Clients ask us how to rank in ChatGPT the way they used to ask about page one of Google, and the question needs a small correction before it can be answered. ChatGPT does not keep a ranked list of documents for a query. When it browses, it issues a search, pulls a handful of pages, reads them, and writes an answer that may or may not name the sources it used. Your goal is not a position number. Your goal is to be one of the pages that gets pulled, and then to be the page that is easiest to quote once it is open.
That split matters because the two halves reward different work. Getting pulled is still mostly classic retrieval: you need to exist in the search index the assistant is querying, and you need to look relevant for the phrasing a person actually types into a chat window, which is longer and more conversational than a keyword. Getting quoted is an editing problem. A model summarising four open tabs will lean on whichever one states the answer plainly, close to the top, in a sentence it can lift without rewriting. Most pages we audit fail the second half even when they pass the first.
How to rank in ChatGPT starts with retrieval, not prompts
There is no submission form and no secret tag. Browsing assistants ride on conventional web search, so the fundamentals still carry: crawlable pages, a clean sitemap, fast responses, titles that match the question, and enough external references that your domain is not a stranger to the index. If a page cannot be found by a normal search for the topic, no amount of formatting will get it in front of the model. Check that first, before you rewrite anything.
The second retrieval lever is query shape. People chat in full sentences. They ask what something costs, whether a tool works for a specific case, how two options compare. Pages built around a single two word keyword often do not surface for those phrasings at all. We usually find a gap between the keyword list a site was built on and the questions it is actually asked, and closing that gap is the cheapest win available.
Write so a model can lift one clean sentence
Open your page and read the first hundred words as if you were the assistant. Is the answer there? On most commercial pages the first hundred words are throat clearing: history of the industry, a promise that this article will explain everything, a paragraph about why the topic matters. A model skimming for something to cite finds nothing and moves to the competitor who led with the answer. Put the direct statement in the opening, then earn the reader’s time with the detail underneath it.
Structure helps for the same reason. Headings phrased as the question, short self contained paragraphs, a definition that does not depend on the sentence before it. When a claim needs a number, put the number and its unit in the same sentence, because a figure stranded in a table cell three sections away is hard to attribute and often gets dropped.
What we check before rewriting a page for AI answers
Our audit is short on purpose. Five checks catch most of the reasons a page never gets cited:
- The page answers its own headline within the first two sentences, without a preamble.
- Key facts are stated in prose, not only in images, tables or PDFs.
- Claims carry a date and a source, so the model has something to attribute.
- The page is reachable without JavaScript rendering and returns quickly.
- Author and organisation details are present and consistent across the site.
That last item does more than it looks like it should. Assistants hedge when the source is anonymous. A named author with a real role, a company page that matches the schema, and an address that lines up with the rest of the web make it easier for a model to say where the sentence came from, which is the whole point of the exercise.
Measuring something that has no rank tracker
You cannot check position seven. What you can do is ask your priority questions on a fixed schedule, from a clean session, and record which domains come back. Twenty questions checked monthly is enough to see movement. Keep the phrasing stable, and expect variance between runs.
Server logs are the other half. Assistant crawlers identify themselves, and a rise in those requests after a content push tells you retrieval improved even if citations lag. In our accounts the pattern is usually a few weeks between the crawl increase and the first citations, which is roughly what you would expect while the index catches up.
What this is worth, honestly
Citation traffic is small compared to search traffic for most businesses, and it converts differently. Someone arriving from an AI answer has usually had the basics explained already, so they land deeper into the decision and bounce faster from thin pages. Treat it as qualification rather than volume.
The good news is that almost nothing on this list is wasted if the assistants change how they retrieve. Clear answers near the top, sourced claims, fast crawlable pages and a real author identity are the same things that have always made a page worth ranking. Figuring out how to rank in ChatGPT mostly means doing the editing work that was easy to skip when a keyword and a backlink were enough.
Keep reading: GEO