Quick Answer: You cannot submit your business to ChatGPT. There is no directory, no form, no paid placement. ChatGPT names businesses it finds in sources it already trusts: places databases, review sites, forums, press, and indexed pages. The work is becoming the thing those sources describe, then waiting for a fresh re-read.
Most people arrive at this question after asking ChatGPT to recommend a business like theirs and watching it name four competitors. The instinct that follows is to look for the place where you sign up. That place does not exist, and searching for it is the biggest waste of time in this category.
What follows is the chain as best it can be reconstructed from what OpenAI publishes, what independent measurement shows, and what turns up repeatedly in audit work. Where the honest answer is that nobody outside these companies knows for certain, this page says so and marks the inference. That distinction matters more here than in most marketing topics, because the confident-sounding claims in this niche are mostly unfalsifiable.
One boundary before we start. This page is about the mechanism and what to do about it. If your question is how to measure whether you are being mentioned, and how to track share of voice against competitors over time, that is a different discipline and it lives on our companion page on brand visibility in AI and ChatGPT.
Why there is no way to submit your business to ChatGPT
Search engines trained everyone to expect a submission step. Google has a Business Profile you claim, Bing has Places, directories have listing forms. ChatGPT has nothing equivalent, and the reason is structural rather than an oversight someone will fix in a future release.
A search engine maintains an index of documents and returns a ranked list of them. Its product is the list, so it needs a way for you to tell it you exist. An assistant writes prose instead, generated from two pools: the statistical knowledge baked into the model during training, and whatever it fetches while answering your specific question. Neither pool has an intake desk. The first is closed by the time the model ships. The second is retrieval over material that already exists on the open web and in licensed data feeds.
So the honest reframing is this. You are not trying to get into ChatGPT. You are trying to get into the sources ChatGPT reads, in a form clear enough that it can repeat what they say about you without hedging. That is slower and less satisfying than a submission link. It is also the only version that survives contact with how the system is built.
What actually happens when someone asks ChatGPT to recommend a business
When a question arrives that the model cannot answer confidently from memory, and current-day business recommendations are exactly that kind of question, ChatGPT reaches outward. OpenAI documents the machinery in its crawler overview, and the useful detail is that the crawlers are separate and do different jobs.
GPTBot is described as crawling content that may be used in training generative AI foundation models. OAI-SearchBot is described in OpenAI's own words as being "used to surface websites in search results in ChatGPT's search features." ChatGPT-User covers user-initiated fetches, and OpenAI notes it "is not used for crawling the web in an automatic fashion." A fourth, OAI-AdsBot, validates the safety of pages submitted as ads. OpenAI states that each setting is independent of the others, which means a site can be visible to ChatGPT search while being withheld from model training.
That separation is worth checking on your own site: a robots.txt written in 2023 to keep AI out may be quietly keeping you out of the surface you now want to be in. OpenAI notes such changes take about 24 hours to register.
There are four distinct routes by which your company name can end up in a generated answer. They move at different speeds, and conflating them is why so much advice in this category contradicts itself.
| Route into the answer | What it is | Can you influence it | Speed |
|---|---|---|---|
| Trained memory | Facts absorbed during training from a web-scale corpus, recalled without any live lookup | Only indirectly, by being widely described before the next training run | Slowest, tied to model release cycles |
| Live web retrieval | ChatGPT search fetching and citing pages at answer time from its own index | Yes, directly: allow OAI-SearchBot and publish quotable answers | Weeks to a few months |
| Places and review records | Structured business data used for location-shaped questions | Yes, directly: claim, correct, and maintain the listings | Often the fastest to shift |
| User-supplied context | The person pasted your link or asked about you by name | Not through optimization, this is demand you already earned | Immediate |
Almost every argument you will read online about ChatGPT visibility is really an argument between people describing different rows of that table as though it were one thing.
Why your own website is rarely the citation
Here is the part business owners find hardest to accept. You can write the best page on the internet about your services and still watch an assistant cite a forum thread instead.
The clearest public measurement of this comes from Ahrefs. Si Quan Ong, publishing in September 2025 from the company's Brand Radar database of 9.6 million ChatGPT queries, ranked the most-cited domains by raw mention count. Reddit led with 847,338 mentions and English Wikipedia followed with 431,710. Amazon, Forbes and Business Insider filled out the top ten. Independent company websites do not feature anywhere near the top of that distribution.
What the citation distribution implies
In the Ahrefs sample, the two leading domains were community and reference sites, not brand-owned properties. This does not say your website is worthless. It says the assistant prefers a source that describes you from the outside over one that describes you in your own words. Your site establishes the facts. Somebody else's page gets quoted.
The owner asking how to get mentioned is usually thinking about their website. The mechanism is telling them to think about their footprint: the review corpus, the local press, the trade association page, the supplier directory, the genuinely useful answer someone left in a community thread. Our deeper treatment of that distribution layer, including how the pre-training corpus differs from real-time retrieval, sits in the research piece on earning citations in the LLM corpus.
An honest caveat on all citation-share figures, including the ones above. They are third-party samples of a moving system, not disclosures from OpenAI, and they move sharply. Reddit's measured share across assistant answers has been reported swinging by large margins inside a matter of weeks as retrieval partnerships and ranking behaviour changed. Use these numbers to understand the shape of the system. Do not build a plan that depends on any single domain staying where it is.
The local layer almost every guide misses
If your business serves a geography, there is a fourth data path that most articles on this keyword never mention, and it is often the one that decides whether you appear at all.
In December 2024, Foursquare publicly announced a partnership with OpenAI covering ChatGPT search, describing its Places API as carrying more than 100 million points of interest across over 200 countries and delivering "relevant real-world results to user queries." That is a first-party statement from one of the two parties involved. It confirms that a licensed places database sits underneath location-shaped questions, alongside whatever the assistant retrieves from the open web.
What that announcement does not establish is how much of any given answer comes from that feed. You will see a figure quoted confidently across dozens of agency blogs claiming that some large majority of ChatGPT's local business names are pulled from Foursquare. We have not been able to trace that number to any first-party source from OpenAI or Foursquare, and we are not going to repeat it as fact. Mark it as folk knowledge until someone produces the measurement.
This maps to Vector 2: Anchor
Entity validation is the work of making every record that describes your business agree with every other record. Name, address, phone format, categories, hours, service list. It is the least interesting item on any GEO plan and, for local businesses, reliably the one that changes machine output fastest, because places records are structured and cheap for a system to trust.
The practical consequence holds even though the exact weighting is unknown. A business with a thin, stale or absent record in the places layer is asking an assistant to describe something it has no reliable data about. Assistants handle that by naming somebody else.
What the research says actually moves your odds
Most advice on this topic is assertion. There is one substantial peer-reviewed study that tested content changes against generative engines and measured the result, and it is worth more than every listicle combined.
Aggarwal and colleagues, in the paper that coined the term generative engine optimization and was accepted to KDD 2024, built a 10,000-query benchmark and tested nine content modifications. The headline: adding direct quotations improved visibility by roughly 41 percent, adding statistics by roughly 32 percent, and adding citations to sources by roughly 27 percent. Keyword stuffing, the reflex that a generation of SEO advice was built on, performed worse than the untouched control.
Read that last finding twice. The tactic most associated with gaming search engines actively hurt performance in generative ones. Evidence formatting beat keyword repetition.
The counterweight comes from Google's own documentation on AI features, which is blunt in a way that is inconvenient for a lot of vendors: "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary." It adds that you do not need to create new machine readable files, AI text files, or markup to appear in these features, and that there is no special schema.org structured data you need to add. That is Google describing Google's surfaces rather than OpenAI's, so do not over-extend it. It remains the clearest first-party rebuttal to the market of AI-specific technical add-ons currently being invoiced to small businesses.
The two findings sit together comfortably. Structure and schema are disambiguation tools, not entry tickets. Evidence, quotable claims and verifiable sourcing are what measurably shift generative visibility. For the academic detail on how models decide what to attribute, our page on how LLMs decide what to cite works through the attribution literature, and where AI engines read on your page covers why position within a document changes extraction odds.
The work, in the order that it pays
Sequenced deliberately. The early items cost nothing but attention and move the fastest routes from the table above. The later ones compound slowly.
Start here
- Read your own robots.txt. Confirm OAI-SearchBot is not disallowed, whether by name or by a blanket rule. This takes two minutes and occasionally explains everything.
- Fix the record everywhere it exists. One canonical version of your name, address and phone number, replicated exactly across your site, Google Business Profile, Apple and Bing listings, Foursquare, and every industry directory you can still log into. Same abbreviations, same suite number, same phone format.
- Get your reviews current, not just numerous. A steady trickle of recent reviews reads as an operating business. A cluster of five-star reviews from four years ago reads as a business that may have closed.
- Write one page per real question. Not per keyword. Take the questions customers genuinely ask before buying, and answer each one so completely that the first paragraph could be lifted out of context and still be accurate with your name attached.
- Add the citation trio to your most important pages. A dated statistic from a named source, a direct quote from a real person, and a link to an authoritative reference. This is the exact profile the GEO study measured as most effective, and it is an afternoon of work.
- Earn descriptions you did not write. Trade association profiles, supplier and partner pages, local press, genuine participation in the communities where your category gets discussed. This is the slowest item and the one that most directly addresses the citation distribution above.
- Publish the boring facts plainly. Service area, what you do and do not do, pricing structure, credentials. Assistants hedge when facts are ambiguous, and hedging reads as another business getting named instead of you.
Notice what is absent from that list. No llms.txt file, no content rewritten "for AI", no schema package sold as an AI requirement. If you want this executed rather than explained, our services page sets out how engagements are structured.
Get your free AI visibility audit
We will run the assistant checks on your business and send you what we find: how ChatGPT, Perplexity, Gemini and Google's AI Overviews currently describe you, which records they appear to be drawing on, and where those records contradict each other. Findings within one business day, no charge and no obligation.
A diagnostic you can run in ten minutes
Before spending money on any of this, find out which problem you actually have. Open ChatGPT and ask three questions the way a customer would phrase them, without naming your company: one asking for a recommendation in your category and city, one asking who the options are, and one about the specific problem you solve. Then match what comes back against the four outcomes below.
Match your result to the diagnosis
- It names competitors and never you. Your third-party footprint is the bottleneck. The record layer and outside descriptions are where the work is, not your website copy.
- It names you, but the details are wrong. You have conflicting records. Somewhere in your listings, an old address, a dead phone number or a retired service is still being read as current. This is the cheapest problem on the list to fix and the most satisfying.
- It says it cannot find the business, or confuses you with a similarly named company. You have an entity problem. The structured places layer either has nothing for you or cannot tell you apart from someone else. Claim and correct the listings first, then add JSON-LD to your homepage as disambiguation.
- It names you accurately. Good. Your question is no longer how to get mentioned; it is how often you are mentioned relative to competitors and whether that share is moving. That is measurement work, and it is covered on our AI brand visibility page.
Run the same three prompts monthly and keep the answers. A dated log beats any tool subscription in the first six months, because it is the only thing that tells you whether your changes worked.
What nobody outside OpenAI can tell you
This section exists because the alternative is pretending. Several things in this field are genuinely unknown outside the companies operating these systems, and a page that blurred that line would be worth less than one that admits it.
OpenAI does not publish the criteria by which ChatGPT selects which retrieved sources to cite. There is no ranking documentation, no equivalent of the decades of accumulated knowledge about Google's behaviour. Every citation-share study, including the Ahrefs work above, is an outside observer sampling outputs and inferring the rules. Those samples are real evidence and also snapshots of a system that changes without announcement.
Two consequences follow. First, treat anyone quoting precise percentages about ChatGPT's internal weighting with suspicion, including anyone who tells you exactly how much of a local answer comes from any single data partner. Second, nobody can guarantee that your business will be mentioned, and a provider who does is either misunderstanding the architecture or counting on you not to check. What can be promised is the work itself and honest reporting of what changed. That is the reasoning behind our own commitment: if your existing domain shows no measurable organic search results after 12 months of work with Formative Digital, we work for free until you see them. Note the shape: continued work, scoped to an existing domain, with success defined in writing before we start.
"The first thing owners ask me is where the submission form is. There isn't one, and the disappointment in that moment is actually the useful part, because it moves the conversation to the real problem. A pattern I run into again and again in audits is that the business is not missing from the internet at all. It is described inconsistently across the handful of records these systems actually read. An old phone number on one directory, a former address on another, a service list that stopped being true two years ago. None of that is glamorous to fix. It is also the part that reliably changes what the machines say about you, and it is usually free."
Matt Griffin, Founder, Formative DigitalHow long it takes, honestly
Different routes move at different speeds, which is why a single timeline answer is always wrong. Listing and review corrections tend to show up fastest. New content has to be crawled and then selected at answer time, which realistically means weeks to a few months. Anything sitting in the model's trained memory rather than its live retrieval will not update until a new model ships, and that schedule belongs to OpenAI.
The clearest cold-start data point in our own files is not an AI-citation figure, and it would be dishonest to present it as one. A brand-new foundation repair company in a major Alberta metro began with a fresh domain, no reviews and no third-party record of any kind, and logged 30 leads within its first three months (internal lead tracking, 2026). That is a lead count from the engagement's own form submissions, not a measurement of assistant mentions. It is relevant only for what it says about sequencing: the record layer had to be built before anything downstream could work. Lead volume varies with trade, market and season, and foundation repair is a high-ticket category where 30 enquiries is a material pipeline.
Set expectations in quarters. Established businesses move faster than new ones, because the work is editing records that already exist rather than creating them from nothing.
Frequently Asked Questions
Can I pay to have ChatGPT recommend my business?
Not into the conversational answer itself. OpenAI documents a separate crawler, OAI-AdsBot, whose stated job is to validate the safety of web pages submitted as ads on ChatGPT, so advertising exists as its own surface. The recommendation an assistant writes in prose is assembled from retrieved sources, and no listing fee inserts a company into it. Anyone selling guaranteed placement inside the answer text is selling something the documented architecture does not offer.
How long does it take before ChatGPT mentions my business?
It depends which route you are moving. Corrections to places and review records tend to surface within weeks. New pages have to be crawled by OAI-SearchBot and then chosen at answer time, which is realistically weeks to a few months. Anything that depends on the model's trained memory rather than live retrieval moves on the release cycle of new models, which is outside your control entirely. Plan in quarters, and treat any specific promised date as a warning sign.
Does my website need schema markup for ChatGPT to mention me?
It helps machines disambiguate you, but it is not an entry requirement. Google states plainly in its AI features documentation that you do not need to create new machine readable files, AI text files, or markup to appear in these features, and that there is no special schema.org structured data you need to add. OpenAI publishes no schema requirement either. Treat JSON-LD as insurance against being confused with a similarly named company, not as a lever that buys mentions.
Why does ChatGPT recommend my competitors instead of me?
Usually because they appear in the sources it reaches for and you do not. That means a claimed and current places record, reviews with enough volume and recency to look alive, and mentions in pages an assistant would retrieve when researching your category. It is rarely a judgement about quality of service. It is a judgement about which companies had a describable, verifiable record available at the moment the answer was written.
Should I block or allow OpenAI's crawlers?
If you want to appear in ChatGPT's search results, allow OAI-SearchBot. OpenAI documents its crawlers separately and states that each setting is independent of the others, so you can permit search inclusion while disallowing GPTBot, which collects content that may train foundation models. Check your robots.txt for a blanket disallow before assuming you are visible. OpenAI notes that robots.txt changes can take roughly 24 hours to register.
Sources
- OpenAI (2026). Overview of OpenAI Crawlers: GPTBot, OAI-SearchBot, ChatGPT-User, OAI-AdsBot. OpenAI developer documentation. Link
- Aggarwal, P., Murahari, V., Rajpurohit, T., Kalyan, A., Narasimhan, K., & Deshpande, A. (2023, rev. 2024). GEO: Generative Engine Optimization. arXiv preprint 2311.09735, accepted to KDD '24. Link
- Google (2025). AI features and your website. Google Search Central documentation. Link
- Ong, S.Q. (2025, September 10). 100 Most Cited Domains in ChatGPT. Ahrefs, from the Brand Radar database of 9.6 million ChatGPT queries. Link
- Foursquare (2024, December 16). Announcement of the Foursquare and OpenAI partnership on ChatGPT search, Places API. Foursquare. Link
Find out what the machines say about you
Formative Digital, Brantford, Ontario
Every fix on this page works better once you know which of the four diagnoses applies to your business. Run the three-prompt check yourself, or ask us to run it and walk you through which records are producing the answer you are getting. Reach us through the contact page or by phone at 226-450-2065.
Written by Matt Griffin, founder of Formative Digital, Brantford, Ontario. Published 2026-07-20.