OpenAI touches the web with three different crawlers, and most robots.txt files treat them as one. GPTBot collects training data, OAI-SearchBot builds the index behind ChatGPT search, and ChatGPT-User fetches a page live when a person asks about it. Blocking the first is a policy choice about model training. Blocking the second removes you from an answer surface that already sends measurable, well-tagged referral traffic. For a business site those are separate decisions, and the rest of this article treats them separately.
Set crawler policy deliberately
Read your robots.txt and check all three user agents by name. Most commercial sites should allow OAI-SearchBot, because it governs whether ChatGPT search can retrieve and cite you at all, and allow ChatGPT-User, because it acts on a real person’s request. GPTBot is the genuine judgement call, and either answer is defensible. Then confirm reality matches policy: search the server logs or CDN dashboard for OAI-SearchBot hits, and validate suspicious traffic against OpenAI’s published IP ranges, since impostor bots borrow the name. If your pages need JavaScript to show their main content, assume the crawler sees none of it and fix the raw HTML.
How citations actually behave
A ChatGPT search answer is a short composed response with a small set of linked sources, typically a handful rather than a results page. Every follow-up question triggers fresh retrieval, so a conversation can cite different sites at each turn, and the page cited is the one that answers the specific sub-question, not necessarily the biggest brand in the space. Clicks through those citations arrive tagged with utm_source=chatgpt.com, which makes this one of the easiest new channels to measure in GA4. The volumes are small for most sites today, but they land deep in the consideration stage.
Why mentions and original data earn inclusion
Cited pages tend to share two ingredients. The first is corroboration: the brand shows up consistently across directories, industry publications, comparison articles and community discussion, so quoting it carries low risk. Mentions do this work even without links, which quietly rehabilitates PR as an SEO activity. The second is original material: numbers you measured, surveys you ran, teardowns you performed, prices you publish openly. An engine composing an answer needs something concrete to attribute, and a page that only restates common knowledge offers it nothing that the model does not already have.
Publish for retrieval, not just ranking
The page patterns that win here are familiar from snippet work: one clear question per section, the conclusion in the first sentence, self-contained passages that survive being lifted out of context, a named author, and a visible, truthful updated date. Add one asset per quarter that contains facts existing nowhere else, even something as modest as a price benchmark drawn from your own sales data with client names removed. Original numbers are the cheapest citation bait a small site can produce.
The quarterly check, in one hour
Put a recurring hour in the calendar with four items. One: robots.txt still says what you decided it should say, since plugins and migrations silently rewrite it. Two: the logs show OAI-SearchBot fetching your commercial pages, not just the homepage. Three: a GA4 exploration filtered to session source containing chatgpt shows the referral trend, annotated whenever it steps up. Four: a fixed set of fifteen buyer questions run through ChatGPT search, recording who gets cited on each. Keep the answers in one sheet so the trend becomes visible by the second quarter.
Do the log check first
Before any content work, find out whether OAI-SearchBot has visited in the last month. If it has, you are competing for citations already and should act like it. If it has not, one line in robots.txt may be the highest-return edit you make this quarter.