TL;DR: ChatGPT names brands for two different reasons, and they need two different fixes. When it runs a web search, it names whatever the retrieved pages name, so content and third-party placement work. When it answers from memory, no page you publish will ever be read, so only broad presence moves the needle: reviews, directories, and being written about. Find out which mode your buyers' questions trigger before you spend a dollar.
The short answer
There is no ranking to climb. ChatGPT assembles an answer from two sources: what the model already knows, and what it retrieves at question time. Getting named means winning one of those, and they are not the same game.
Mode one: the model searched
When a question is current, specific, or commercial, ChatGPT runs a search and reads a handful of pages before answering. Those pages decide the answer. If a Reddit thread, a G2 category page, and two listicles get pulled in, whoever those pages name is who ChatGPT names.
This is the winnable mode, and the work is concrete:
- Be on the pages that get retrieved. Not any pages, but the specific ones. You can only know which by measuring.
- Publish content that answers the question directly in the first paragraph. Retrieval favours extractable answers over throat-clearing.
- Get comparison and alternatives pages of your own live, because those are the shapes that get pulled for high-intent questions.
Mode two: the model already knew
Ask "what's a good CRM" and the model often will not search at all. It answers from training. Nothing you publish this month will appear, because nothing is being read.
The lever here is presence at scale over time. Review volume on the sites that got scraped, directory listings, mentions in articles other people wrote, a Wikidata entity. Slower, less satisfying, and the only thing that works.
| Retrieval mode | Knowledge mode | |
|---|---|---|
| Triggered by | Current, specific, commercial questions | Broad, evergreen questions |
| What decides the answer | The pages retrieved right now | What the model absorbed in training |
| What works | Content, placement on cited pages | Reviews, directories, third-party mentions |
| Time to impact | Days to weeks | Months |
| What is wasted | Nothing | Every article you publish for it |
The mistake almost everyone makes
Publishing articles at a question the model never searches for. The article is fine. It ranks, even. It is simply never retrieved, because retrieval never happens for that question.
This is why measuring whether a search actually fired matters more than measuring your score. A tool that reports "you are absent from this prompt" without telling you whether a search ran has handed you a number you cannot act on.
What I would actually do first
Speaking plainly, from watching this across a lot of brands: get your review count up before you write anything. It is the least fun item on the list and it outperforms everything else, because it feeds both modes at once. Review pages are the most-retrieved sources in retrieval mode, and review volume is a large part of what the model absorbed about your category in the first place.
Second, check whether your own site is readable. A surprising number of sites block GPTBot in robots.txt without anyone having decided to, it arrives via a security plugin or a default. If your pricing page is blocked, the model will keep saying your pricing is unclear, forever.
Third, and only third, write something.
Key takeaways
| Do this | Why |
|---|---|
| Measure whether a search fires per question | Decides whether content can ever work |
| Fix crawler blocks | Cheapest, fastest, highest certainty |
| Build review volume | The only lever that moves both modes |
| Get onto the pages that get cited | Beats publishing your own page nobody retrieves |
| Publish answer-first content | Extractable beats eloquent |
FAQ
Does ChatGPT use backlinks?
Not as a ranking factor the way Google historically did. See does ChatGPT use backlinks.
How long does it take to show up?
Retrieval-mode wins can land in one to three weeks. Knowledge-mode shifts take months and follow review and mention volume.
Can I just pay to be included?
No, and anyone selling that is selling a directory listing with a markup. What works is being genuinely present on the sources the model reads.
How do I know which questions matter?
Start from what you already get search traffic for, then expand into the comparison and alternatives phrasings buyers actually type. That is what prompt tracking is for.