How AI assistants choose which businesses to recommend
Short answer
For questions about products, services and local businesses, AI assistants usually search the web first, read a small number of pages, and recommend the businesses those pages describe most clearly and consistently. A business is more likely to be named when its own site is readable by AI crawlers and answers the question directly, and when independent sites such as reviews, directories and articles say the same things about it.
Two sources: live search and memory
When you ask an AI assistant “Who are the best wedding photographers in Jaipur?”, it can answer from two places.
- Live web search. The assistant writes one or more search queries, reads several of the results, and builds its answer from them, usually with links to the pages it used. For commercial and local questions this is now the normal route in ChatGPT, Gemini, Perplexity, Claude and Copilot.
- Training data. Everything the model learned before its cut-off date. Well-known brands are “remembered”; small and new businesses usually are not. This layer changes only with new model versions, so you cannot influence it quickly.
For most businesses, the live search layer is where the opportunity is.
Which search each assistant relies on
| Assistant | Where its web results come from | Crawler to allow |
|---|---|---|
| ChatGPT search | OpenAI’s own crawler and search partners | OAI-SearchBot |
| Google AI Overviews, AI Mode, Gemini | Google Search | Googlebot |
| Microsoft Copilot | Bing | Bingbot |
| Perplexity | Perplexity’s own index | PerplexityBot |
| Claude | Web search, plus Anthropic’s search crawler | Claude-SearchBot |
Two practical conclusions follow. First, if you are not indexed by Google and Bing, several assistants cannot find you at all. Second, each AI company’s own search crawler must be allowed through robots.txt and your firewall. The AI crawlers guide lists them.
What makes a page get picked
Once the assistant has search results, it chooses which pages to read and quote. The pattern across assistants is consistent: they prefer pages that answer the exact question quickly and specifically.
- A direct answer near the top. “We are a wedding photography studio in Jaipur covering destination weddings across Rajasthan, from ₹1.5 lakh” is easy to use. Three paragraphs about passion and quality are not.
- Concrete facts. Areas served, prices or price ranges, years in business, credentials, numbers of projects, turnaround times.
- Coverage of follow-up questions. What is included, how long it takes, how to book, what happens if something goes wrong.
- Evidence. Named clients, case studies, reviews, photos of real work.
- Freshness. Current prices and dates. Old pages that look abandoned are used less.
Why other websites matter as much as yours
For “best” and “top” questions, assistants rarely rely on a business’s own claims. They lean on sources that compare businesses: review platforms, directories, local news, industry associations, “top 10” articles and forums. When you look at the links in an AI answer, these are often most of them.
So part of AEO is simply finding out which third-party pages the assistants cite for your questions, and then earning a fair place on them: claiming directory listings, asking happy customers for reviews on the platforms that get cited, and offering useful information to the writers of comparison articles.
Consistency builds confidence
Assistants combine information from many pages. If your site says you open at 9, Google says 10, and a directory lists an old phone number, the assistant may repeat any of them, or avoid naming you. Keep these identical everywhere:
- business name (one spelling),
- address and service areas,
- phone number and email,
- opening hours and prices,
- a one-sentence description of what you do.
Structured data (Organization or LocalBusiness markup with sameAs links to your official profiles) helps machines connect these sources. See Structured data for AI search.
What does not work
- Keyword stuffing or hidden text. Assistants read pages the way people do; padding makes a page less useful to quote.
- Fake reviews or seeded mentions. Google says its systems are built to ignore inauthentic mentions, and review platforms remove fake reviews.
- Pages written only for AI. Google’s 2026 guidance says there is no special way of writing for AI search; helpful pages for people are what get used.
- Special files as a shortcut. Files such as llms.txt are optional and are not a ranking signal in Google. See llms.txt explained.
Questions
Does ChatGPT use Google to find businesses?
ChatGPT search uses OpenAI’s own crawler, OAI-SearchBot, together with search partners. Google’s AI Overviews, AI Mode and Gemini draw on Google Search. Microsoft Copilot draws on Bing. Being well indexed in both Google and Bing therefore covers most AI assistants.
Why does the AI recommend my competitor and not me?
Usually because the pages the assistant read describe your competitor more clearly: their own page answers the question directly, or they appear on the directories and “best of” lists the assistant cited. Comparing the cited pages with yours shows the gap.
Do reviews affect AI recommendations?
Often, yes. Assistants frequently cite review platforms and quote ratings when recommending local businesses and services. A strong, recent, genuine review profile on the platforms your customers use helps.
Why do I get a different answer every time I ask?
AI assistants generate text with some randomness, and the search results they read can differ between runs. That is why one test proves little; ask the same question several times and look at how often you are named.