How to rank in ChatGPT search: how it picks sources, and what you can change

How to rank in ChatGPT search: the short answer

You cannot rank in ChatGPT search the way you rank in Google, because there is no list of ten results to climb. What you can do is make your pages easy to find, easy to read and easy to quote at the moment ChatGPT goes looking.

ChatGPT writes one answer and links the pages it used. Ask the same question twice and the pages can change. So the useful question is not "what position am I in". It is "how often am I one of the pages it uses, and for which questions".

This guide walks through what OpenAI has said about how ChatGPT search works, what outside studies have measured, and which parts of the process a website owner can actually change. Every claim links to the page it came from.

Does ChatGPT search the web, or answer from memory?

It does both. ChatGPT answers some questions from what its model learned in training, and searches the web for others. Only the second kind can cite your page today.

OpenAI's help page says that "ChatGPT may search the web automatically when your question would benefit from current information", and that a person can also pick Search from the tools menu (OpenAI help, Searching the web with ChatGPT). Nobody outside OpenAI knows exactly when it chooses to search.

Outside estimates differ. Seer Interactive wrote in February 2025 that its own small study found close to 30% of questions triggered a search, and it reported that a larger Semrush study put the figure at 46% (Seer Interactive). Both figures are old now and come from different question sets, so treat them as a sign that search is common, not as a rate for your questions.

What happens when ChatGPT searches

ChatGPT turns your question into one or more search queries, reads the results, and writes an answer with links. That middle step, the rewrite, is the one most owners miss.

1. ChatGPT decides that the question needs current information, or the person turns on Search.
2. It rewrites the question into shorter search queries. OpenAI’s own example turns a long question about cancer drugs into "CCR8 immunotherapy drug development 2025".
3. It sends those queries to one or more search providers, along with a general location worked out from the IP address.
4. After reading the first results, it may send more specific follow-up queries to other providers.
5. It writes the answer and attaches citations. A Sources button lists the cited pages and other relevant links.

Every step above is from OpenAI's help page (OpenAI help). The same page lists the privacy policies of Microsoft and Shopify as search providers that may process those queries. Microsoft runs Bing, and Shopify runs online stores, so this tells you where some of the results come from. It does not say how much weight each provider gets.

ChatGPT does not search for your customer’s exact question. It searches for its own shorter version of it, so your page has to answer the short version clearly.

Location matters too. The help page says ChatGPT may rewrite "good restaurants near me" into "top restaurants San Francisco" when the IP address points there. If your business serves a place, name the place on the page in plain words.

Where the results come from

ChatGPT draws on third-party search providers and on content from partners, and one outside study found its citations overlap heavily with Bing’s results. That makes ordinary search visibility, in Bing as well as Google, the floor for ChatGPT visibility.

When OpenAI launched search in October 2024 it wrote that "ChatGPT search leverages third-party search providers, as well as content provided directly by our partners" (OpenAI, Introducing ChatGPT search). The partners it named on that page were news and data providers for weather, stocks, sports, news and maps.

Seer Interactive matched more than 500 citations from about 100 questions against Google and Bing results for the same questions. It found that "more than 87% of ChatGPT search citations matched Bing’s top 20 organic results", against 56% for Google (Seer Interactive). Seer also noted that many matches sat in positions 11 to 20, and that most of its questions used "best" or "top", which favoured affiliate, news and roundup sites.

Microsoft's guidelines say they cover "how Bing discovers, crawls, indexes, evaluates, and surfaces content across Bing search experiences, Copilot, and grounding API results" (Bing Webmaster Guidelines). Grounding means giving a model search results to base its answer on. Our page on how ChatGPT uses the web covers the same ground from OpenAI’s documents. OpenAI has not said ChatGPT uses that exact service, so read this as a strong hint rather than a fact.

What you can change

Five things are in your hands: crawler access, search presence, the shape of your pages, where else you are written about, and how current your facts are. The table puts them in the order to check them.

On access, OpenAI's crawler page is direct: "To help ensure your site appears in search results, we recommend allowing OAI-SearchBot in your site’s robots.txt file and allowing requests from our published IP ranges" (OpenAI, Overview of OpenAI Crawlers). A firewall rule that blocks unknown bots can undo a correct robots.txt. The full check is in our guide to getting your website to show up in ChatGPT.

On page shape, write the question your buyer asks as a heading, and answer it in the first sentence under it. Then give the detail. Microsoft's own guide for its AI search says assistants "break content into smaller, usable pieces" and warns against key facts placed only in images, tabs or PDFs (Microsoft Advertising). Put prices, sizes and dates in text so a model reading the page can quote them.

On third-party mentions, search your own "best" question in Bing and look at what ranks. If the top results are roundup articles, the route into ChatGPT’s answer may run through those articles rather than through your own site.

What does not work, or is not proven

Stuffing keywords, buying ads to change the answer, and tracking a single "position" are the three common mistakes. Each one has evidence against it.

• Keyword stuffing does not help. In the GEO study, adding more query words to a page gave little or no gain, and on Perplexity it did about 10% worse than doing nothing (Aggarwal and others, KDD 2024).
• Ads do not buy a place in the answer. OpenAI says "Ads do not influence the answers ChatGPT gives you" and that ads are labelled and shown apart from the answer (OpenAI, Testing ads in ChatGPT).
• A single position is the wrong measure. The answer is written fresh each time, so one check is a sample of one. Ask the same question several times and count how often you appear.
• Blocking OAI-SearchBot while hoping to be cited cannot work. OpenAI says opted-out sites can still appear as plain navigational links, but not in search answers.

The GEO study is worth reading with its limits in mind. It tested a research setup that fetched the top five Google results and wrote answers with an older OpenAI model, plus a smaller test on Perplexity. It did not test ChatGPT search itself. Its finding that adding sources, quotations and statistics helped is a useful direction, not a promise.

How to check where you stand

Write down the questions your buyers really ask, put each one to ChatGPT several times, and record who is named and which pages are cited. That gives you a count you can compare next month, which a single screenshot never will.

1. List 20 to 50 questions a buyer would type before choosing a product like yours. Leave your brand name out of most of them.
2. Ask each question in ChatGPT with search on, at least three times, in a fresh chat each time.
3. Record the brands the answer names and the pages it cites, for every run.
4. Count how often you were named and how often your pages were cited, and keep the count per question.
5. Repeat on the same day each week or month, with the same questions, so the counts can be compared.

Keep questions that name your brand apart from questions that do not. A buyer who already typed your name will usually see you. The questions that matter are the ones where they did not. Our guide to finding the questions buyers ask shows how to build the list.

Agentcites does this for you. It reads your website, writes the questions a buyer would ask, and puts them to ChatGPT, Perplexity, Gemini and DeepSeek on a weekly schedule. It keeps every answer word for word, records which brands were named and whose pages were linked, and reports branded and non-branded questions separately. How we measure explains each count, and the pricing page lists what the one plan includes.