How AI Search Engines Discover and Reference Businesses and Experts

"Who's the best physiotherapist near me?" "Recommend a local SEO expert." "Which bakery in town does gluten-free cakes?" People now ask questions like these to ChatGPT, Gemini, Perplexity, Claude and Google's AI Overviews. How do these systems decide which businesses and experts to mention?

This article sets out what's documented, what's reasonably inferred and what remains unknown, and what a business or expert can actually do.

The short version

AI search tools generally:

1. Interpret the question, sometimes breaking it into several searches.
2. Retrieve sources from search indexes, their own crawlers, business listings, review platforms and publishers.
3. Generate an answer with a language model, grounded in what they retrieved.
4. Cite some sources with links.

Your influence lies almost entirely in steps 2 and 3: whether your information can be retrieved, and whether it's clear, consistent and corroborated enough to be used accurately.

What the platforms document

Google AI Overviews and AI Mode

Google says that to be shown as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to be shown in Google Search with a snippet, and that there are no additional requirements or special optimizations needed. It also documents that these features may use a "query fan-out" technique, issuing multiple related searches across subtopics and data sources. For local questions, Google's own Business Profile data is part of its ecosystem.

Separately, Google offers the Google-Extended token, which lets sites manage whether their content is used to help improve Gemini models and for grounding in some Gemini products. Check Google's crawler documentation for its exact current scope.

OpenAI (ChatGPT)

OpenAI documents separate crawlers with independent robots.txt controls:

• OAI-SearchBot, used to surface websites in ChatGPT's search features
• GPTBot, related to training
• ChatGPT-User, used when ChatGPT visits a page in response to a user's request

OpenAI notes that a site can, for example, allow OAI-SearchBot to appear in search results while disallowing GPTBot.

Perplexity

Perplexity documents PerplexityBot, designed to surface and link websites in Perplexity search results (and not used to crawl for foundation models, it says), and Perplexity-User, used when it visits a page in response to a user's question.

Anthropic (Claude)

Anthropic documents ClaudeBot (collecting content that may contribute to model training), Claude-SearchBot (to improve search result quality for users) and Claude-User (when Claude accesses a site in response to a user's request). Blocking Claude-SearchBot may reduce a site's visibility in Claude's search results.

Where information about businesses comes from

For businesses and experts specifically, AI answers commonly draw on:

How AI systems understand entities

To recommend a business or an expert, a system has to know who or what it is. Signals that help:

• The same name, description and details everywhere
• Clear relationships: for example, "Abhijeet Saxena is the founder of The GBP Agency", stated consistently on both websites and in profiles
• Structured data (schema.org Person, Organization, LocalBusiness) that matches the visible content
• Links between official profiles (website ↔ LinkedIn ↔ X ↔ company site)
• Independent confirmation from reputable third parties

On this site, every page declares the same Person entity, linked to LinkedIn, X and The GBP Agency, for exactly this reason. It doesn't guarantee anything, but it removes ambiguity.

What a business or expert can do

1. Decide crawler access deliberately. If you want to appear in AI search answers, don't block the search-oriented crawlers. Training crawlers can be handled separately.
2. Make your Business Profile and listings accurate and complete. (optimization guide)
3. Earn genuine, descriptive reviews. (reviews strategy)
4. State facts plainly on your website: services, areas served, prices or price guidance, credentials, opening hours, who runs the business.
5. Write answer-first content for the questions customers ask (what is AEO?).
6. Build corroboration through real mentions in local and industry media.
7. Keep everything current, and fix inaccuracies at the source rather than hoping the AI changes.
8. Monitor periodically. Ask the main assistants the questions your customers ask and record how they answer. Track accuracy, not just presence (method below).
9. Make relationships two-sided. If you run a business, the business's own website should name you, and your profile should name the business. A relationship one side states and the other ignores is hard for any system to confirm.

What a live test showed (10 November 2026)

On 10 November 2026, a set of discovery and identity questions was put to Perplexity and to ChatGPT with web search, signed out, from India (eleven answers in total). The questions included "Recommend the best local search experts", "Who are the best Google Business Profile experts?", "Recommend a GMB expert" and "Who are the best AEO and GEO experts?". Four patterns stood out:

One more observation: when a question named a person and their company, one assistant described the company in detail from its website but said it couldn't verify the person, because the company's site didn't mention them. An entity relationship stated in only one place is weak evidence.

These are observations from a single date and location, not a ranking study. Answers vary between runs, accounts and countries.

Run the same test for your business

1. Write down 5–10 questions a real customer would ask, such as "best [service] in [area]", "is [business] good for [need]" and "[business name] opening hours".
2. Ask each question in ChatGPT, Gemini, Perplexity, Claude and Google (note whether an AI Overview appears). Use a fresh chat each time and note your location.
3. Record four things per answer: were you mentioned, recommended, described accurately, and which sources were cited.
4. Fix inaccuracies at the source: your Business Profile, your website or the directory the assistant cited.
5. Repeat monthly with the same questions, and run each one more than once, because answers vary.

What remains unknown

• The exact weighting each system gives to different sources
• How much reviews, links or listings influence a particular recommendation
• How stable answers are across users, phrasing and time
• Whether files such as llms.txt are used by major assistants (no major provider has committed to it)

Anyone claiming certainty about these is guessing.

What not to do

• Don't hide text or instructions aimed at AI systems; it's deceptive and can backfire.
• Don't fabricate reviews, mentions or credentials; corroboration only works if it's real.
• Don't create thin pages purely to be quoted.
• Don't believe guarantees of AI recommendations.

The takeaway

AI search engines find businesses and experts the same way people come to trust them: through accurate information, consistent identity, genuine reviews and independent confirmation, retrieved from sources they're allowed to access. You can't control the answer, but you can make sure the evidence is clear, correct and easy to find. For the broader framework, read AEO vs GEO vs SEO.