How do you get your store cited by ChatGPT and Perplexity?
Start by letting the search crawlers in: OAI-SearchBot for ChatGPT and PerplexityBot for Perplexity, plus a working sitemap. Then publish pages that answer the exact questions buyers ask, and get named on the lists those engines already cite. Measure with a fixed set of buyer prompts every week. When we checked 15 prompts for Mully, our own site was cited in 3, all from one comparison post.
In October 2026, AI search sent mymully.com 32 visits: 21 from ChatGPT, 10 from Perplexity and 1 from Claude (Drew's thread on X). That is small next to Google, but it is the channel buyers are moving toward, so we built an SEO and GEO agent to measure it every week and fix what it finds.
This guide covers what actually controls whether ChatGPT and Perplexity can cite your store, what our first check found, and how to measure it without guessing.
What decides whether ChatGPT or Perplexity can cite your store?
Their search crawlers have to be able to read your pages. ChatGPT search uses OAI-SearchBot, and OpenAI says sites that opt out "will not be shown in ChatGPT search answers" (OpenAI). Perplexity uses PerplexityBot to surface and link sites, and recommends allowing it in robots.txt (Perplexity).
Google is simpler. It says there are "no additional requirements" to appear in AI Overviews or AI Mode: a page has to be indexed and eligible for a snippet in normal search (Google Search Central). So regular SEO health is the price of entry for all three.
| Engine | Crawler that matters for citations | Separate training crawler |
|---|---|---|
| ChatGPT search | OAI-SearchBot | GPTBot |
| Perplexity | PerplexityBot (plus Perplexity-User for live fetches) | None, per Perplexity |
| Google AI Overviews and AI Mode | Googlebot, normal indexing | Not covered here |
Blocking GPTBot does not block ChatGPT search. OpenAI treats the two settings as independent, so you can opt out of training and still be citable.
What did our 15-prompt check find?
Our own site was cited in 3 of 15 buyer prompts. The agent ran 15 questions a golf subscription buyer might ask through AI search and logged what came back:
| Result | Prompts | What it looked like |
|---|---|---|
| mymully.com cited | 3 of 15 | All subscription or brand prompts. The same page every time: our subscription comparison post, at position 2 |
| Mully named, but someone else's page cited | 7 of 15 (includes the 3) | Third-party lists, like a golf site ranking MullyBox first and a Golfweek gift guide |
| Mully not found | 8 of 15 | Every gift, polo, outfit and stylist prompt |
Three lessons came out of it:
- One honest comparison page did all the work. The page that got cited compares golf subscriptions side by side. AI answers to "which subscription" questions want exactly that.
- Third-party lists count. More than half our appearances were someone else's page naming us. Getting on the lists that already rank is part of the job.
- Check your brand name. For "is mully worth it", the top results were a film, a podcast and a different brand called Mully London. If your name is generic, write the page that answers that question about you.
What technical fixes come first?
Make sure the crawlers can find and read everything. When the agent started, the live site had no sitemap.xml and no robots.txt, and thousands of old links from before our replatform returned 404s. The fixes, in order:
- robots.txt that allows all crawlers on public pages and points to the sitemap. Ours now does.
- A sitemap listing every live page. Ours now lists 98 URLs.
- Fix dead links that used to earn traffic. We restored 12 old blog posts at their original URLs and added 130 permanent redirects.
- Tell Bing. ChatGPT search does not publish which index it uses, but Bing is easy to notify. IndexNow submissions are shared across all participating engines, including Bing (IndexNow). We ping it every time a page on agentcocoon.com goes live.
None of this gets you cited on its own. It just removes the reasons you cannot be.
What kind of page gets cited?
A page that answers one buyer question directly, in its first paragraph, with specifics. Our cited post works because it names the options, compares them, and says who each one is for. That is the same structure we use for these guides: the question as the title, a short answer that stands alone, then sections that each open with an answer.
What does not work is volume. Our agent has a hard rule against mass-producing AI pages. Ten thin pages on "best golf gifts" lose to one page that actually picks.
How do you measure AI search visibility?
Use two numbers: AI-referral visits and a fixed prompt check. Visits come from your analytics, filtered by referring domain such as chatgpt.com and perplexity.ai. We track them in PostHog next to Google clicks, as one monthly KPI: 648 in October (616 Google clicks plus 32 AI visits), against a goal of 2,000 a month by March 31, 2027.
The prompt check is the same 15 questions every week, logged as cited, named or not found, with the page that won. Keep the list fixed. If you change the questions every week you cannot tell whether you improved or just asked easier ones.
Common questions
Do I need an llms.txt file to get cited by ChatGPT?
Not as far as any official documentation says. OpenAI and Perplexity both point to robots.txt and their crawlers, and Google says there are no special requirements for its AI features. We publish llms.txt on agentcocoon.com because it is cheap, but we do not count on it.
Does blocking GPTBot keep me out of ChatGPT search?
No. OpenAI says GPTBot (training) and OAI-SearchBot (search) are controlled separately. You can block training and still allow OAI-SearchBot so your pages can appear in ChatGPT search answers.
How long does it take to start getting cited?
There is no published timeline. Crawler rule changes take about 24 hours to register at OpenAI and Perplexity, but being cited depends on whether your page is the best answer to the prompt. We rerun the same prompts every week and only count a change once it holds.
Which prompts should I test?
The ones a buyer types before they know your name, like "best golf subscription box" or "gift for a golfer", plus a couple of brand checks like "is [brand] worth it". Keep the list fixed so week over week changes mean something.