Why ChatGPT Never Mentions Your Business (And What Actually Changes That)

An accountant in the United States publishes unglamorous articles on her firm's blog: sales-tax nexus rules in California, payroll best practice, the kind of writing nobody reads for pleasure. In August 2025 her husband described what had started happening. Qualified leads were arriving because language models were citing those articles when someone asked a question like “What is the sales tax nexus policy in California?” As he told it, a reader follows the citation, lands on the firm's site, and arrives already half-convinced.

Eleven months later, on the same forum, an independent publisher who has run a free Berlin city guide since 2017 wrote something else: “It is locally famous and beloved… Now traffic is down 75% from two years ago. LLMs and AI overviews are wrecking me.”

Same technology, opposite outcomes. The difference is not luck, and it is not a growth hack. AI assistants do not keep a private list of good businesses. They answer by retrieving live web pages through crawlers they name publicly and a search index, then quoting whatever is easiest to lift as a clean, attributable answer. So if ChatGPT never mentions you, it is usually one of three things: the search-facing crawler cannot get in, your pages hold no liftable answer to the question your customer actually typed, or somebody else answered it more cleanly. What follows is the mechanism, the handful of moves that genuinely affect it, how to check your own results this afternoon, and two popular fixes that stopped working.

What actually happens when someone asks an AI about your industry?

The assistant does not search its memory. It searches the web, roughly the way you would, and then writes a summary with citations attached.

Google is unusually direct about this. Its guidance for site owners states there are “no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary”. Your content simply has to be indexed and eligible to appear in ordinary Google Search. There is no separate AI index to buy your way into. Google does describe one wrinkle worth knowing: AI features use “query fan-out,” issuing several related searches across subtopics at once, which is why an AI answer often cites pages you would never have surfaced with the original phrasing.

ChatGPT, Claude and Perplexity each run their own crawlers, named publicly in their documentation. That distinction is where most of the confusion, and most of the bad advice, lives.

Why “should I block the AI bots?” is the wrong question

OpenAI, Anthropic and Perplexity each run several separately named crawlers, doing different jobs. Blocking the wrong one costs you the thing you were trying to protect.

• Training crawlers collect material for building models. OpenAI’s GPTBot is one; Anthropic’s ClaudeBot is another. OpenAI puts it plainly: “Disallowing GPTBot indicates a site’s content should not be used in training generative AI foundation models.” OpenAI presents that as a training decision, not a search one.

• Search crawlers are the ones that determine whether you can appear at all. OpenAI’s OAI-SearchBot “is used to surface websites in search results in ChatGPT’s search features,” and the documentation is blunt about the consequence: “Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers.” Anthropic runs Claude-SearchBot, which it says “navigates the web to improve search result quality for users”; Perplexity runs PerplexityBot, which it says “is not used to crawl content for AI foundation models.”

• User-triggered fetchers visit a page because a person asked a question that needs it. ChatGPT-User, Claude-User and Perplexity-User sit here. This is the uncomfortable category: OpenAI notes that because these actions are user-initiated, “robots.txt rules may not apply,” and Perplexity states outright that Perplexity-User “generally ignores robots.txt rules.”

So the widely repeated advice that blocking GPTBot stops AI from reading your site is wrong twice over. It does not remove you from ChatGPT’s answers, because that is a different crawler, and it does not reliably stop a user-triggered visit either.

One related correction, because it trips up owners constantly: robots.txt is an instruction about crawling, not a way to hide. Google says so explicitly. A page disallowed in robots.txt “can still appear in search results, but the search result won’t have a description.”

Does FAQ schema still get me those dropdown answers in Google?

No. It stopped in May 2026, and a great deal of the advice still being sold has not caught up.

The timeline is in Google’s own documentation changelog. In 2023 Google restricted the FAQ rich result to “well-known, authoritative government and health websites.” On 8 May 2026 it added a deprecation notice: “This feature will no longer appear in Google Search starting May 7, 2026.” On 15 June 2026 it removed the documentation entirely. FAQPage no longer appears in Google’s gallery of supported structured data.

Keep writing the FAQ content. A page that answers the questions a customer genuinely asks is exactly the kind of page that gets quoted, by search engines and assistants alike. Just do not expect a dropdown in Google, and do not pay anyone who promises you one.

The structured data still worth your attention is Organization. Google says it helps “disambiguate your organization in search results,” using properties like name, url, logo, sameAs, address and telephone. That is a modest, documented benefit: making it unambiguous which business you are. Worth doing once, per Google’s spec, then forgetting about. One caution about how it is often sold: Google describes structured data as helping its own search layer understand a page. None of the crawler documentation quoted in this article mentions structured data at all, so “schema gets you cited by ChatGPT” is an unproven claim, not a documented one.

Do I need an llms.txt file?

Almost certainly not, and the companies who would have to read it do not say they do.

llms.txt is a proposal published in September 2024 by Jeremy Howard, intended to give language models a compact guide to a site. Its own page describes the specification as “open for community input.” It is a proposal, not a standard. Google’s John Mueller, quoted in June 2026, was unsentimental: “it’s purely speculative for now (the file has existed for years, yet none of the AI systems use it).” That is one Google employee posting on a forum, not company policy — but nobody has answered him with a counter-example.

There is a quieter piece of evidence too. The crawler documentation from OpenAI, Anthropic and Perplexity is where a file like this would have to be honored, and none of it says their crawlers read an llms.txt on the sites they visit. There is an irony worth noticing: OpenAI and Perplexity each publish an llms.txt for their own developer documentation. They use the format themselves, pointing at their own material. Neither says it reads yours.

Publishing one costs nothing and harms nothing. It is simply not a lever, and any checklist that opens with it is a checklist written from other checklists.

So what actually works?

Four things, none of them clever.

Let the search-facing crawlers in

Open your robots.txt and look for the search agents by name: OAI-SearchBot, Claude-SearchBot, PerplexityBot, Googlebot. If a developer or a plugin blocked “AI bots” in a burst of caution, this is where you find out. You can still disallow the training crawlers if you want to; those are separate lines and separate decisions.

Publish answers, not brochures

Go back to the accountant. What got cited was not her firm’s services page. It was an article that answered one specific question a stranger typed in full sentences. An assistant assembling an answer needs a passage it can lift and attribute without editing. A page about how good your team is offers nothing to lift.

The practical version: write down the ten questions customers ask you before they buy, including the awkward ones about price and timelines, and answer each properly on its own page. That list is usually sitting in your inbox already.

Producing them is its own problem, and a different one from being cited. If that is where you get stuck, Content Marketing When You Have No Time covers the production side: how to get useful pages written when nobody on the team has a spare afternoon.

Make it obvious who wrote it

Google’s guidance on helpful, people-first content uses the framework it calls E-E-A-T: experience, expertise, authoritativeness and trustworthiness. The document is clear about the ranking of those: “Of these aspects, trust is most important.” It then asks questions you can answer today. “Is it self-evident to your visitors who authored your content?” “Do pages carry a byline, where one might be expected?” “Do bylines lead to further information about the author or authors involved, giving background about them and the areas they write about?”

A real name, a real page about that person, and a real company identity behind the site. E-E-A-T is not a ranking dial you can turn, and anyone selling it as one is guessing. It is a description of what a trustworthy page looks like, and it is cheap to satisfy honestly.

If putting your own name on the work makes you uneasy, that is a common and reasonable reaction; A Personal Brand Without Selling Your Soul is about doing it without turning into a content personality.

Get new pages seen quickly

IndexNow is a free protocol for telling search engines a URL changed, instead of waiting to be re-crawled. Its documentation describes the deal: search engines adopting the protocol agree that “submitted URLs will be automatically shared with all other participating search engines.” The participants it names are Amazon, Bing, Naver, Seznam.cz, Yandex and Yep.

Two honest caveats. Google is not named anywhere on that page, so do not count this as a Google lever. And an HTTP 200 response, in IndexNow’s own words, “only indicates that the search engine has received your URL”; submitting a URL “does not guarantee immediate indexing.” If your site runs on WordPress, there are plugins that do the submitting for you.

One more thing worth knowing: Google has no general “push my new page” API. Its Indexing API can only be used to crawl pages with either JobPosting or BroadcastEvent. If somebody offers to submit your blog posts through it, they are selling you something that does not exist.

How do you check whether any of this is working?

For a long time you could not. That changed in February 2026, when Microsoft opened a public preview of the AI Performance report in Bing Webmaster Tools. It shows “the total number of citations that are displayed as sources in AI-generated answers,” and “grounding queries”: the phrases the AI used when retrieving your content. Microsoft says the report covers Copilot, AI-generated summaries in Bing, and “select partner integrations” — its own surfaces, in other words. It is not a window onto Google, and it makes no claim to cover every assistant.

On Google’s side, Search Console now counts AI Mode towards your totals, but does not break it out as a separate line. You cannot see AI Mode clicks on their own.

Then there is the test that costs nothing. Open ChatGPT, or Perplexity, and ask the question a customer would ask, phrased the way they would phrase it, without your brand name anywhere in it. “Who does emergency plumbing in Cartagena on a Sunday?” Read the answer and the citations. That is your actual position, and if a competitor is there instead of you, now you know which page of theirs is doing the work.

The part nobody selling AI visibility wants to say

Being cited is not the same as being visited. The Pew Research Center tracked the real browsing behavior of 900 US adults across 68,879 Google searches in March 2025. Around 18% of those searches produced an AI summary. When one appeared, people clicked a traditional result in 8% of visits, against 15% when no summary appeared. And they clicked a link inside the summary itself in just 1% of visits.

It is one country, one panel, and a snapshot now more than a year old. Read that last number again anyway. Citation is mostly a consideration channel, not a traffic firehose. It shapes whether a stranger has heard of you before they ever search your name.

Nor does ranking well guarantee it. In an analysis of 863,000 keyword SERPs and four million AI Overview URLs, Ahrefs found that only 38% of AI Overview citations came from pages ranking in Google’s top ten. Its authors caveat it carefully — AI Overviews are “probabilistic,” they note, and their own parsing method improved between studies — and Ahrefs sells an AI-visibility product, so weigh it accordingly. But “just rank first” is clearly not the whole mechanism.

Some people think the whole pursuit is overblown. “LLMs aren’t going to shill for your business,” one commenter wrote, with a qualifier worth keeping: “Unless you’re a household name or very large.” They had not seen the needle tip in their own acquisition channels, and would rather see web search work well than optimize for placement in a model’s weights. They may be right about their own business. The honest position is that the work described above is cheap, it is the same work that makes your site legible to ordinary search, and it does not require you to abandon anything else. Treat it as maintenance, not as a channel that will replace your existing ones.

Which is the practical reason to keep this in proportion. If you need customers this quarter rather than this year, the direct routes still outperform it — Finding First Customers When You Have No Marketing Budget is the honest short game; this article is the slow one.

If you would rather not do this by hand

Everything above is doable by an owner with an afternoon and a text editor. The problem is that it is not an afternoon’s work once. It is a small amount of work, repeatedly, for months, which is exactly the kind of task that loses to a busy week.

Full disclosure before anything else: Laspi is built by GradeBuilder S.L., the same company that publishes moinaki and this blog. Treat what follows as a description of a tool we make, not a neutral recommendation.

Laspi is aimed at owners of small local businesses who will not do content work by hand — its own site names restaurants and cafés, salons, gyms and studios, schools and clubs, small shops and local services. The premise is that you talk for about two minutes about what actually happened in the business this week, and it turns that into social posts and blog articles, publishing the articles to your WordPress automatically. It keeps what it calls a rolling memory of your last 60 topics, 30 angles and 30 offers, with older facts fading as the business changes, and it builds a voice profile from your own recordings rather than from a prompt.

The part relevant to this article is what it calls its AI-visibility audit, and the method matches what I would tell you to do manually. It generates the buying questions a real customer would type, deliberately without your brand name in them, asks them the way a customer would, and reports how often you appear and which competitors take your place. You can repeat it every 7, 14 or 30 days and watch the line move. The dashboard also shows which AI bots crawled your pages and when, which is the one signal most owners have no way to see.

On the technical side it does the unglamorous parts inside WordPress rather than handing you a PDF of advice: structured business data, an explicit allowlist for AI crawlers, and a ping to search engines on every publication. Those three map directly onto the crawler, Organization and IndexNow sections above. There is also a free public page in its AI-readable business directory that requires no subscription and no website, which is a reasonable thing to take without paying for anything.

Pricing at the time of writing runs from €19 a month for 15 publications up to €99 for five projects, with the AI-visibility audit included from the €69 tier. Those are promotional prices; the list prices shown beside them are higher. The honest caveat is the one this whole article has been making: no tool, ours included, can promise that ChatGPT will recommend you. What a tool can do is make the checkable things stay done and tell you where you currently stand.

If you would rather do it yourself, do it yourself. The list is short and it is all above.

Where should you start tomorrow?

Open ChatGPT and ask the question your best customer would ask, without your name in it. Whatever comes back is your starting line. Then check your robots.txt for the search crawlers, write the answer to one question you get asked every week, and put your name on it.

That is a real afternoon. The accountant did not do anything cleverer than this; she simply kept doing it, about the boring questions, in public.