The short answer
- We found no page from Google, OpenAI, Perplexity or Microsoft that says how an AI answer picks one local business over another.
- What they do publish is eligibility: Google says a page must be indexed and able to show a snippet, and OpenAI says its search crawler must be allowed.
- The best measurements available test a source that is already inside the model's answer, not how it got there, and almost none test local businesses.
- One preprint looked at 4,776 venues in two Bali markets. Most were never recommended by any of the four assistants it tested.
- Our own count so far: Montrelia was named in 0 of 8 answers to buyer questions on Ask Brave, one engine.
What each company says
| Assistant | What its own pages say | What they do not say |
|---|---|---|
| Google AI Overviews and AI Mode | A page must be indexed and eligible to show in Search with a snippet. There are no additional requirements for these features. They rest on Google's core ranking and quality systems. Keep Business Profile and Merchant Center information up to date. No special markup or AI text files are needed. | How it picks one local business over another in an answer. |
| Google local results | Local results are mainly based on relevance, distance and popularity. There is no way to request or pay for a better local ranking. | The page does not mention AI answers, and Google does not say these factors govern them. |
| ChatGPT search | It ranks results using multiple factors, and placement is not guaranteed. To be eligible, allow its search crawler. It sometimes partners with other search providers and may share location with third-party providers for nearby business listings. Results can be incomplete, outdated or incorrect. | Which providers it uses, or how businesses and local listings are chosen or ranked. |
| Perplexity | Answers carry numbered citations to the original sources. It labels the website a source comes from on some sites. Its search API post describes retrieval followed by reranking, and keeping documents from authoritative domains fresh. That post describes the API, not the consumer app. | Anything about local businesses. |
| Microsoft Copilot and Bing | Copilot relies on the same core crawling, indexing and ranking foundation as Bing. Optimizing for AI does not guarantee grounding or citations. Local businesses can register with Bing Places for Business. Microsoft Advertising says there is no secret strategy for being selected. | How a local business is chosen in a Copilot answer. |
Sources: Google: AI features and your website, Google: AI optimization guide, Google Business Profile: improve your local ranking, OpenAI: ChatGPT search, OpenAI: crawlers, Perplexity: how it works, Perplexity: source labels, Perplexity: search API, Bing: webmaster guidelines, Bing: AI Performance in Webmaster Tools and Microsoft Advertising: AI search answers.
What researchers measured
| Study and status | What it tested | What it does not show |
|---|---|---|
| Aggarwal et al., KDD 2024. Peer reviewed. | Rewriting one source per query with quotations, statistics or citations, then giving the model the top five Google results. Rewritten sources were used more prominently in some tests, and results varied by topic. | Whether content changes affect which sources are retrieved. The authors say they did not evaluate search rankings. No local businesses. |
| C-SEO Bench, NeurIPS 2025. Peer reviewed. We read only the abstract. | Content-optimization methods on question answering and product recommendation. The abstract reports most current methods as largely ineffective. | Anything about local businesses. |
| Liu et al., EMNLP Findings 2023. Peer reviewed. | Whether the sentences in generative search answers are supported by their citations. 51.5% were fully supported. | Why a source was chosen. It shows that a citation does not prove the cited page supports the sentence. |
| Pitenin, August 7, 2026. Preprint. | 4,776 venues in two Bali markets and 2,208 answers from four assistants over seven days. 85.6% of venues were never recommended (the authors' figure). Entry into answers was associated with review volume, an own website, price information and third-party mentions. Star rating had no effect at entry. Foursquare presence had no positive effect. Answers from different assistants overlapped only partly, and reruns two weeks apart looked like reruns on the same day. | Cause. It is observational, covers restaurants and cafes, and covers one region. |
| Ibrahim and Zaki, September 16, 2026. Preprint. | Provider recommendations in local service markets. With search, 64 to 71 percent of recommendations in the domains tested matched a real provider. Without search, the models largely made providers up. | How a real provider is chosen over another real one. |
Sources: GEO: Generative Engine Optimization, C-SEO Bench, Evaluating verifiability in generative search engines, Pitenin, August 2026 and Understanding AI Provider Recommendations in Local Service Markets.
What we see ourselves
We ask five buyer questions on Ask Brave, which needs no account, in a new conversation each time. We have run them twice, on September 28 and October 2, 2026.
- Montrelia was named in 0 of the 8 answers to the four buyer questions (four questions, two runs).
- In the branded question we were named both times, described as a New York studio. That is wrong; we are based in the DC area.
- For the question about which Washington, DC agencies publish prices, 8 of the 9 agencies named in run 1 were named again in run 2. For the question about who to hire for a plumber's website, 1 of the 4 was.
- For the Google Ads question, the same two Virginia agencies were named both times. Both say in writing that the client owns the account.
That is one engine, five questions and two runs. It shows where we started. It does not show why anyone was or was not named, and we have not yet run ChatGPT, Perplexity, Gemini or Google AI Mode. Full answers and the runs where we were not named are on the AI visibility scoreboard.
What holds either way
None of these depends on a mechanism being proven. Each one is useful if the mechanism turns out to be different.
- Be indexable. Google says a page must be indexed and eligible to show a snippet, and OpenAI says to allow its search crawler. Our robots.txt allows all crawlers.
- Keep your Google Business Profile current, and register with Bing Places for Business. Google lists Business Profile information among the things to check, and Bing names Bing Places for local businesses.
- Publish what a buyer asks first: your price, how long it takes, where you work and who you are not for. The Bali preprint links price information and an own website to entering answers. We publish ours because buyers ask, not because a mechanism is proven.
- Ask the assistants your buyer questions yourself, in a clean session. Save each answer with its date, repeat on different days and count who is named. Answers vary, so we report counts, not percentages.
- Do not buy a guaranteed mention. None of the four companies publishes a way to pay for one, and anyone who promises it is promising something they do not control.
We publish an llms.txt file that lists our prices and terms. Google says you do not need one, and whether any engine uses it is not established. It stays in sync with the price pages because it is built from the same data.
What is not known
- How any assistant picks one local business over another. We found no primary source.
- Whether relevance, distance and popularity, which Google describes for local results, also apply inside its AI answers.
- Which local data providers ChatGPT uses. OpenAI names none.
- Whether changing a page affects whether it is retrieved at all. The main measurement skipped retrieval.
- The effect of reviews, ratings, directories or markup inside the assistants.
- How much one assistant's recommendations change from run to run. No company publishes a number.
You will also see claims that one data provider supplies most of ChatGPT's local results, that ChatGPT searches mainly through Bing, or that Copilot relies on specific sites. We found those only on agency blogs, with no primary source. We do not repeat them.
For the difference between SEO and GEO, see GEO vs SEO. For how we work on it, see the GEO page and how we work.
Sources
- Google: AI features and your website, page updated December 10, 2025, read October 2, 2026.
- Google: AI optimization guide, page updated July 10, 2026, read October 2, 2026.
- Google Business Profile: improve your local ranking, no date shown, read October 2, 2026.
- OpenAI: ChatGPT search, read October 2, 2026.
- OpenAI: crawlers, no date shown, read October 2, 2026.
- Perplexity: how it works, modified September 3, 2026, read October 2, 2026.
- Perplexity: source labels, September 9, 2026, read October 2, 2026.
- Perplexity: search API, September 25, 2025, read October 2, 2026.
- Bing: webmaster guidelines, no date shown, read October 2, 2026.
- Bing: AI Performance in Webmaster Tools, February 10, 2026, read October 2, 2026.
- Microsoft Advertising: AI search answers, October 8, 2025, read October 2, 2026.
- Aggarwal et al., GEO, KDD 2024, read October 2, 2026.
- C-SEO Bench, NeurIPS 2025, abstract read October 2, 2026.
- Liu et al., verifiability in generative search engines, EMNLP Findings 2023, read October 2, 2026.
- Pitenin, preprint, August 7, 2026, read October 2, 2026.
- Ibrahim and Zaki, preprint, September 16, 2026, read October 2, 2026.
Questions people ask
Can I pay to be recommended by an AI assistant?
Not for the organic answer, as far as the companies say. Google states there is no way to request or pay for a better local ranking on Google, and OpenAI states that placement in ChatGPT search is not guaranteed. ChatGPT Ads is a separate paid product and is shown as an ad.
If I rank first on Google, will ChatGPT recommend me?
Nobody has published that link. OpenAI says ChatGPT search sometimes partners with other search providers and does not name them, and we found no source tying a Google ranking to a ChatGPT recommendation.
Do I need an llms.txt file?
Google says you do not need new machine-readable files or AI text files to appear in its AI features. We publish one anyway because it costs nothing and stays in sync with our price pages. Whether any engine uses it is not established.
Why does the same question give different answers?
Google says the responses and links its AI features show will vary. In our own Ask Brave runs, 8 of the 9 agencies named for one question came back four days later, and only 1 of the 4 agencies named for another did.
Can anyone guarantee an AI assistant will name my business?
No. We do not promise rankings, leads or return on ad spend, and none of the companies publishes a way to buy a guaranteed mention.
What does Montrelia do about it?
We ask buyer questions of each engine separately in clean sessions, save the full answers with dates and report raw counts. Our own results, including every run where we were not named, are on the AI visibility scoreboard.
Last updated October 2, 2026