How to show up in ChatGPT: what gets your business cited in 2026

ChatGPT answers some questions from what its model already knows and searches the web for others. OpenAI’s help page on searching the web says it “may search the web automatically when your question would benefit from current information”, and a person can also start a search by hand. Web search is available on ChatGPT’s free plan as well as its paid ones, and to people who are not signed in. Answers that use search can include citations, and a Sources button lists the pages behind them. Being one of those links is what “showing up in ChatGPT” means for a business. The audience is large: in March 2026 OpenAI said ChatGPT had more than 900 million weekly active users and that search usage had nearly tripled in a year.

OpenAI runs separate crawlers for separate jobs, and its crawler documentation is explicit about which one matters here. OAI-SearchBot “is used to surface websites in search results in ChatGPT’s search features”, and “sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers, though can still appear as navigational links.” GPTBot is a different crawler, used to gather training data. The settings are independent: a site can allow OAI-SearchBot and block GPTBot, and still be eligible to appear in ChatGPT search. ChatGPT-User is the agent ChatGPT uses when a person’s request makes it visit a page; because a person starts those visits, OpenAI says “robots.txt rules may not apply”, and it says ChatGPT-User “is not used to determine whether content may appear in Search.” For search, OAI-SearchBot is the setting that counts.

OpenAI’s October 2024 announcement of ChatGPT search adds two details that matter. It says “Any website or publisher can choose to appear in ChatGPT search.” And it says ChatGPT search “leverages third-party search providers, as well as content provided directly by our partners”, which suggests being indexed by the major search engines still matters. One more timing detail from the crawler page: after you change robots.txt, it “can take ~24 hours” for OpenAI’s systems to adjust.

What OpenAI does not publish is a ranking recipe. Its help page says only that “ChatGPT ranks search results using multiple factors intended to help users find relevant, reliable information. Placement is not guaranteed.” It does not say how it chooses between two pages that both answer a question. Everything past access in this guide is therefore evidence about how AI answer engines behave in general, from controlled studies, plus our own measurements, rather than a documented ChatGPT rule.

This is the step that decides everything else, and it is the one most businesses already pass without knowing it. When we read the robots.txt files of 395 Calgary business sites in July 2026, only about 4% blocked any AI crawler, and only about 2% blocked a crawler that powers live answers. So a deliberate block is rarely the problem. These are the checks worth running.

JavaScript is a separate problem from robots.txt, and a harder one to spot. In an analysis of AI crawler traffic across Vercel’s network, published in December 2024, Vercel and MERJ found that ChatGPT’s and Claude’s crawlers fetch JavaScript files but “don’t execute them. They can’t read client-side rendered content.” A site whose text only appears after scripts run can look empty to the crawler that decides whether you get quoted. Crawlers change, and that measurement is nearly two years old, so test your own pages: view the source, not the rendered page, and look for your headline and your prices.

Firewalls are the quiet one. Some bot protection settings challenge anything that is not a regular browser, which is sensible against scrapers and can stop OpenAI’s crawlers reaching your pages. OpenAI publishes the IP ranges its crawlers use, so a firewall can be told to let them through, and its help page makes this part of eligibility: allow OAI-SearchBot to crawl and “confirm that the website host or content delivery network allows traffic from OpenAI’s published searchbot IP addresses.” If your host offers AI crawler controls, check what they are set to.

A buyer does not ask ChatGPT for “comprehensive marketing solutions”. They ask which Calgary agency can build a website for a restaurant, what SEO costs for a small business, or whether a contractor needs Google Ads. Answer that question directly, near the top, in plain words. Our Calgary restaurant page, the one that brought the lead, is built that way: one kind of buyer, a direct answer under a label near the top, and a short list of the questions operators ask.

Controlled evidence on what else helps comes from a paper by Pranjal Aggarwal and colleagues, accepted at KDD 2024, the Association for Computing Machinery’s data mining conference. They rewrote web pages in nine different ways and measured how much of each page generative engines used in their answers. On their own benchmark, adding citations from credible sources, quotations and statistics did best. In a smaller test of 200 queries on Perplexity, adding quotations raised the paper’s main visibility score by about 21% and adding statistics by about 9%. Keyword stuffing, the oldest SEO habit there is, offered “little to no improvement” on the benchmark, and on Perplexity it lowered that score by about 9%.

Four cautions keep that finding honest. The main results come from the authors’ own benchmark of queries, not from ChatGPT, and the Perplexity test gave it the source texts as uploaded files rather than letting it search the web. The paper scores visibility two ways, and on its second score keyword stuffing actually rose on Perplexity, so the measures do not always agree. The gains were largest for sources that ranked lower to begin with, while some methods reduced visibility for the top-ranked source. And later controlled tests on newer models mostly did not repeat the gains: C-SEO Bench, presented at NeurIPS 2025, found most of these rewriting methods “largely ineffective” and often harmful to a page’s ranking. What survives is narrower: keyword stuffing does not help, and specific, sourced writing is worth doing because readers need it, not for a guaranteed lift in AI answers.

Google’s position on its own AI features lands in the same place. Its guide to optimizing for generative AI features says “optimizing for generative AI search is optimizing for the search experience, and thus still SEO.” It says there is no requirement to break content into tiny pieces for AI, that structured data “isn’t required for generative AI search”, and that creating content for every variation of how people search, “primarily to manipulate rankings or generative AI responses in Google Search”, violates its scaled content abuse policy. That guidance is about Google, not ChatGPT, but Google has written its rules down, and they describe the same kind of page: one written for the person asking.

In practice that means one strong page per real buyer question, not forty thin ones. Put the answer in the first paragraph. Give the number, the range or the date. Name who said it and link where you got it. Say what you do not know. Put a real author’s name on it: OpenAI does not say how ChatGPT weighs authorship, and Google says E-E-A-T “itself isn’t a specific ranking factor”, but Google strongly encourages accurate bylines, and readers trust a name they can check. Those are the habits of a page a person trusts, and the most repeatable finding in a 2026 survey of this research is that answer engines favour the page that most directly answers the question asked.

This is the step most businesses skip, and the research gives good reason not to. In a 2025 study across multiple industries, languages and phrasings, Mahe Chen and colleagues found that AI search services showed a “systematic and overwhelming bias towards Earned media (third-party, authoritative sources) over Brand-owned and Social content”, a sharper tilt than Google’s results showed. The paper was a preprint, not yet published in a journal, when this was written, and its authors also describe a “big brand bias” that smaller businesses have to work against. The practical reading: AI search engines lean on what credible third parties say about you, not only on what you say about yourself.

Earned mentions for a local business look like this: complete and consistent listings on the directories your industry uses, reviews on the platforms buyers read, a mention in a local publication, a supplier or partner page that links to you, and original data that other people cite. Original data works because a number nobody else has published is a reason for a journalist, a blogger or an answer engine to mention you, which is why we publish our Calgary studies with their methods, and the data where we hold it, on our research page.

Plenty of what is sold as AI search optimization has no published evidence behind it, and some of it can backfire. These are the common ones.

We keep an llms.txt file on our own site because it costs nothing. We do not sell it as a way to get cited, because we have not seen evidence that it does.

Ads are a separate question. ChatGPT began testing ads on February 9, 2026, starting in the United States, and OpenAI’s help page on ads says “Ads do not influence ChatGPT’s answers.” Ads can appear below a response, clearly labelled and shown separately from it. An ad is a paid placement next to the answer, not a citation inside it.

ChatGPT has no stable ranking you can check the way you check a Google position, but you can see what it sends you. The visit that brought our first ChatGPT lead was tagged utm_source=chatgpt.com, as our lead handler’s records show. We could not find a live OpenAI page that documents the tag, so confirm it in your own data: any analytics tool that records campaign sources will show tagged visits. In Google Analytics 4, look at traffic acquisition by session source, and add a comparison for chatgpt.com. Look for visits referred by perplexity.ai, gemini.google.com, copilot.microsoft.com and claude.ai as well.

The step most people miss is carrying that source into the lead itself. Our lead handler now records the AI source on every form it forwards, so an enquiry from ChatGPT is labelled when it arrives. Without that, a lead from ChatGPT can look like any other direct enquiry, and the channel stays invisible in exactly the report that matters.

Expect small numbers at first, and judge them by what they lead to. A visitor who clicks through from an AI answer has already seen an answer that mentions you. For a business that wants leads, one enquiry tells you more than a thousand impressions.

It is tempting to treat “AI search” as one thing. It is not. When Pew Research Center tracked 900 American adults’ browsing in March 2025, people who saw a Google AI summary clicked a regular result in 8% of visits, against 15% when no summary appeared, and clicked a link inside the summary itself in just 1% of visits.

But not every search gets an AI summary. When we pulled 38 Google results pages on August 1, 2026, none of 18 local service searches captured an AI Overview, and 17 showed a map pack. Those were a city plus an agency service, such as “calgary seo” or “denver web design”. For local service searches like these, that points to Google’s map listing and reviews. ChatGPT is where people can ask the longer questions, who is good, what it should cost, what to watch out for, and it needs its own measurement. The work overlaps, because a clear, sourced, well-reviewed business gives all three something to show, but the scoreboard is different for each.

Then ask ChatGPT your buyers’ questions yourself, several times each on different days over a few weeks, and record the result as mentioned in X of N runs. One run proves little, because answers vary between runs, with wording and between users, but it is the most direct view you have of the shelf your business is competing for.

OpenAI’s crawler behaviour and publisher guidance are quoted from its crawler documentation, its help pages on searching the web and on ads, and its October 2024 announcement of ChatGPT search, all read in October 2026. The usage figures are from OpenAI’s funding announcement of March 31, 2026. An OpenAI publishers and developers FAQ we read in September 2026 has since been taken offline, and nothing in this guide rests on it, a reminder that OpenAI can change its guidance without notice. The content findings are from Aggarwal and colleagues, presented at KDD 2024, whose main results come from their own benchmark of queries rather than from ChatGPT, from C-SEO Bench, presented at NeurIPS 2025, from Chen and colleagues (2025), which had not been peer reviewed when this was written, and from Olivier Martinez’s survey of the research from 2023 to 2026, posted to arXiv in July 2026. The JavaScript finding is from Vercel and MERJ’s analysis of crawler traffic, published in December 2024. Google’s statements are from its guide to optimizing for generative AI features, which describes Google Search, not ChatGPT. The click figures are Pew Research Center’s, from 900 U.S. adults in March 2025. Our own figures come from our published robots.txt and AI Overview studies, each of which states its sample and date, and from our own lead records and Google Search Console data. OpenAI does not publish how ChatGPT ranks sources, so the steps here combine its documented access requirements with what research and our own measurements support; they do not guarantee a citation, and anyone who guarantees ChatGPT placement is selling something they cannot control.

• The AI blocking myth: what Calgary business sites block in robots.txt.
• llms.txt explained: what the file does and what it does not do.
• AEO tools for AI search visibility: tools that run the same prompts for you.
• Which searches get an AI Overview: local service queries against AI tool queries.
• The reviews-domain ChatGPT hack: why buying a reviews site does not get you cited.
• How much does AEO cost: budgets for AI search optimization.
• How to scale your business: when getting found is the constraint, and when it is not.