Getting cited by ChatGPT isn’t mysterious once you see the steps: it searches, it retrieves, it selects. Each step is something you can be built to win.
When ChatGPT answers with sources, it works in steps: it runs a search for the query, retrieves a set of candidate pages, and selects the ones that most directly and reliably answer the question. It favors content it can actually read, that resolves the query clearly, and that comes from sources it has reason to trust. Understanding those steps is the whole of optimizing for it — each one is a place you either qualify or get filtered out.
The first thing to know is that this only happens in browsing or search mode. Without it, ChatGPT answers from training data, with no live retrieval — and presence there depends on having a broad, established footprint across the web rather than any single optimized page. AEO targets the browsing path, where current, well-built content can be found and chosen in real time.
Search. ChatGPT issues a query to a search layer and gets back candidates. If your content isn’t retrievable there — blocked, unindexed, or invisible — you never enter consideration. Retrievability is the entry ticket.
Retrieve and read. It fetches candidate pages and reads them. Content that renders only in JavaScript it can’t run, or that buries the answer, gives it little to work with. Clean, readable HTML with the answer near the surface does well.
Select. Among what it read, it chooses the passages that answer the query most directly and that come from sources it can trust. This is where the answer capsule and corroboration earn their keep: a clear, standalone answer from a corroborated source is the safe pick.
Win each step in order. Be retrievable, so you enter the candidate set. Be readable, with content in the HTML and the answer near the top, so you survive the read. Be selectable, with direct, standalone answers from a source corroborated elsewhere, so you’re the safe citation. None of this is specific trickery — it’s the same well-built foundation AEO rewards everywhere, applied to the specific path ChatGPT takes to an answer. For the broader version of this, see how to get cited by AI.
The single most useful distinction to hold onto is which ChatGPT you’re trying to win. In training-data mode, there is no live retrieval — the model answers from what it absorbed during training, where your presence depends on having a broad, long-established footprint across many sources, not on any page you can optimize today. In browsing mode, it retrieves live and your current, well-structured content can be found and cited in real time. This is why AEO concentrates on the browsing path: it’s the one where deliberate work changes the outcome. A business with no training-data footprint can still be cited the moment ChatGPT browses, provided it’s retrievable, readable, and clearly answers the question.
When browsing, ChatGPT runs a search, retrieves candidate pages, and selects the ones that most directly and reliably answer the query — favoring clear, well-structured, corroborated content from sources it can read. Without browsing, it draws on training data, where presence depends on broad, established footprint.
It depends on the mode. With browsing or search enabled, ChatGPT retrieves and cites live web pages. Without it, responses come from training data with no real-time retrieval. AEO targets the browsing path, where current, well-structured content can be selected and cited.
Be retrievable and selectable: ensure crawlers and the search layer can read your content, answer the specific questions clearly and standalone, and build the corroboration that makes you a safe source. ChatGPT cites what it can find, read, and trust.
We test what ChatGPT actually returns for your buyers' questions and show you where in the search-retrieve-select path you're getting filtered out.