How an AI answer gets made, and why your rankings do not carry over

When someone types "best CRM for a small business" into Google, you get a list of ten links and your job is to be one of them. When they type the same thing into ChatGPT, they get a paragraph that names three or four companies, and your job is to be one of the names. Those are different jobs, and the second one is decided somewhere you probably have not looked.

The sequence

For a buying question, most engines now do roughly the same thing before writing a word:

1. Rewrite the question into two or three searches. "Best CRM for a small business" becomes something like "small business CRM comparison 2026", "CRM pricing for small teams", "HubSpot vs Pipedrive small business".
2. Search, using their own index or a partner's, and take the top handful of results for each. Not ten pages. Usually four to eight.
3. Read those pages, or the parts of them that survive being stripped to text.
4. Write the answer from what those pages say, and cite some of them.

The names in the answer are, almost entirely, the names that appear across those handful of pages. If g2.com's "Best CRM Software 2026" is read and you are on it, you tend to be in the answer. If it is read and you are not, you tend not to be. That is the whole mechanism, and it is why a site can rank fourth on Google for the same query and be absent from the answer.

What this means in practice

Three things follow, and they are the three places you can act.

The pages that get read are a small, stable set. For any given category the same dozen domains come up over and over: review aggregators, one or two publications, a Reddit thread, the incumbents' own comparison pages. You can find out which ones by asking the engines your buyers' questions and looking at the citations. Once you know the set, "get named on those pages" is a concrete list of five or six outreach tasks, not a strategy.

Your own pages have to be readable as text. Engines fetch a page and strip it. If your pricing page is a JavaScript application that renders nothing without a browser, the engine reads a blank page and moves on to a competitor whose pricing page is HTML. If your product page is a paragraph of positioning with no plain sentence saying what you are and who it is for, the engine has nothing to lift. This is boring and fixable and it is where most of the quick wins are. (We built a free readiness check that reads twelve of your pages and lists exactly this.)

Blocking the crawler is blocking the answer. A surprising number of sites disallow OAI-SearchBot, PerplexityBot or Google-Extended in robots.txt because someone added a blanket rule against AI in 2023. The training bots and the answer bots are different agents; blocking the first is a policy choice, blocking the second removes you from the answers entirely. Check yours: it takes two seconds and the fix is four lines.

What does not carry over

Backlinks as a number, domain authority as a number, and most of the on-page ritual: none of it is read by an engine writing an answer. What carries over is the part of SEO that was always about being genuinely present where buyers look: being on the pages people trust, saying plainly what you are, and being reachable. AI search did not invent a new discipline. It removed the layer between the buyer and the pages, and it removed your ranking with it.

How to see where you stand

Take five questions a buyer would ask before choosing you. Ask each one in ChatGPT, Perplexity and Gemini. Write down who gets named and which pages get cited. That is your visibility, by hand, and it takes twenty minutes.

If you would rather not do it by hand every week, that is what we do, every run, on every engine, for the questions you choose. Either way, the mechanism above is what you are working with.