How AI Assistants Find and Recommend Businesses (and How to Be Findable)
When someone asks ChatGPT for "a good accountant near me" or asks Claude to "find a photographer for a small wedding," the assistant answers with a shortlist. Some businesses make that shortlist. Most never do. The difference is rarely luck: it comes down to whether the assistant can find, read, and trust your website. This guide explains how AI assistants actually source their recommendations in 2026, then walks you through five checks you can run today.
Where AI recommendations actually come from
There is a common myth that ChatGPT "knows" businesses from its training data and that is the end of it. That was mostly true in 2023. In 2026 it is not. When you ask a modern assistant a question with a local or commercial answer, it usually runs a live web search behind the scenes, reads a handful of pages from the results, and builds its answer from what it just read.
That means AI recommendations flow through a pipeline, and your website has to survive every stage of it:
Fail any stage (not crawled, not indexed, not readable) and you cannot appear in the last box.
The practical takeaway: being recommendable by AI is mostly the same job as being findable in normal search, plus a few new rules about which robots you let in. That is good news. You do not need an "AI SEO" agency. You need five checks.
First, know your robots: training bots vs search bots
In 2026 the major AI companies run separate crawlers for separate jobs, and you can allow or block each one independently in your site's robots.txt file. This is the piece most small business advice gets wrong: blocking "AI bots" as one bucket can silently remove you from the AI answers that send you customers.
The two jobs to keep apart:
Training crawlers (GPTBot from OpenAI, ClaudeBot from Anthropic, CCBot from Common Crawl) collect content to train future models. Blocking them does not affect whether assistants can recommend you today.
Search and retrieval crawlers (OAI-SearchBot for ChatGPT search, Claude-SearchBot for Claude, PerplexityBot for Perplexity) index your site so the assistant can find you, cite you, and link visitors back. Blocking these removes you from AI answers.
| Crawler | Run by | Job | If you want AI referrals |
|---|---|---|---|
| OAI-SearchBot | OpenAI | ChatGPT search results | Allow |
| Claude-SearchBot | Anthropic | Claude web answers | Allow |
| PerplexityBot | Perplexity | Perplexity answers | Allow |
| GPTBot | OpenAI | Model training | Your call |
| ClaudeBot | Anthropic | Model training | Your call |
Whether to allow training bots is a business decision about your content, not a visibility decision. Many publishers block training and allow search. For a small business site whose whole point is being found, the simplest safe posture is: allow everything, or at minimum allow all three search bots.
The 5-check findability audit (about 30 minutes)
1. Check robots.txt is not locking AI search out
Visit yourdomain.com/robots.txt in a browser. Look for lines like User-agent: OAI-SearchBot followed by Disallow: /. That pairing means blocked. No mention of a bot at all means it is allowed, which is fine. We checked our own file at storypress.app/robots.txt while writing this: it contains no Disallow rules for any AI crawler, so all search bots can read the site. That is the state you want.
Verify it worked: paste your robots.txt into any robots tester, or simply confirm there is no Disallow: / under an AI search bot's user-agent.
2. Confirm you are in the classic search indexes
AI assistants lean on Google and Bing results. Search site:yourdomain.com on both. If your key pages (home, services, contact) do not appear, fix indexing first. Our guide to Crawled, currently not indexed covers the most common blocker.
Verify it worked: your homepage and every service page show up in the site: search.
3. Put the facts an assistant needs in plain text
Assistants extract facts literally. If your prices live only inside an image, a PDF, or a booking widget, the assistant cannot quote them and will recommend a competitor whose page says "Haircuts from $35, open Tuesday to Saturday." Every service page should state in normal text: what you do, where you are, who it is for, what it costs (even a range), and how to book.
Verify it worked: select-all and copy your service page into a blank text file. If the essentials are not in the pasted text, the assistant cannot see them either.
4. Keep your business facts consistent everywhere
Assistants cross-check your site against Google Business Profile, Yelp, and directories. Conflicting hours or addresses make you look unreliable, and unreliable answers get dropped. Match your name, address, phone, and hours everywhere. Structured data helps machines confirm these facts; we covered the how-to in Structured Data and Schema.
Verify it worked: search your business name and check the top 5 listings show identical hours and phone number.
5. Test it: ask the assistants yourself
This is the audit nobody runs. Open ChatGPT, Claude, and Perplexity and ask the question your customer would ask: "best [your service] in [your town]". Note who gets recommended and why. Then ask directly: "What does [your business name] offer and what does it cost?" If the assistant answers wrong or draws a blank, the gap is almost always something from checks 1 to 4.
Verify it worked: the assistant describes your business accurately and cites your website as its source.
What helps vs what wastes your time
Worth doing
Allowing AI search crawlers in robots.txt
Plain-text prices, hours, and services on real pages
Consistent name, address, phone across the web
One clear page per service, written for humans
Asking the assistants about your business monthly
Mostly hype
Paying for "AI SEO" packages that just add an llms.txt file
Keyword-stuffing pages "for the AI"
Blocking all bots out of caution, then wondering where AI traffic went
Chasing every new AI-visibility tool
Rewriting your whole site in robot-speak
A note on llms.txt, the much-hyped "menu file for AI": one 2026 study that monitored over 500 million AI bot visits across 90 days found only 408 requests for llms.txt. Google has said it does not use the file at all. It costs little to add, but it is nowhere near as important as the five checks above, so do those first.
“AI assistants do not recommend the best business. They recommend the best business they can read.”
The quiet advantage: a site machines can read
Most of the work above is undoing things that make websites hard for machines: text trapped in images, facts buried in widgets, robots locked out by accident. Some platforms handle this for you. StoryPress ($5/month) renders every page as clean, AI-readable content out of the box and leaves AI search crawlers unblocked by default, so the pipeline in the diagram just works. Whatever platform you use, the five checks are the same, and they are all free.
Run the audit this week. The businesses showing up in AI answers right now are not the biggest ones. They are the ones whose websites an assistant can actually read.
Get one practical website guide like this every week
'Crawled - Currently Not Indexed': What It Really Means
Google visited your page, read it, and decided not to keep it. Here is how to tell which of those pages need fixing, and what actually gets them indexed.