ChatGPT finds a business by searching the web when a question needs current information, reading the public pages it retrieves, and citing some of them. It has no private directory and no ranking you can buy. A professional firm controls whether its pages can be crawled, how clearly they state what it does, and whether other sites describe it consistently.
No method guarantees a mention. Answers differ by wording, run and moment, and OpenAI itself says search results and citations can be incomplete, outdated or incorrect. What a firm can do is remove obstacles, make its facts easy to retrieve and check the result periodically. Each of those is a few hours of work, not a campaign, and none of it needs a specialist tool.
What happens between a question and a named firm
When someone asks "who can help a small business with a cross-border tax question?", ChatGPT may decide to search. OpenAI's help page says it can search the web automatically when a question would benefit from current information, and users can also start a search manually. The rest of the path, as far as OpenAI documents it, looks like this.
- The question is turned into search queries. OpenAI says ChatGPT search sometimes works with other search providers, and when it does it typically rewrites the question into one or more targeted queries for them. The help page links to Microsoft's privacy statement as one of those providers' policies, which is a good reason to check that your site is indexed in Bing as well as Google.
- Results come back from a search index. OpenAI's own crawler for this purpose is OAI-SearchBot, described as the one used to surface websites in ChatGPT's search features.
- Pages are read and summarised. The model reads what it retrieved and writes an answer that may cite sources.
- Sources are shown. Answers that use search may include citations and a Sources panel.
- Local questions use location. OpenAI says ChatGPT may use an approximate location based on the user's IP address, and that sharing a precise device location is optional and off by default.
That last step matters more for professional firms than it looks. If the model only knows roughly where the user is, the area you serve has to be stated in words on your pages; a map embed, or an address that appears only in an image, tells it little about which towns you cover. So a firm appears when it is retrievable (crawlable, indexed somewhere ChatGPT looks), relevant (its pages say what the question asks about) and corroborated (other places agree). None of these is a trick. Each can be checked.
The three crawlers, and why they are different settings
OpenAI documents separate robots with separate jobs. Mixing them up is the most common technical mistake, and it can cut a firm out of search or leave content open to training against its wishes.
| User agent | What OpenAI says it does | If you block it |
|---|---|---|
| OAI-SearchBot | Surfaces sites in ChatGPT search results | Your site will not be shown in ChatGPT search answers, though it can still appear as a navigational link |
| GPTBot | Crawls content that may be used to train its models | Signals that your content should not be used for training |
| ChatGPT-User | Visits a page when a user's action in ChatGPT asks for it | OpenAI says robots.txt rules may not apply because the visit is user-initiated, and it is not used to decide search inclusion |
OpenAI also says each robots.txt setting is independent: a site can allow OAI-SearchBot to appear in search while disallowing GPTBot. It recommends allowing OAI-SearchBot, and allowing requests from its published IP ranges, to help a site appear. Firms behind a security service that blocks unknown bots sometimes lock out the search crawler without knowing, so ask whoever runs your website to check the firewall as well as the robots file.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowChecking your own door: a five-minute robots test
Open yourfirm.example/robots.txt in a browser, using your real address. You are looking for lines that mention these names. A healthy arrangement for a firm that wants to be findable but not used for training might look like this (illustrative):
User-agent: OAI-SearchBot
Allow: /
User-agent: GPTBot
Disallow: /
User-agent: *
Disallow: /client-portal/The risky pattern is a blanket rule written by a plugin or a developer years ago:
User-agent: *
Disallow: /That tells every compliant crawler to stay away from the whole site, including the one that feeds ChatGPT search. Also check whether the site uses a "noindex" setting site-wide, usually left on by mistake after a redesign. If you change the file, OpenAI says it can take about 24 hours for search systems to adjust, so wait before testing again. If you have never looked at the file, this single check is worth doing before anything else.
What a firm controls, and what it does not
| You control | You partly influence | You do not control |
|---|---|---|
| Crawler access in robots.txt and firewall | How other sites describe you | Whether ChatGPT searches at all for a question |
| Clarity of your service pages | Reviews and directory listings | Which sources it picks on a given run |
| Names, addresses and descriptions on your own site | Press or association mentions | The wording of its answer |
| Page speed, working links, current content | Whether people link to your guides | Competitors' pages or the search index itself |
Notice that paying is nowhere in that table. OpenAI's help page on ads says ads do not influence the answers ChatGPT gives, run on separate systems from the chat model, and are labelled. Ads can appear for people on Free and Go plans, and OpenAI's ad policies restrict some professional categories, such as financial, health and legal services, to approved advertisers on a case-by-case basis. If you are weighing ads, ChatGPT ads for law firms covers one professional category in detail, but ads are a separate channel from being recommended in answers.
Why clear pages get retrieved: what to put on them
ChatGPT reads text. A page that states, in plain sentences, who you help, with what, and where you are based gives it something to retrieve. A page that relies on a slider image and the words "bespoke solutions" does not. Think of three page types.
The "who we help and how" service page
One page per service, opening with a two-sentence description a stranger could repeat. For an illustrative small accountancy practice: "We prepare year-end accounts and tax returns for sole traders and small limited companies, and advise on payroll set-up. Fixed fees are agreed before we start." That reads as an answer to a question. A page that begins with a mission statement does not.
The people page
AI answers about professional firms often name individuals. Give each professional a short paragraph: role, qualifications you can state accurately, and the types of client you serve. Keep the spelling of names, firm name and credentials identical everywhere.
The question-and-answer page
The questions clients ask before hiring you, answered in full sentences: what it costs, how long it takes, what you need from them. The thing that rarely appears on competitors' pages is a stated range or a process, because many firms avoid saying anything specific. Specific text is more useful to a model and to a person.
Google's own guidance on its AI features makes a related point: it says unique, non-commodity content is likely to matter more over the long run than the other suggestions, and that special files or markup are not needed for Google's generative features. OpenAI does not publish an equivalent checklist, so treat page clarity as good practice rather than a documented ranking rule. For firms wanting the practical checklist, how to improve AI SEO and get your firm cited turns this into steps.
Consistency across the places that describe you
ChatGPT's answers draw on more than your website. A firm's description on a professional body directory, a review site, a local chamber listing and a social profile are all pages it might retrieve. When they disagree, an answer can mix them up: an old address, a partner who left, a service you stopped offering.
Write the firm's facts once, on a single page: legal and trading name, address, phone, hours, the services you actually sell, each professional with the credentials you can state, and a description short enough to paste into any directory's "about" box. Then audit five listings against it. Illustrative example: a three-partner consultancy found its directory page still listed a founder who had retired two years earlier, and one aggregator showed a former office address. Fixing those two entries took less than an hour and removed the two statements most likely to be repeated wrongly.
A test you can run this week
You cannot see inside ChatGPT, but you can ask it the questions your clients ask and log what happens. Use five prompts that cover how a client might phrase the need, and run each on three different days.
1. Who are good options for [service] for a small business?
2. What should I look for when choosing a [profession] for [situation]?
3. Tell me about [your firm name]. What do they do?
4. Is [your firm name] a good choice for [client type]?
5. Compare three firms that offer [service] for [client type].Prompts three and four are the most revealing: they show what ChatGPT says when it already knows your name. Log each run in a simple sheet, one row each.
| Date | Prompt | Firm named? | Sources cited | Facts correct? | Note |
|---|---|---|---|---|---|
| Day 1 | 3 | Yes | Own site, directory | Old phone number | Fix directory |
| Day 1 | 1 | No | Two competitors, one guide | n/a | Write comparison page |
| Day 4 | 1 | No | Different guide | n/a | Pattern holds |
(Illustrative log.) After a week you have 15 data points. If prompt three gets your name wrong or says it cannot find you, you have a retrieval or consistency problem. If prompt one never names you but others, you have a relevance problem: your pages do not say what the question is about, or others are better corroborated. For tracking over time, how to track whether AI search is sending you customers covers the analytics side.
A worked example: a four-person consultancy over four weeks
Here is an illustrative run, with invented but plausible numbers, to show how the test and the fixes fit together. A four-person management consultancy serving small manufacturers runs the five prompts on three days, so 15 runs. Prompts three and four are 2 prompts across 3 days, which is 6 runs, and the firm is named correctly in 4 of them; in the other 2 it is confused with a similarly named firm. Prompts one, two and five are 9 runs, and the firm is named in 1.
The diagnosis has two parts. First, a name clash: a differently spelled firm with a similar name shares a directory category. The fix is to make the firm's own pages state the full name, the founding year and the town in the first sentence of the About page, and to correct the directory entry. Second, the firm appears in 1 of 9 category runs, because its service pages open with a values statement rather than a plain description. The fix is to rewrite the first two sentences of the three service pages and add a fees-and-process page.
Four weeks later the same 15 runs give: 6 of 6 correct on the name prompts, and 3 of 9 on the category prompts. The movement from 1 to 3 is too small to prove anything, and a different set of days could have produced different numbers, which is exactly why the log exists. What the firm can say with confidence is that the name confusion is gone and that its pages now answer the questions asked. It should keep logging monthly rather than celebrate or panic.
The same question across three kinds of firm
The mechanism is the same, but the content that helps differs by profession. Three short illustrations:
- A small law practice. Prospective clients ask situational questions ("can a landlord do this?"). Pages that explain a situation in plain language, then say which matters the practice takes and which it does not, are more retrievable than a list of practice areas. Statements about outcomes carry regulatory risk, so keep to process and scope.
- An accountancy. Clients ask about fees, deadlines and software. Fee ranges, onboarding steps and the software you support are the specifics that make pages quotable. Keep the facts sheet current at filing season, when the questions spike.
- An architecture or surveying practice. Searchers ask about process, permissions and timescales. Project write-ups with stated scope, stage and typical duration, written without client names, give a model concrete material. Image-heavy pages need real text captions, because the model reads words, not photographs.
In each case the pattern is the same: say what you do and for whom in sentences a stranger could repeat, and keep the facts identical everywhere.
How a sample answer reads, and what it tells you
Look at what a run might return for prompt three, shown here as an illustration rather than a real output:
"[Firm name] appears to be a small advisory practice that works with
manufacturers. According to its website, it offers operations reviews and
costing support. A directory listing also mentions an office in a different
town, which may be out of date. Sources: firm website, directory listing."Read it as a diagnostic. The first sentence says the model understood the firm's niche, which is good. The second shows which pages it relied on, so you know which to keep strong. The third is a warning: the model noticed a conflict between sources and hedged. A hedge like "may be out of date" points at precisely the listing to fix. Each clause maps to an action, which is more useful than a general feeling about "AI visibility".
When the answer about your firm is wrong
Sooner or later a run will state something false: a service you do not offer, a fee, a partner. Do not argue with the chat. Open the cited sources and find where it came from. Often the error lives on a third-party page. Fix the page if you control it, or ask the site owner to correct it. If the model stated something with no source, note the date and wording in your log and move on, because you cannot instruct it to remember a correction. OpenAI's help page says that if a result looks wrong, a user can ask it to search again using a specific source, date or location, so link your own corrected page in future when you test.
For regulated professions, an incorrect statement about services or credentials is worth recording in a file note, with the date and prompt, in case a client relies on it. This is practical hygiene, not legal advice; take your own regulator's rules on advertising and on public statements as the guide.
Things that look clever and do not help
- Stuffing the same keyword on every page. It makes pages worse for people and gives a model nothing new.
- Writing a page for every phrasing. Google calls scaled pages written mainly to manipulate responses spam, and models gain little from near-duplicates.
- Buying fake reviews or mentions. They are fragile and risky for a professional whose reputation is the product.
- Assuming a special file fixes it. Google says its generative features do not use llms.txt; whether OpenAI's systems read such files is not documented on the pages I checked, so do not rely on one. Does llms.txt work? weighs the evidence.
- Treating one screenshot as proof. Answers vary; only a log over time tells you anything.
A four-week plan for a small firm
- Week one: unblock and baseline. Check robots.txt and the firewall; run the five prompts on three days; start the log.
- Week two: fix facts. Build the facts sheet; correct the five most visible listings; update people pages.
- Week three: answer pages. Rewrite the opening lines of your main service pages; add one question-and-answer page covering fees and process.
- Week four: re-test. Wait the 24 hours after any robots change, repeat the prompts, compare with week one and decide what to do next.
That plan costs mostly time. It changes nothing about how clients experience you, and each step improves the firm's visibility to people using ordinary search as well. If you want a wider view, how to get your business recommended by ChatGPT and AI search covers the same ground for any small business.
Common questions about being found in ChatGPT
Can I pay to be recommended in ChatGPT answers?
No. OpenAI says ads do not influence the answers ChatGPT gives and are shown separately and labelled. Ads on Free and Go plans are a separate paid product with their own rules, and professional categories such as legal and financial services face extra restrictions. Check OpenAI's ad policies before planning spend.
Should I block GPTBot to protect my content?
Blocking GPTBot only signals that your content should not be used to train OpenAI's models. It is separate from OAI-SearchBot, which controls whether your pages can appear in ChatGPT search answers. Many firms block one and allow the other. Decide each on its own.
How long after changing robots.txt do changes take effect?
OpenAI's crawler documentation says it can take about 24 hours for its systems to adjust after a robots.txt update for search. Do not judge the result the same day. Re-run your test prompts after a few days, and record the date of the change.
Does ChatGPT always show the same firms?
No. Answers vary between runs, accounts, locations and wording, and search results change over time. That is why a single screenshot proves very little. Run each test prompt several times on different days and log what comes back before drawing a conclusion.
Further reads
- Can Estate Agents Be Found in AI Search? What to Fix Now — How another local professional trade approaches it.
- Best AI Visibility Service Providers: What a Small Firm Should Ask — Questions to ask before paying a visibility provider.
- How AI Overviews Are Changing Local Search for Small Businesses — How Google's AI answers affect local firms.
- PR for AI Search: How Agencies Earn Citations in ChatGPT Answers — How third-party coverage earns citations in AI answers.
- How to Monitor Brand Mentions in ChatGPT for a Client, Week by Week — A weekly schedule and log for checking mentions.
- AI SEO for Plumbers: Get Named by ChatGPT and Google AI Overviews — See the same checks applied to a trade business.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: OpenAI Help Centre (Searching the web with ChatGPT), OpenAI crawler documentation (OAI-SearchBot, GPTBot, ChatGPT-User), OpenAI ad policies and Help Centre article on ads in ChatGPT.