What "appearing" actually means

Appearing in ChatGPT means being named in an answer, either from the model's memory or from pages it retrieves live, and only the second route is open to most businesses.

From the model's memory. The name was frequent enough in training data that the model produces it unprompted. You cannot edit this, and for most businesses it does not apply.

From live retrieval. The model runs a search, fetches pages, and composes an answer from them. This is the path that is open to a business of any size, and it is what this page is about.

The distinction sets the bar. You are not trying to change what a model believes. You are trying to be present, legible and unambiguous in the small set of documents it fetches at the moment someone asks. This article covers that first gate; the full order of work for a small company is in the series' main guide.

The concentration problem, and why it is not hopeless

AI answers concentrate on fewer and larger websites than classic search does, but that concentration is measured per question, which leaves room for a business with a specific position.

Researchers at MIT (Aral, Li, and Zuo, 2026) found that AI search refers to the top 1,000 websites by traffic significantly more than traditional search does, and to the long tail significantly less. They also measured markedly lower response variety than in classic search results: ask the same question repeatedly and AI answers converge, where ten blue links diverge.

Read carelessly, that says a small business has no chance. Read properly, it says something more useful: the answer set is small, so the question is whether you are in it for your query, not for every query. Nobody wins "best CRM". A supplier with a genuinely specific position can be the retrieved answer for the narrow question its buyers actually type.

The four checks your site has to pass

Every business that fails to appear in ChatGPT fails at one of four checks, and usually at the first one.

Check What breaks it How to test it Typical fix
1. Crawler access robots.txt rule or CDN bot filter blocking OAI-SearchBot Read robots.txt; look for 403s to OpenAI agents in edge logs Allow OAI-SearchBot; OpenAI says changes take about 24 hours to apply
2. Content without JavaScript Offer, services and location only appear after client-side rendering curl the page and read the raw HTML Server-side rendering or static HTML for key claims
3. One entity Different names, descriptions or addresses across your site and profiles Ask ChatGPT to describe your company and compare with reality One consistent description, corroborated on external sources
4. Statements, not atmosphere Slogans instead of facts about what you do, for whom, where Pick a paragraph and ask whether it could be quoted as an answer Rewrite key pages as plain, specific sentences

Gate 1: The crawlers can reach you

OpenAI runs three user agents with different jobs, according to its crawler documentation: OAI-SearchBot decides whether your pages can be shown in ChatGPT search answers, GPTBot collects content for training foundation models, and ChatGPT-User fetches pages when a user's request triggers it and is not used to decide what appears in search.

Each one is set independently in robots.txt. Blocking GPTBot keeps your content out of training but does not remove you from search; blocking OAI-SearchBot does. A robots.txt written years ago for search engines, a bot filter at the CDN, or a firewall rule can block OAI-SearchBot while your Google rankings stay untouched, so nothing in your analytics tells you it happened. Check what your file says about all three agents and whether your edge returns 403s to them. It is a common cause of total absence, and the cheapest to fix.

Gate 2: The content survives without JavaScript

A retrieval fetch is not a browser session. If your key claims (what you do, who you serve, where you are) only exist after client-side rendering, the fetched document is an empty shell. Curl one of your own pages and read what comes back. That text, and only that text, is what the model has. Why this has turned from a classic SEO detail into an AI visibility problem is explained in JavaScript SEO for AI crawlers.

Gate 3: You resolve to one entity

This is where most technically healthy sites lose. Kargaev (2026) measured which signals actually correlate with generative-engine performance and found brand entity mentions dominant at NIS 0.918, far ahead of anything else in the hierarchy. The same analysis found the classic technical signals close to irrelevant on their own: page speed at NIS 0.000, content length at NIS 0.043.

That is a strong claim and worth stating carefully: speed and length still matter for humans and for classic ranking. They just do not decide whether an AI names you. What decides it is whether "your company" is a single, resolvable thing: one name, one description, one address, corroborated in enough places that the model is not guessing between three plausible versions of you. Our research on brand entity signals goes through what that corroboration looks like in practice.

Gate 4: The text reads as statements, not atmosphere

Ubiratan de Oliveira (2026) calls this generative legibility: content that both a person and an inference process can interpret. A paragraph that says "we deliver excellence through innovative solutions" is legible to neither. A sentence that says "we install industrial flooring in food-production facilities in the Netherlands and Germany" can be extracted, attributed and repeated.

Iyappan (2026) quantifies the gap: long-form contextual content achieves a 92% citation rate in AI answers, entity-rich content 89%, structured-data-heavy pages 85%, FAQ-formatted content 67%, and keyword-focused thin content 41%. The spread between the top and bottom of that range is the difference between writing for a crawler and writing for a reader.

A test you can run this afternoon

Five checks, most of them under ten minutes, show whether ChatGPT can find and name your business today.

  1. Open ChatGPT with web search on and ask the five questions your buyers ask before they contact you. Not your brand name, but the problem.
  2. Record what comes back: who is named, which pages are cited, how each company is described.
  3. Fetch your own most important page with curl and read the raw HTML. Is the offer there in plain text?
  4. Check robots.txt and your CDN logs for OAI-SearchBot, GPTBot and ChatGPT-User.
  5. Search your company name and read how the model describes you. A wrong or vague description is a corroboration problem, not a content problem.

Steps 1 and 5 will tell you more than any tool subscription.

What to do about each outcome

Each result of the test points to one of the four checks, so the outcome tells you where to work next.

Not named, competitors are. You are eligible but not selected. That is a different problem, covered in the entity-clarity article listed below.

Named but described wrongly. The model is assembling you from weak sources. Fix the descriptions you control, then work on corroboration elsewhere.

Nothing at all, for any question. Go back to checks 1 and 2. Absence this total is almost always access, not quality.

Named only when you use your own brand name. You exist as an entity but not as an answer to a problem. That is a content-shape problem: your pages describe the company rather than the question.

In this series

Frequently asked questions

How do I get my business listed on ChatGPT?

There is no listing or submission form that decides inclusion in ChatGPT. What you control is access for OAI-SearchBot, pages that read without JavaScript, one consistent business description, and the external corroboration that makes the model confident enough to name you.

How long does it take to appear in ChatGPT?

Once a blocker is removed, live retrieval can pick up a page within days, because it fetches at query time rather than waiting for a training cycle. OpenAI states that its systems take about 24 hours to adjust to a robots.txt change. Being chosen consistently takes longer and depends on corroboration elsewhere.

Does ChatGPT use Google's index?

ChatGPT search uses OpenAI's own crawler, OAI-SearchBot, to find the sites it can show, separate from Googlebot. Overlap with Google's results exists but is partial, which is why a business can rank on Google and still be invisible in ChatGPT.

Does blocking GPTBot hurt me?

Blocking GPTBot does not remove you from ChatGPT search, because GPTBot only collects content for training OpenAI's models. The crawler that decides whether you can be shown in search answers is OAI-SearchBot. Some publishers block GPTBot deliberately for content licensing reasons; if you want to be recommended, make sure OAI-SearchBot stays allowed.

Does ChatGPT-User follow robots.txt?

Not necessarily: OpenAI says that because ChatGPT-User acts on a user's request, robots.txt rules may not apply to it. It is also not used to decide whether content appears in search, so it is not the agent to focus on for visibility.

Why does ChatGPT name my competitor and not me?

Usually because they read as a specific, corroborated answer to that question and you read as a general company page. Concentration in AI answers is real (Aral, Li, and Zuo, 2026), so the margin goes to whoever is unmistakably the answer to the narrow question.

Do I need structured data?

It helps materially, since structured-data-heavy pages cite at 85% against 41% for thin keyword pages (Iyappan, 2026), but it does not rescue a page the crawler cannot reach or a business the model cannot resolve. Order matters.

Does appearing in ChatGPT bring traffic?

It brings fewer sessions than search, of better quality: AI-referred traffic converts at 14.2% against 2.8% for organic search, according to Iyappan (2026). The value is in who arrives, not how many.

If you want the five-question test run against your own site, with the crawler access and entity checks done properly, the free scan does exactly that and the result is yours either way.