How to rank in ChatGPT: what actually gets a brand cited

What ranking in ChatGPT actually means

There is no results page. When someone asks ChatGPT which CRM suits a ten-person agency, they get a paragraph, a few named products and, when browsing is on, a row of source links. Ranking here means two things: your brand is named in the answer, and a page you control, or a page that describes you fairly, is among the sources. Both are measurable, and neither is guaranteed by classic search rankings. We have watched brands on page one of Google go unmentioned while a smaller brand with one well-written comparison page gets cited.

We do not know how OpenAI ranks anything, and nobody outside the company does. What follows is editorial observation from asking buyer questions and looking at which pages get pulled, plus the general things engines have said publicly: they retrieve from the live web, they cite what they retrieve, and they prefer sources they consider reliable. The models behind those answers change often, too; our note on GPT-6 Sol and Luna covers the latest release and what it changes for marketers.

How the answer gets built

The useful mental model is a three-step pipeline. The model rewrites the question into one or more search queries. A retrieval layer fetches a small set of pages for those queries. The model then reads those pages and composes an answer, citing the ones it leaned on. That means you are competing twice: first to be retrieved at all, then to be the page whose sentences the model finds easiest to lift.

The first competition looks a lot like ordinary SEO. Crawlable pages, a sensible title, an accurate description, no login wall, nothing rendered only after a click. The second is where ChatGPT SEO diverges from the Google playbook, because the reader is a model that wants a clean statement of fact it can quote.

Make the page the obvious answer

Pick the exact question a buyer asks and answer it in the first hundred words. Not a preamble about why the question matters, the answer. If the question is which tools do X for teams of size Y, name the tools and say what they cost.

Then structure the rest so a machine can skim it. Headings that are questions or plain labels. Short paragraphs that each make one claim. Numbers with units and dates. Named entities rather than pronouns: write the product name, the company, the integration, because a sentence with a named subject is far easier to attribute. In our observation the pages that get cited most are not the longest ones, they are the ones where any single paragraph makes sense out of context.

Keep the date visible and honest. Engines are asked "in 2026" constantly, and a page whose last update is stated at the top gets picked over one that looks stale. Update the content when you update the date; a fresh stamp on old text is easy to spot when the facts inside contradict it.

Be described the same way everywhere

Models build a picture of a brand from every page that mentions it. If your homepage says you are an AI visibility platform, your LinkedIn says growth analytics, your G2 listing says social media management and a press release says ad tech, the model has four weak signals instead of one strong one. Write a single two-sentence description of what you do and for whom, and use it verbatim on your About page, your social profiles, your directory listings and your partner pages. Consistency is not glamorous, but it is the cheapest lever on this list.

The same goes for facts about you: founding year, headquarters, pricing tiers, integrations. State them plainly on one page you own and keep them current.

Earn mentions where the engine already looks

Ask ChatGPT a category question and look at what it cites. In most software categories it is a familiar mix: review sites, "best X" comparison posts on established blogs, category directories, and occasionally trade press. Those pages get retrieved first, so a mention there is worth more than another article on your own domain. Practical steps: keep your review-site profiles complete, ask real customers for reviews there, pitch comparison writers with accurate information rather than a request for a link, and make sure directory entries carry that same consistent description.

Your own comparison content still matters, especially for questions where nobody has written a fair one yet. Write it honestly, including the cases where a competitor is the better choice; those are the pages the model treats as an answer rather than an advertisement.

Structured data and a clean About page

Organization, Product and Article schema will not put you in an answer on their own, but they remove ambiguity about who published a page and what it is about. Add them, fill in sameAs with your real profiles, and keep the logo and name consistent with the description you settled on. Then read your About page as if you were a model trying to summarise your company in one line. If it cannot, rewrite it until it can.

Measure it like a channel

None of this is worth doing blind. Write down the twenty questions your buyers actually ask, in their words, and ask them on a schedule. Record whether you are named, in what position, how you are described, and which pages get cited. Watch the trend, not the single answer; these engines vary from run to run.

This is the job Signals is built for. It is designed to ask Gemini and Perplexity the questions buyers ask, then report visibility, position, sentiment, share of voice and the sources cited, so the work above has a scoreboard. Its AI-answer scan is switched off on the live product while we finish it, and it does not cover ChatGPT, though the same practices apply there. If you want the questions asked weekly without a person doing it, you can build a workflow that calls an engine's API on a schedule with your own saved key and sends you the answers.

What not to do

Do not stuff pages with "best" and "top" in the hope of matching prompts. Do not publish hundreds of thin question pages; a model reading them sees the same absence of substance a person does. Do not hide instructions aimed at the model inside your content. None of it has worked in our observation, and some of it damages the search traffic you already have. Answer the question, be described consistently, get mentioned where it counts, and measure.