+
Not when it is built properly: it loads after the page renders and does not replace your indexed content. Indirectly it helps: the questions it cannot answer form a demand-ranked list of content gaps, which is a better content backlog than most keyword tools produce.
How many pages do I need for it to be worth it?+
Roughly twenty is the floor. Below that, visitors find things in the menu and search of any kind is redundant. The benefit grows with catalog size and with how much specialist vocabulary separates your wording from your customers'.
Revdot's AI site search is built for online stores: it replaces the store's search box and shows matching products, categories and pages under the search, so a shopper can type "e-bike for a hilly commute" without knowing a product name. It is a separate product, included from Starter at 39 € a month. If you want a bot that answers in its own words inside the page, that is the inline bot, included in Pro, Business and Scale and sharing one knowledge base, one quota and one dashboard with the chat widget.
Enter your website address and get your own bot by email in 10 minutes, then ask it a few of your zero-result searches. Curious how the answering works under the hood? Read RAG explained without the jargon.
Read next
More from the blog.
PricingAI chatbot pricing in 2026: flat rate vs per-resolution
Three billing models dominate this category, and they produce very different invoices once the bot starts working. What each model bills for, the arithmetic at three volumes, and how to work out which one suits you.Read article
Buying guideFree AI chatbot: what you actually get, and what you don't
Free tiers are real and often useful. They also end abruptly, always at the same three points. How to tell whether free will carry you.Read article
ExamplesAI chatbot examples: 20 that actually do something
Twenty jobs a bot can genuinely do, grouped by purpose, each with the thing that makes it work and the way it typically fails.Read article
Try it on your site
Your own bot tells you more than any article.
Enter your site address, and Revdash builds a bot from your content and emails you the link in 10 minutes. Prices are public, from 39 € a month.
Revdash reads your site · knowledge base · bot in your inbox
Thanks. Your bot is on its way.
Revdash is building a bot from your site, and the link will be in your inbox in 10 minutes. If it does not show up, check your spam folder or email info@revdot.ai.
The Revdot team answers in person:WhatsAppBook a call+358 45 231 1420eelis@revdot.ai
