How do you get cited by ChatGPT when you are a small business?
A small business gets cited by ChatGPT when its page surfaces in the web search ChatGPT runs for itself, not when its brand is well known. In a study of 1.4 million prompts published on April 15, 2026, Ahrefs found that 88% of the URLs ChatGPT ends up citing are pulled straight from its search channel. Fame, then, is not the lever, which is the only good news a ten-person company actually needs: what matters is being findable at the exact moment the question is asked.
This article covers how ChatGPT picks what it cites, what a small business can realistically win, what will stay out of reach no matter what it does, and the common situations where this work is simply the wrong thing to spend money on right now.
How does ChatGPT choose the sources it cites?
ChatGPT does not search for the sentence your customer typed. It rewrites it. OpenAI's own documentation on web search states that ChatGPT typically rewrites the user's query into one or more targeted queries sent to search providers, and that it may send further, more specific queries after reviewing the first results. OpenAI names Microsoft and Shopify among those providers.
Those intermediate queries have a name in the Ahrefs study of April 15, 2026: fanout queries. ChatGPT breaks a broad question into factual sub-questions, then compares the titles of the pages it finds against those sub-questions. Ahrefs measured that semantic closeness between a page title and its best-matching sub-question is markedly higher for cited pages than for ignored ones. In other words, the competition does not happen on your customer's question. It happens on a sub-question your customer never wrote.
The filtering is harsh, and much of it happens before your page is ever opened. According to the same Ahrefs study, ChatGPT cites only around half of the URLs it retrieves, and the citation rate depends heavily on the channel the URL arrived through: 88.46% for the general search channel, 12.01% for news, 1.93% for Reddit. ChatGPT uses Reddit to understand a topic, and almost never to source it.
Why can a small business get cited without being famous?
A small business can get cited because ChatGPT's sub-questions are narrow, and nobody famous is fighting over them. A sub-question like "which dentist near Leeds city centre takes emergency appointments on a Sunday" is of no interest to Wikipedia or Forbes. On that ground, a small business competes against another small business, or against nobody at all.
Location sharpens the effect for anyone serving a local market. OpenAI's documentation states that ChatGPT infers an approximate location from the IP address and may fold it into the rewritten query: a request for a restaurant becomes a query containing the name of the city. The geography is therefore already inside the question sent to the search providers, without the customer having thought to type it. A business that names its city and its service area explicitly on its pages is playing along with that mechanism. A business whose site says "we cover the whole region" is not.
Two more measurements from the Ahrefs study of April 15, 2026 point the same way. Results with human-readable URL slugs showed a citation rate of 89.78%, against 81.11% for the rest. And the median age of a cited page in the search channel sits around 500 days, which means one genuinely useful page that ages well beats a stream of fresh posts. We went through that relationship with time in "How long before you see an effect on your AI visibility?".
What is the access condition most sites overlook?
Before any question of content, your site has to let OpenAI's search crawler through. It is called OAI-SearchBot, and OpenAI's crawler documentation is explicit: sites that have opted out of OAI-SearchBot will not be shown in ChatGPT search answers. A block inherited from a firewall, an anti-bot toggle at your host, or a line of robots.txt written two years ago is enough to remove you from the running, however good your pages are.
That setting is independent of the others, and this is the part people get wrong. OpenAI states that each setting stands on its own: a site can allow OAI-SearchBot, which exists to surface it in answers, while refusing GPTBot, which collects content that may be used to train the models. Allowing yourself to be cited does not commit you to handing over training data. Plenty of sites blocked both in a single gesture, meaning to protect their content, and quietly removed themselves from the answers at the same time.
This condition takes minutes to check, and it is the only item on this list that is genuinely binary. Everything else, content and relevance, is work you do without a guarantee.
What will a small business never get, even doing everything right?
A small business will not get the generic queries, and no technical setting will hand them over. In the Ahrefs ranking of the most-cited websites in ChatGPT, updated September 2, 2026, reddit.com sits first with 16.8% of citations and en.wikipedia.org second with 7.0%, across a sample of US queries. On a question as broad as the best invoicing software, you will not outrank either of them.
That concentration is structural. Reviewing the 1,000 most-cited ChatGPT pages from September 2025, Ahrefs sorted 67.7% of them into categories a company cannot influence through content work: Wikipedia, homepages, and app store listings. The same count carries something you can act on, though: homepages account for 23.8% of that list. Your homepage is not a brochure. It is the page answer engines cite most readily.
Finally, there is no guarantee of citation, and nobody can sell you one. The same question asked twice will not produce the same ChatGPT answer, which is exactly the subject of "Asking ChatGPT whether it knows you measures nothing". What you can work on is eligibility: the probability that your page gets retrieved, then kept. That probability is measurable over time, across repeated questions. It is not a box you tick.
When is this not your priority?
This is not your priority if your site has nothing to cite. A three-paragraph about page and a contact form cannot answer a factual sub-question, and no technical fix compensates for missing content. In that case, write the pages that answer the questions customers already ask you on the phone, and come back to AI visibility afterwards.
This is not your priority either if nobody is asking a question your company could be the answer to. An industrial subcontractor whose five clients renew through tenders has no demand to capture inside ChatGPT, and that budget belongs elsewhere. Nor is it your priority if your platform locks the files you would need: Shopify, Wix, Webflow and Framer all forbid writing certain files at the domain root, which caps what you can fix however hard you try.
The urgency deserves some perspective too. Search is not being replaced, it is being squeezed, and it remains the pipe that feeds the answers. Ahrefs, comparing 300,000 keywords between December 2023 and December 2025, found that when an AI Overview is shown, the click-through rate of the top organic result is 58% lower. That is a real loss, and it sits on the same plumbing: the Ahrefs study of April 15, 2026 puts 88% of ChatGPT citations in the search channel. If your conventional search visibility is broken, fix that first. The two projects are one project seen from two sides.
What does it cost to find out where you stand?
Finding out costs nothing at Biencité. The scan is free and runs without an install or an account: it reads your homepage and your public files, then reports which of our signals were verified, which are impossible on your platform, and which could not be measured at all. An unreachable signal drops out of the calculation instead of scoring zero against you, and the report states how many signals were actually checked.
Beyond the diagnosis, weekly tracking of what five AI assistants answer, ChatGPT, Claude, Perplexity, Gemini and Le Chat, starts at 24 € per month on Starter and 39 € per month on Pro. If you would rather the fixes were installed for you than handed to your developer, installation is 290 € or 490 € depending on the site. The full breakdown is on /en/tarifs. And if the honest answer for your situation is "not yet", the free scan will tell you that just as well as a sales call would.
Frequently asked questions
Do you have to be well known to get cited by ChatGPT?
No. Per the Ahrefs study of April 15, 2026, 88% of the URLs ChatGPT cites come from its search channel: the page cited is the one that surfaces for the rewritten query, not the one belonging to the best-known brand. Fame counts on very broad questions, which a small business was never going to win anyway.
Is an llms.txt file enough to get cited by ChatGPT?
No, no file triggers a citation. OpenAI documents exactly one access condition for ChatGPT search: not having opted your site out of OAI-SearchBot. The rest depends on whether a page exists that answers the sub-question, and whether it appears in the search results being queried.
Does blocking GPTBot stop me being cited in ChatGPT?
No. OpenAI states that each setting is independent of the others: a site can allow OAI-SearchBot, the crawler that exists to surface it in search answers, while refusing GPTBot, which collects content that may be used to train the models.
How long before this has an effect?
There is no guaranteed timeline. Access fixes are verifiable as soon as the crawler next comes through. For content, the median age of a page ChatGPT cites in its search channel sits around 500 days per the Ahrefs study of April 15, 2026, which argues for thinking in months rather than weeks.
How do I find out whether ChatGPT cites my business today?
Not by asking it: the same question produces different answers from one session to the next. You need to re-ask a fixed set of chosen questions, regularly, and count the appearances across several weeks. That is the difference between a measurement and an impression.
- chatgpt
- ai citations
- small business
- geo
- ai visibility
- oai-searchbot