ChatGPT, Perplexity and Google AI Overviews: how each one picks local businesses

Owners hear ChatGPT, Perplexity and Google AI Overviews described as one machine. They are three products, with three published notes on how a page is fetched and when a link can appear. This article stays inside those notes, plus one logged-out ChatGPT run on October 18, 2026. It does not promise that any of them will name your practice.

The checklist that follows from the notes is how to get your med spa recommended by ChatGPT. The local snapshot is what ChatGPT said about Austin, Dallas and Miami.

What ChatGPT search can and cannot be told

OpenAI’s crawler documentation splits three robots that get lumped together as “the ChatGPT bot.”

OAI-SearchBot is the search crawler. OpenAI says it surfaces websites in ChatGPT’s search features. A site that opts out of OAI-SearchBot will not be shown in ChatGPT search answers, though it can still appear as a navigational link. The page recommends allowing OAI-SearchBot in robots.txt, and allowing the published IP ranges, if you want the site eligible for those results. A robots.txt change can take about 24 hours. Eligible means the crawler is allowed in. It does not mean a med spa question will name you.

GPTBot is independent. OpenAI says it crawls pages that may be used to train foundation models. Disallowing it is a training choice, not the search switch. You can allow search and refuse training.

ChatGPT-User is a person-triggered visit. OpenAI says it is not an automatic crawl, that robots.txt may not apply because a person asked, and that this token is not how Search inclusion is decided. A log line for ChatGPT-User is not proof that you opted into search, and blocking it is not the same as blocking OAI-SearchBot.

On October 18, 2026 one logged-out ChatGPT, asked once per city, named med spas and attached source chips. Readable chips included yelp.com on the Austin and Dallas listings and google.com on the Miami answer. Other chips were truncated. Dallas mentioned a Dallas Observer readers’ choice award in the prose. That is one run’s sources, not a directory list or a review minimum from OpenAI.

What Perplexity publishes

Perplexity’s crawler documentation says PerplexityBot surfaces and links websites in Perplexity search results, and that it is not used to train foundation models. They recommend allowing PerplexityBot and their published IP ranges. A robots.txt change may take up to 24 hours.

Perplexity-User is the other token. It fetches a page to answer a person’s question and may link that page. The docs say this fetcher generally ignores robots.txt because a person requested the fetch. A firewall rule is what actually stops or allows that visit. A robots.txt disallow does not, on Perplexity’s page, stop it.

If you run a web application firewall, Perplexity tells you to allow the bot by user agent and by IP together, and to refresh the ranges from its JSON files because they change. A block left over from a redesign keeps new service pages out of the index PerplexityBot builds.

We did not ask Perplexity the three city questions for this article. We will not guess which spas it would have named.

How Google describes AI Overviews and AI Mode

Google’s AI features page and its longer optimization guide describe AI Overviews and AI Mode as part of Google Search. The guide says they are rooted in core Search ranking and quality systems. The method it names is retrieval augmented generation: ranking systems retrieve pages from the index, then a model writes from those pages and shows links. It also describes query fan-out. Google’s example is a lawn question that becomes several related searches, about herbicides, non-chemical removal and prevention, so the answer can use a wider set of pages. A med spa question can fan out into an injector, a treatment, aftercare or a neighborhood without you publishing a doorway page for every wording. The guide says manufacturing those pages to chase variants can violate its scaled content abuse policy.

The AI features page says the two features can show different links because they can use different models. AI Overviews often do not appear, because they are for queries where the system decides an overview adds something. A search for your exact practice name may never show one. That is the product as described, not a penalty with a switch.

Eligibility on that page is the normal Search bar: the page is indexed and can show a snippet. There are no additional technical requirements. Meeting them still does not mean Google will crawl, index or serve the URL. The longer guide adds that a site must be included in Search generative AI features in Search Console to be eligible for those features. Treat that as a Search Console check. If the two pages sound different, use each page’s wording and your own console, not a blended slogan.

Crawl control, the AI features page says, is robots.txt for Googlebot. To limit what Search shows from a page, Google lists nosnippet, data-nosnippet, max-snippet and noindex. The same page points to Google-Extended for training and grounding in some of Google’s other systems. It does not say Google-Extended is the AI Overviews switch. This article stops at that sentence.

For local facts, both pages talk about the Business Profile. The optimization guide says a Google Business Profile can help a business be visible in AI responses and in other Search results. The AI features page lists an up to date Business Profile with crawl access, internal links, text rather than text trapped in images, and structured data that matches the visible page. None of those lines promises that you will be included.

What Google says will not move the answer

The optimization guide says Google Search, including generative features, does not use llms.txt or other special AI text files. Creating llms.txt neither helps nor harms visibility in Google Search, because Google Search ignores it. Structured data is not required for generative AI search, and there is no special schema.org type you must add. Structured data still matters for ordinary rich results, which is a different surface. The AI features page says the same thing briefly: no new machine-readable file, and no special schema, for AI Overviews or AI Mode.

The guide also says you do not need to slice pages into tiny chunks or rewrite them in a special dialect. Inauthentic mentions are less helpful than they look. A real local award, like the Dallas Observer sentence in our Dallas capture, is a different object. One capture does not prove that awards cause inclusion.

Google points measurement at Search Console, including a Generative AI performance report, and says no third-party tool has its internal systems. An outside score is someone else’s product.

The owner’s side of the line

You can change access. Allow Googlebot, OAI-SearchBot and PerplexityBot if you want those indexes able to read the site. Decide GPTBot on OpenAI’s page, because training is a different switch. Expect ChatGPT-User and Perplexity-User to follow the person-triggered rules those docs describe, which include ignoring robots.txt.

You can change the record patients already see: the Google Business Profile, Yelp, the name and phone and address on the site, service pages written as text, and reviews from real visits. Our October 18 screenshots pointed at that layer. They did not point at a secret file.

You cannot dictate the sentence a model writes tomorrow, buy a stable mention, or force an AI Overview to trigger. You also cannot turn llms.txt or a new schema type into Google distribution, because Google says it does not use them that way. Pixelmosa keeps an llms.txt as a plain summary for tools that want one. The file is housekeeping. It is not a claim.

Free AI visibility check

The check shows one current answer for your med spa in your city, and the sources it pointed to. We will not tell you the answer is stable, and we will not promise a ranking.