How AI assistants pick which businesses to name
ChatGPT, Perplexity, Google's AI answers and Copilot use four different pipelines to decide who gets recommended. An intervention that works on one can do nothing on another, which is why they have to be measured separately.
AI visibility is usually discussed as though the assistants are one thing. They are not. There are four distinct retrieval pipelines behind the major assistants, and they disagree about who to recommend because they are reading different sources through different indexes.
This matters commercially rather than academically. If you are told an intervention improves your AI visibility, the first question is which pipeline it acts on, because several popular ones act on none of them.
ChatGPT: Bing is the gatekeeper
- Retrieval is RAG over the live web, but URL discovery runs through the Bing Search API. Microsoft handles indexing, ranking and serving of candidate URLs.
- That makes Bing indexation a hard gate. If you are not in Bing, ChatGPT structurally cannot cite you.
- There are three separate crawlers, and they are not interchangeable. GPTBot builds the training corpus, OAI-SearchBot builds the search index, and ChatGPT-User fetches pages live during a conversation. Blocking OAI-SearchBot removes you from citation eligibility; blocking GPTBot does not.
- A separate, opaque training-data layer also influences answers. Wikipedia dominates ChatGPT's citation share, reported between 26% and 48% of top-ten citations.
OpenAI has content licensing deals with selected media partners whose material may be preferentially surfaced. That is a lever no consultant can pull on your behalf, and it is worth knowing exists when someone explains your absence purely as a fixable technical problem.
Perplexity: its own index, and very heavy on Reddit
- Perplexity runs its own index, independent of Bing. PerplexityBot indexes; Perplexity-User fetches during a conversation.
- It respects robots.txt, but a blocked page may still have its domain, headline and a factual summary indexed.
- Reddit is reported at roughly 46.7% of Perplexity's top-ten citation share — by far the most concentrated dependency on a single source of any assistant.
Google AI Overviews, AI Mode and Gemini: grounded in Search
- These are grounded in Google's core Search ranking and quality systems, in Google's own words. A page must be indexed and snippet-eligible to appear at all.
- The link to classic rankings is weakening. The share of cited URLs drawn from Google's top ten fell from roughly 76% in mid-2025 to roughly 38% in 2026. Ranking first no longer guarantees citation, and not ranking in the top ten no longer excludes you.
- Google explicitly names Google Business Profile and Merchant Center as inputs for local and ecommerce answers. That is the clearest documented lever available to a local business.
AI Overviews and Gemini are commonly treated as the same product. They are not, and they routinely name different businesses for the same question. AI Overviews is the answer at the top of an ordinary Google results page, which most people encounter without ever deciding to use AI. Gemini is a separate assistant you open on purpose. Both are Google, both run on Gemini models, and they need to be measured separately or the result is an average of two different populations.
Claude: the least documented of the five
Claude is the surface with the least public detail about how retrieval works, and it is worth saying that plainly rather than filling the gap with confident description. Anthropic documents ClaudeBot as its crawler and respects robots.txt, but does not publish the equivalent of Google's grounding statement or OpenAI's crawler breakdown.
It is widely reported in this industry that Claude's web search is served by a third-party index rather than one Anthropic operates itself. We have not been able to verify that from provider documentation, so we are not going to assert it.
What we can report is our own measurement. Across 29 answers, Claude cited 66 distinct domains — more than any other assistant and three times ChatGPT's 22. Its overlap with the others was small in every direction: 7 shared domains with Google AI Overviews, 6 with Perplexity, 5 with Gemini and 2 with ChatGPT. Whatever it is reading, it is not the same pool as the rest.
This matters if you are being sold ChatGPT-focused work. Bing indexation is genuinely a gate for ChatGPT and Copilot, and on our data it does very little to explain Claude, which shared only two source domains with ChatGPT out of 86 between them.
Copilot: Bing again
Copilot is Bing-index native, which makes its discovery gate effectively identical to ChatGPT's. Two of the five surfaces most people care about therefore share a single point of failure.
What is actually in the loop
Across engines, Reddit is the single most-cited domain, appearing at roughly 40% frequency and ranking first on every major engine. Wikipedia, YouTube, LinkedIn and Forbes follow. The top fifteen domains together capture roughly 68% of all citations across ChatGPT, Claude, Gemini, Perplexity and AI Overviews.
The counterweight is that no single domain exceeds about 5% of any platform's total. Concentration at the top does not mean a handful of pages decide everything, and for local businesses the sources that matter most are Google Business Profile, Yelp and category directories, where inconsistent name, address and phone details directly degrade what the assistants say about you.
The base rate nobody quotes
SOCi's 2026 Local Visibility Index, covering more than 350,000 locations across brands with 50 or more sites, found ChatGPT recommends only 1.2% of brand locations, against 35.9% appearing in Google's local three-pack. AI local discovery is roughly thirty times more selective than traditional local search.
That number cuts both ways, and any provider quoting it only in one direction is selling. The opportunity is real, because almost nobody is there yet. The base rate of success is also low, and you should hear that before you sign anything.
What follows from all this
- Measure each assistant separately. Averaging them hides the pipeline that is actually broken for you.
- Check the gates first: Bing indexation for ChatGPT and Copilot, crawler access for Perplexity, indexing and Google Business Profile for Google's surfaces.
- Treat third-party sources as the main event rather than a footnote, because that is where the majority of citations come from.
- Ask any provider which pipeline a proposed intervention acts on. If the answer covers all of them equally, it is probably acting on none of them.
Common questions
- Why does ChatGPT recommend my competitor but not me?
- The most common structural reason is discovery. ChatGPT's URL discovery runs through the Bing Search API, so a business absent from Bing's index cannot be cited regardless of how good its website is. After that, the usual causes are blocked crawlers, client-side rendering, and an absence from the third-party sources the assistants read.
- Are Gemini and Google AI Overviews the same thing?
- No. AI Overviews appear at the top of an ordinary Google results page, which most people see without choosing to use AI. Gemini is a separate assistant. Both are Google, both run on Gemini models, and they routinely name different businesses for the same question, so they have to be measured separately.
- How often does ChatGPT actually recommend a local business?
- Rarely. A study of more than 350,000 locations found ChatGPT recommends about 1.2% of brand locations, against 35.9% appearing in Google's local three-pack. AI local discovery is roughly thirty times more selective than traditional local search.
Sources
Every claim above should be checkable. Where a study has limits, they are stated rather than left out.
- Google Search Central — AI features grounded in core ranking systemsGrade A. Provider documentation.
- Perplexity Help Center — how Perplexity follows robots.txtGrade A. Provider documentation.
- OpenAI — GPTBot, OAI-SearchBot and ChatGPT-User crawler documentationGrade A. Provider documentation.
- SOCi 2026 Local Visibility Index — 350,000+ locations, 120+ metricsGrade A for the measured base rate.
- Profound — citation share analysis across 680 million citationsGrade B. Large-sample observational.
More on getting cited in ai answers
- Why the five AI assistants disagree about who to recommend
We measured every source five assistants used to answer the same questions. Not one domain was cited by all five, and 89% were cited by only one. Our own original data on why a single AI visibility number is misleading.
- The source ecosystem: how AI actually decides who to name
When we asked five assistants who to hire in our own category, 59 of 108 citations were other people's roundup lists. Five of the twelve most-named companies sat on one directory. Being named is mostly about other people's pages.
- What actually works in AI visibility, graded by evidence
Twenty interventions sold as AI visibility work, each rated by how strong the evidence behind it actually is. Three are gates you must pass. One has a rigorous causal study behind it, and that study found nothing.
- What to fix first for AI search
A running order based on what the evidence supports rather than what is easiest to sell. Three gates before anything else, then off-site work, then your own pages — which come later than almost every agency will tell you.
- Does llms.txt do anything?
Over 844,000 sites have adopted llms.txt. No major AI provider has committed to reading it in production retrieval, and the largest correlation study found no relationship with citations. Here is what it is actually for.
Want to know where you stand?
We ask five AI assistants for a company like yours and send you a free report showing how often you came up and who came up instead.
Get my free report