How to Rank in ChatGPT: What Actually Gets You Into the Answer

People ask me how to rank in ChatGPT the way they used to ask how to rank in Google, and the honest first answer is that the question is slightly wrong. There is no page one in ChatGPT. There is a paragraph, written on the spot, that names three or four options and describes each in a sentence. You are either in it or you are not, and if you are in it, the sentence is either accurate or it is not.

That changes the job. Ranking was a contest for position. This is a contest for inclusion, and then for framing. Here is how the assistant decides, and what moves it.

How ChatGPT chooses who to name

When someone asks ChatGPT for the best reputation firm, the best accountant in Boise, or whether a particular executive is credible, the model does two things. It draws on what it learned in training, which is a snapshot of the public web up to a cutoff, and, for most questions now, it runs a live web search and reads a handful of pages before answering. The answer is a blend: what the model already believes, corrected or confirmed by what it just read.

Three patterns show up consistently in the sources those answers cite.

First, entity consistency. The model is far more confident naming a company or person whose description is the same everywhere it looks: the website, LinkedIn, a Crunchbase or Wikidata entry, a couple of directory listings, a news mention. When those disagree, or when half of them are missing, the model hedges or skips you.

Second, direct answers. The pages that get quoted are the ones with a sentence that answers the question outright. “X provides Y for Z, based in W” gets lifted into an answer. A page that takes four paragraphs to get there does not, even if it ranks first in Google.

Third, independent agreement. A model weighs a third party saying you are good at something far more than you saying it. Comparison articles, “best of” lists, trade coverage, and review profiles are the sources that most often sit behind a category answer. When a competitor is in the answer and you are not, it is usually because they are on those pages and you are not.

What does not work

Keyword density does nothing. Buying links does nothing, and can get the linked page filtered. Publishing fifty thin pages does nothing, because the model is not counting pages, it is looking for the one sentence it can trust. Stuffing a page with “best” and “top” does nothing, and reads as exactly what it is.

Prompt tricks do not work either. You cannot inject instructions into the model by putting them on your website. The people selling that are selling something that was patched before they heard of it.

The order of operations

If I were starting with a brand or a person who is absent from the answer, I would do these in this order.

1. Ask the questions first

Write down the ten to twenty questions your buyers, critics, and reporters actually ask. Not keywords, questions: “who are the best X for Y,” “is Z trustworthy,” “what does X company do.” Run them through ChatGPT with web search on, and through Gemini, Perplexity, and Claude, because they differ. Record who is named, how you are described, and which sources are cited. That is your baseline, and everything after is measured against it.

2. Fix your identity everywhere it appears

Write one paragraph that says what you are, for whom, where, and who runs it. Put that paragraph, word for word, on your site, your LinkedIn, your Google Business Profile, Crunchbase, Wikidata, and any directory that matters in your field. Make the structured data on your site say the same thing. This is unglamorous, it takes an afternoon, and it is the single most common reason a model will name a competitor and not you.

3. Give every important page a one-sentence answer

For each question on your list, make sure some page on your site answers it in the first sentence of a section, plainly, with the question itself as the heading where that reads naturally. Then support it with the detail a human reader wants. This is the content the assistants quote. It is also, not coincidentally, good writing.

4. Get onto the pages the answer is built from

Look at the sources cited in your baseline. If the category answer draws on three comparison articles and a review platform, the work is to be present on those, accurately described. Sometimes that is outreach to the author. Sometimes it is a profile you never claimed. Sometimes it is a piece of coverage you have to earn. There is no shortcut here, and this is where most of the real effort in generative engine optimization goes.

5. Keep the record moving

The assistants prefer current sources. A site that has not changed since 2023 reads as abandoned. Publish something with a point of view regularly, keep dates and facts current, and retire pages that no longer describe you.

6. Re-run the questions

Monthly, same questions, same assistants. Record the changes. Answers drift when models update and when new sources appear, and the only way to know you have moved is to look. When a good answer degrades, trace it to the source that changed it and fix the source.

How long it takes

Faster than Google, usually. Because the assistants read live pages, a fixed identity and a well-written answer page can show up in responses within a few weeks. Getting into a competitive category answer, where you need the third-party sources to agree with you, is a three to six month build. Holding the position is the monitoring in step six, indefinitely, because the answer will not hold itself.

Where this is heading

Every month a larger share of the people who matter to you form their first impression from a generated paragraph rather than a list of links. The paragraph is written from the public record, and the public record is something you can shape, honestly, by being consistent, by answering the question, and by earning the sources that agree with you. That is the whole method. The firms that do it well treat the AI answer as a result to be measured, not a mystery to be feared.

If you want to see what the assistants say about you today, that is where every engagement we run starts, and the baseline is free.