LLM search on your website: add AI search for visitors
Short answer LLM search lets visitors type a plain question into your site's search box and get a short answer written from your own pages, with links. It pays off on sites with hundreds of pages or products. Shopify stores already have a version; others add a search service. Feed it only your pages, and test it before launch.
Want help applying this to your business? Talk it through on a free 30-minute call.
Book a callLLM search on a website: what changes for visitors
An ordinary search box matches words. LLM search (search run by a large language model, the kind of AI behind ChatGPT and Gemini) works out what the visitor means and can write the answer.
| Keyword search | Meaning-based search | AI answer search | |
|---|---|---|---|
| Visitor types | "water heater" | "no hot water in the morning" | "Do you fix tankless heaters on weekends?" |
| What comes back | Pages containing those words | Pages about the problem, even in other words | A two- or three-sentence answer, with links to the pages it used |
| Fails when | Visitors use different words than you | Your pages don't cover the topic | Your pages are wrong, missing or out of date |
Meaning-based search is often called semantic search. Google's own version is in semantic search; this page is about putting it on your site.
Does a small business website need AI search?
Usually not yet. On a site of 10 to 50 pages, a clear menu and a few hub pages get visitors where they're going faster than any search box; site search and content hubs shows how.
It starts to pay when:
- You have hundreds of pages: a help center, a resource library, a long FAQ, past project galleries.
- You sell many products and shoppers describe needs, not product names.
- Your search log shows many searches that return nothing.
If what visitors want is a conversation (price, area served, booking), a chat assistant fits better than search. See AI web components.
How AI site search works
Three steps, often called RAG (retrieval-augmented generation):
- Index. The service reads your pages and stores them in a form it can search by meaning.
- Retrieve. When a visitor asks something, it pulls the few passages that fit best.
- Answer. It hands those passages to a language model, which writes a short reply and points to its sources.
Step 2 is what keeps the answer about your business. Google Cloud explains:
"Providing "facts" to the LLM as part of the input prompt can mitigate "gen AI hallucinations."" (Google Cloud: What is RAG?)
A hallucination is a confident answer the AI made up. Retrieval lowers the risk; it doesn't remove it. The answer can only be as good as your pages.
How to add AI search: your options
Prices as listed on each page on November 16, 2026, in US dollars.
| Option | Fits | What it does | Cost |
|---|---|---|---|
| Shopify store search | Shopify stores | Meaning-based product search, built in | Included; Shopify says you don't need to turn it on |
| Algolia Agent Studio | Help centers and large catalogs, built by a developer | AI answers and an AI search bar on top of Algolia search | Its searches count toward your Algolia plan, and you pay your own AI provider for usage |
| Google Cloud Agent Search | Large sites, built by a developer | Website search with AI answers | 10,000 queries a month free per account, then $4.00 per 1,000 on the edition with website search, plus storage |
| A chat assistant that answers from your pages | Most service businesses | Answers questions in a chat window | See AI web components |
Shopify
Shopify calls its version semantic understanding and says it "uses related words, concepts, categories, and other contexts to improve and expand search results" (Shopify Help Center: Modifying search). It reads product title, description, type, category and tags, not images. So the fix for weak results is better product text: write descriptions in the words shoppers use.
Algolia Agent Studio
Algolia's AI answer tool, Ask AI, now ships inside Agent Studio; Algolia says Ask AI "isn't available as a standalone feature for new applications" (Algolia docs: Ask AI cost optimization). Agent Studio connects an AI model you choose (OpenAI, Anthropic, Google Gemini and others) to your Algolia search (Algolia: Agent Studio). Algolia says "Production requires your own LLM provider" and "Search requests made by agents count toward your Algolia usage" (Algolia docs: Agent Studio LLM providers). That means two bills: Algolia's plan, and the AI provider's usage. Algolia's free plan lists 10,000 search requests a month (Algolia: Pricing).
Google Cloud Agent Search
Google lists $1.50 per 1,000 queries for the Standard edition and $4.00 per 1,000 for the Enterprise edition, which adds website search and "core Generative Answers" at no extra cost. Storage for a website is counted as 500 KiB per page; Google's own example puts a 1,000-page site at $2.38 a month (Google Cloud: Agent Search pricing). A developer sets it up and maintains it.
How to keep AI answers accurate
You own what the search box says. In 2024 a British Columbia tribunal held Air Canada responsible for wrong fare advice from the chatbot on its website, rejecting the idea that the bot was separate from the airline (Moffatt v. Air Canada, 2024 BCCRT 149).
OWASP, the security group whose AI risk list developers follow, describes the root cause:
"Hallucinations occur when LLMs fill gaps in their training data using statistical patterns" (OWASP: LLM09:2025 Misinformation)
Before launch:
- Feed it only current pages. Remove old offers, discontinued services and draft pages from what it indexes.
- Turn on source links. Every answer should link the page it came from, so visitors can check.
- Fence off risky topics. Prices for custom work, legal, medical or tax questions: show the relevant page and a contact button, not a written answer.
- Test 20 real questions. Take them from your inbox, phone calls and current search log. Fix the page, not just the answer, when one comes back wrong.
- Read the log monthly. Questions with no good answer are pages you haven't written yet.
Example
Illustrative cases. Volumes and prices are invented for the example.
A lighting store on Shopify, 1,200 products
The owner rewrites the top 100 product descriptions in customers' words ("warm light for a reading nook"). Built-in search starts matching those phrases. Cost: her time.
An HVAC company with 40 pages
Its search log shows 30 searches a month. It skips AI search, adds a "Repairs" hub page, and puts a chat assistant on the site for after-hours questions.
A software firm with a 600-article help center
A developer sets up Google Cloud Agent Search. At 20,000 searches a month, the first 10,000 are free and the rest cost about $40; storage for 600 pages is under $2 a month. The developer's setup fee is the main cost.
FAQ
Is LLM search the same as ranking in ChatGPT?
No. This page is about a search box on your own site. Getting your business named in ChatGPT or Gemini answers is a different job; see LLM search optimization.
Will AI search slow down my website?
It adds a script, like any widget. Load it only when someone clicks the search box, and compare page speed before and after.
Can AI search answer from PDFs?
Many services can index documents as well as web pages. Check the vendor's list of supported file types, and remove outdated PDFs first.
Do visitors trust AI answers on a business site?
They trust answers they can check. Show the source link under every answer and a clear way to reach a person.
Different for you?
- Your site has fewer than about 50 pages: site search and content hubs.
- You want a chat assistant, not a search box: AI web components.
- You want to know if chat brings leads, and the bot disclosure rules: live chat for a website.
- A clinic or health practice handling patient questions: is ChatGPT HIPAA compliant.
- A law firm answering legal questions online: law firm AI chatbot ethics.
- You'd rather have site search set up and kept accurate for you: Website Agent.