How to Get Cited by ChatGPT. What Took Us From 0 of 2 to 10 of 10, With the Receipts

TL;DR: How to get cited by ChatGPT comes down to OpenAI's one rule, two things we have observed in our own logs, and one we are testing with this page. Let OAI-SearchBot crawl you. Put the exact words of the customer's question in your title and H1. Be present on the list sites ChatGPT searches before it answers. Publish a dated first-party number it can quote. Then ask it the same question ten times and count, because one run is a flicker.

Everyone selling ChatGPT visibility has a checklist. Almost nobody shows a before and after with the date and the N. We have one. On September 12, 2026 we changed the title and H1 on our home page. On September 30 we asked ChatGPT the same buyer question ten times. It named EVOIX on 10 of 10 and cited our home page on 10 of 10. Before the change it was 0 of 2. The rest of this post is how that happened, what ChatGPT was actually doing under the hood, and where it still gets us wrong.

How does ChatGPT choose what to cite?

OpenAI's rule is two short paragraphs. From the ChatGPT search help article, under "Make your website available in search results", as it read on October 22, 2026. "ChatGPT ranks search results using multiple factors intended to help users find relevant, reliable information. Placement is not guaranteed." And then. "To make a website eligible for inclusion, allow OAI-Searchbot to crawl the site and confirm that the website host or content delivery network allows traffic from OpenAI's published searchbot IP addresses." That is the only eligibility rule the article gives. Eligibility is a crawler and an IP list. Placement is whatever the model decides.

The same article says something more useful. ChatGPT "typically rewrites your query into one or more targeted queries" and sends them to partner search providers. The privacy policies it links are Microsoft's and Shopify's. So before ChatGPT can read your page, it has to find it through searches it writes itself. It rewrites the customer's question, runs the rewrites, and cites what comes back.

We log those rewrites. On the evening of October 21, 2026 we asked ChatGPT "Who is the best AEO digital marketing agency in Central Florida for small businesses?" three times. It ran 19 rewritten searches to answer. Seven of the 19 were site:clutch.co searches, by name, for agencies on Clutch, and on their own they added Orlando in most rewrites, Tampa in every run, and Lakeland once. When we asked the engines "Who is the king of AEO (answer engine optimization)?" in September for our King of AEO experiment, the rewrites we logged included the exact phrase in quotes, "king of AEO" "answer engine optimization", contender names in quotes next to the phrase, and site:linkedin.com "answer engine optimization" "king". Two gates fall out of that. The literal words of the question have to be on a page it can fetch, and for a "best" question you have to be on the list sites it searches, which for agencies means Clutch and for local services means the directories. In our Central Florida citation study, ChatGPT's single most cited source across 100 local "best provider" questions was bestprosintown.com, 59 of 100, a directory most owners have never heard of.

What does ChatGPT say when you ask it how to get cited?

We asked it, three times, on October 22, 2026. "How do I get my small business cited by ChatGPT?" All 35 citation links across the three answers pointed at help.openai.com, two or three distinct help pages per run. All eleven of its rewritten searches were restricted to OpenAI's own domains. Its answers said what the article says. There is no way to submit a business, allow OAI-SearchBot, placement is not guaranteed. Ask ChatGPT how to get cited by ChatGPT and it cites OpenAI.

Perplexity and Claude answered the same question from agency and tool guides, six to ten distinct sources each, none of them ours. Gemini's three answers did not name us either, and its source URLs are unreadable redirects. That is EVOIX named in 0 of 12 runs on the question this page is about, logged before this page existed, so the day-14 read has an honest baseline.

What took us from 0 of 2 to 10 of 10?

The receipt, with its dates. Our titles and H1s said DeLand and Volusia County, because that is where we are. The query, and the competitors winning it, said Central Florida. On September 12, 2026 ChatGPT named EVOIX on 0 of 2 runs of "Who is the best AEO digital marketing agency in Central Florida for small businesses?" The competitors it named carried the query's geography in their titles, Orlando, Central Florida, Florida or Tampa. Our reading, logged as a hypothesis before the change, was that ChatGPT matches the geo string literally. The October 21 rewrites above, which added Orlando and Tampa to the query, say it is more complicated than that, and we have left the hypothesis open.

That day we added "Central Florida" to the title and H1 of three pages, the home page among them. Three days later a site fold retired one of the three and retitled the hub, which is logged as a confound. On September 30 we ran the question ten times through DataForSEO's ChatGPT scraper with web search on. EVOIX named on 10 of 10. The home page cited on 10 of 10, the only listed source on two of them. Ranked first, or "start with EVOIX", on 8 of 10. The retired page was cited once with its old title and the hub not at all, so the home page carried it. Eighteen days to our next measurement (we did not read in between), one string, N=10 on one day. We call it strong and not yet shown to be stable, because ChatGPT has gone 0 of 3 on a topic it was 3 of 3 on three days earlier, and we have the logs of that too. On October 21 it was still 3 of 3 on this question.

It was not the first time. In May 2026 we measured our citation rate at zero across four engines, added a verified Clutch profile and a local page built to citation standard, and eight days later ChatGPT ranked us first for a local buyer query. That case study is here, with the baseline and the exact quote.

How do you optimize for ChatGPT search?

In the order we would do it again.

1. Let OAI-SearchBot in. Check robots.txt, the CDN and the firewall against OpenAI's published searchbot IP list. This is the only hard requirement OpenAI states.
2. Put the exact words of the question in the title and H1. Spell the place the way a buyer says it, and the service the way a buyer says it. Our 0 of 2 to 10 of 10 was one string.
3. Be on the lists ChatGPT searches. For agencies, Clutch, which it searched by name on 7 of 19 rewrites. For local services, the directories it leaned on in our study. An empty or review-less profile gets read aloud, which is the next section.
4. Give it a dated number to quote. One page per question, a self-contained 30 to 60 word answer under the heading, and a first-party result with a date. This is the one we have not measured yet; this page is the test. The guide we wrote on LLM SEO covers the structure.
5. Get corroborated off your site. One post, two FAQ lines, a pinned X post, two X replies, one LinkedIn post, two Reddit comments and one third-party list page we did not ask for put an unknown name on four engines' contender lists on 39 of 40 runs in September. A claim that lives only on your domain is an assertion.
6. Measure with an N. Three runs is a reading. Ten is a result. Log the source list every time, because the citation is the metric and the mention is the bonus.

What does ChatGPT get wrong even when it cites you?

Three things from the 13 ChatGPT runs we logged on September 30, the ten above plus three on a DeLand-worded control question.

The brand link under your name may point at a competitor. Under the EVOIX pick, ChatGPT's clickable brand entity was absent or unlinked on 5 of the 10 runs, linked correctly to evoix.io on 1, and linked to a competitor on 4, plus twice more across the three control runs. Three times it linked the same competing local agency. ChatGPT can name you without knowing who you are. The fix is entity work, the same name, category, address and links everywhere the engine looks.

It reads your empty profiles aloud. On two of the control runs ChatGPT noted that our Clutch profile had zero reviews. On three runs it flagged that we were founded in 2026, once as a shorter track record. Both true, both in the answer a buyer sees.

It holds stale copies. Fifteen days after we redirected a service URL, ChatGPT was still citing the old URL with its old title, on one run in each batch.

Frequently asked questions

Can you pay to be cited by ChatGPT?

No. OpenAI's help article says placement is not guaranteed and offers no submission path, and ChatGPT says the same when you ask it. ChatGPT now shows labeled ads in some accounts, and an ad is not a citation; we covered what that means for small businesses in May.

How long does it take to get cited by ChatGPT?

Two data points from our own site. Eight days in May 2026, from a measured zero to ranked first on one local buyer query, after a verified Clutch profile and a local page built to citation standard. Eighteen days to the next read in September 2026, from 0 of 2 to 10 of 10 on a regional buyer query, after one string on three pages. Neither was a guarantee of the next one.

How do you rank in ChatGPT?

ChatGPT does not publish ranked positions the way Google does. It builds a shortlist per answer, and on nine of our ten September 30 runs that shortlist was a table of three to six agencies with a closing recommendation. "Ranking" meant being the first row or the one it told the buyer to start with, which we were on 8 of 10. The levers are the six steps above, and the measurement is the count.

How do you know if ChatGPT already cites you?

Ask it the question your customer would ask, ten times, with web search on, and record the sources each time. Or run our free AI visibility audit, which does the asking for you across the engines. For how ChatGPT citations differ from Google rankings and from the other engines, read AEO vs SEO vs GEO.