Does ranking first in Google mean an assistant will name you?

What do the assistants actually retrieve?

Each assistant has its own route to the web, and only some of those routes pass through Google. Before asking whether your ranking helps, it is worth knowing which index each one reads.

Gemini can ground an answer in Google Search. Google’s developer documentation for groundingTying an answer to retrieved sources, so the model writes from them rather than from memory. with Google Search describes the flow: the model decides whether a search would help, writes one or more queries itself, runs them, and builds the answer from the results with citationsA source an assistant links to under or inside an answer. A mention names you; a citation points to a page. attached. The index is Google’s; the queries are the model’s.

Google’s AI Overviews and AI Mode draw on the same index. Google’s guidance on AI features and your website says a page needs to be indexed and eligible to show with a snippet to be a supporting link, and that there are no extra requirements beyond that. It also says both features may use “query fan-out” — running several related searches across subtopics while the answer is built.

ChatGPT uses OpenAI’s own search crawler. OpenAI’s crawler documentation names OAI-SearchBot as the bot that decides which sites can appear in ChatGPT’s search features, and states that sites which block it will not be shown in ChatGPT search answers. A separate bot, GPTBot, collects material for training. Your Google position plays no part in either.

Perplexity also runs its own crawler. Its bot documentation says PerplexityBot exists to surface and link websites in Perplexity’s results, and recommends allowing it in robots.txt if you want to appear there.

Claude searches only when it judges that the question needs it. Anthropic’s documentation for the web search tool says Claude decides when to search based on the promptThe question as it is actually typed. Two wordings of the same intent can return different names., may search several times in one request, and answers directly without searching when the question draws on stable knowledge.

The honest reading of this table: a Google ranking is a direct input to one family of products, and an indirect one at best for the rest.

Does the assistant search for the keyword you rank for?

Usually not. The buyer types a sentence, and the assistant turns it into its own searches — sometimes one, often several.

Both Google documents above describe this: the Gemini model “automatically generates one or multiple search queries”, and AI OverviewsGoogle's generated summaries above the ordinary results. They name the sources they used, which makes them the one surface where citations are visible. and AI ModeGoogle's conversational search surface: you ask, it answers, and the next question continues the same thread. fan a question out into related searches across subtopics. None of the five assistants publishes the queries it writes for a given question.

That matters more than it sounds. You rank for invoicing software for freelancers. A buyer asks what should I use to send invoices if I bill clients in three currencies and hate spreadsheets? The searches behind that answer might be about multi-currency invoicing, about tools reviewed for small businesses, about comparisons. Your page could rank first for none of them.

So the first limit on a ranking is simple: it is a ranking for a phrase, and the assistant is free to search for different phrases. The comparison between keyword thinking and question thinking is laid out in GEO vs SEO.

Where does ranking still help?

Wherever the assistant searches Google, and wherever your page is the reference copy of your own facts. That covers more ground than the sceptics say.

Getting onto the shortlist. Gemini and Google’s AI features read search results. A page that ranks for the searches they write is more likely to be fetched than a page on page five. You cannot be quoted from a page nobody read.

Proving the page is readable. Ranking well already implies the basics that every engine needs: the page can be crawled, renders without JavaScript gymnastics, has a clear subject and loads. Google’s own guidance for AI features says it needs nothing beyond what search already requires, which is another way of saying a well-ranked page has cleared the technical bar.

Being the source others repeat. A page that ranks gets found by the people who write directories, roundups and comparison articles. Those pages then describe you to the assistants that never searched Google at all. This path is indirect and slow, and it is the only one by which a ranking reaches an assistant that does not read Google.

Gemini in particular. Of the four assistants, Gemini is the one whose documented search route runs through Google, so it is where a ranking has the most direct path into an answer. The comparison between engines is in Gemini and Claude visibility, and the Google surfaces have their own article on AI Overviews optimization.

Where does ranking stop helping?

At the point where the assistant stops reading Google, stops reading pages, or starts reading other people’s pages about you. There are four such points.

When the assistant uses a different index

ChatGPT search and Perplexity run their own crawlers. A first-place Google ranking means nothing to an index that never fetched your page, and a site that blocks OAI-SearchBot or PerplexityBot in robots.txt is invisible to that index regardless of how it ranks elsewhere. This is the cheapest gap to close and the one most often left open by accident, usually by a blanket rule written to keep AI training bots out.

When the assistant answers from memory

Claude’s documentation is explicit that it answers without searching when a question draws on stable knowledge. Plenty of buying questions look like that to a model: what is a good accounting package for a small agency? reads as general knowledge, not current events. In that case the answer comes from what the model learned in training, and today’s ranking is irrelevant to it. What matters is how often, and how consistently, your business was written about before the model was trained.

When the answer is about a business, not a page

Search ranks pages. An assistant names businesses. In one local market we measured, a single assistant answer rested on eleven pages on average — the business’s own pages alongside directories, review sites, roundups and trade listings. The run is published in full.

A ranking speaks for one of those eleven. If the other ten describe a competitor more completely, the competitor gets named, and your first place in search has done its whole job — got your page read — without changing the outcome.

When being first on the shortlist stops counting

Even when an assistant reads Google’s top results, the order of those results does not carry over into the answer. Research on this is still thin, but one paper tested it directly.

Why can a lower-ranked page beat the first one?

Because once a page is in the set the model reads, what it says matters more than where it ranked. That is the finding of the clearest experiment on the question so far.

The researchers behind the Generative Engine Optimization paper built a test engine that fetched the top five Google results for each query and had a language model write an answer from them. They then rewrote pages in different ways and measured how visible each source was in the answer.

The result that matters here: lower-ranked pages gained much more from the rewrites than top-ranked ones. When the page in fifth place added citations to credible sources, its visibility in the answer rose by 115.1%, while the top-ranked page lost visibility under the same treatment. In the authors’ words, “lower-ranked websites, which typically struggle for visibility, benefit significantly more”.

Read that with its limits. It is a research setup from 2023, with an older model and a fixed shortlist of five, not a measurement of any of the assistants your buyers use. What it shows is the mechanism: inside the shortlist, position in Google is not a strong advantage, and content that is easy to quote can overtake it.

What is a first-page site usually missing?

Usually not anything on the page that ranks. The gaps tend to sit around it.

Answers in the buyer’s words. Pages that rank are often built around a keyword and build up to their point. An assistant extracting two sentences needs the answer at the top, phrased the way the question is asked. A pricing page whose prices sit in a table three screens down ranks fine and gets quoted rarely.

Coverage of the questions behind the keyword. Price, comparison, urgency, fit for a specific kind of customer. A site that ranks for the head term and says nothing direct about these is absent from exactly the searches an assistant writes when the buyer asks a real question. The question types are explained in what AI visibility is.

An off-site description that matches the site. This is the largest gap and the one SEO reporting never shows. Directories with an old service list, review platforms with no description, roundups that do not mentionYour business named in the text of an answer, with or without a link to you. you at all. Strong pages and a thin or contradictory off-site record is the most common shape of “we rank well and assistants ignore us”.

Access for the other crawlers. robots.txt rules written to block AI training sometimes block the search crawlers too. GPTBot and OAI-SearchBot are different bots with different jobs, and OpenAI documents them separately for that reason.

Something specific enough to cite. Concrete numbers, stated prices, named services, a dated method. The GEOGenerative engine optimization: the work of being named inside answers that assistants write, instead of ranked in a list of links. paper above found that adding citations to credible sources helped visibility; the general principle is that a page saying something checkable is easier to quote than a page saying something impressive.

What should you check on your own site?

:::checklist title=“Five checks, in order of cost, and the first two take minutes”

• Read your robots.txt. Confirm that OAI-SearchBot and PerplexityBot are allowed if you want to appear in ChatGPT search and Perplexity. If you block GPTBot to keep content out of training, make sure the rule does not catch the search bots as well.
• Check snippet controls. Google’s guidance makes snippet eligibility the condition for being a supporting link in AI Overviews and AI Mode. A nosnippet directive on a page you want quoted removes it from that shortlist.
• Take your best-ranking page and read its first two sentences. Do they answer the question a buyer would ask, in the buyer’s words? If the answer is under a heading halfway down, move it up.
• Write out the questions behind your main keyword. Price, comparison, urgency, fit. Find the page on your site that answers each one directly. If there is none, that question is one you cannot be quoted for.
• Ask the assistants and read what they cite. Ask your buying questions in Perplexity, where sources are always shown, and in Google with AI Overviews. List every site cited. Then check whether you appear on each one, and whether what it says matches your own pages. :::

The fifth check is the one SEO tools cannot do for you, and it is where most of the remaining work is. The ordered version of that work is in how to improve AI visibility.

An illustrative example

The figures below are illustrative. They show the shape of the problem rather than reporting a result anyone measured.

Say the search side looks like this. An online bookkeeping service ranks first for its main commercial keyword and has done for a year. Organic traffic is steady. By every SEO measure, the account is healthy.

Say the assistant side looks like this. Named in roughly half of general recommendation answers on Gemini, a fifth on ChatGPT, and almost never when the question is about price or about a specific kind of business.

What the checks would find. robots.txt blocks every OpenAI bot, because a rule was added to keep content out of training. The pricing page puts its prices in a table at the bottom. The two comparison sites that Perplexity cites for price questions list the service with an old plan structure.

What the work would be. Allow the search crawlers. Move the answer to the price question into the first two sentences of the pricing page. Ask the two comparison sites to correct the listing, using the plan names from the service’s own page. Nothing about the ranking would change, and nothing needs to.

What movement would look like. Gemini drifting a few points either way, which is no change. ChatGPT rising over the following weeks once the crawler can read the site. Price answers moving from almost never to occasionally, on Perplexity first. And a report that states what was changed and when, without claiming the changes caused the movement.

Should you change your SEO because of this?

:::myth No. Keep it, and add the part it does not cover. The pages that rank are the raw material for every assistant that reads the web, and the technical work behind them is the same work these engines need. :::

What changes is what you report and where you spend the next hour of effort. A ranking report tells you how one family of products sees one page. It does not tell you whether ChatGPT can read your site, whether Claude learned about you before its training cut-off, or what the directories in your category say. Those need to be looked at directly, and reported separately from search. The full comparison of the two disciplines is in GEO vs SEO.

Questions people ask

Does SEO help with ChatGPT? Partly. The groundwork carries over: pages that can be crawled, read and quoted. But ChatGPT search runs on OpenAI’s own crawler rather than on your Google position, and the answer weighs what other sites say about you, so a strong ranking does not predict a mention.

Do AI assistants use Google rankings? Gemini and Google’s AI Overviews and AI Mode search Google, so a good position makes your pages more likely to be read there. ChatGPT search and Perplexity run their own crawlers, and Claude decides per question whether to search at all. None of them copies the Google order into its answer.

I rank first. Why does the assistant name someone else? Usually because the assistant searched for different words than you rank for, or because the sites it read about your category describe a competitor more completely than they describe you. A ranking gets a page read; it does not decide which business the answer names.

Where to start

Find out where the two pictures disagree before deciding anything. Ask your three most valuable buying questions in each assistant, several times each, and compare the businesses named with the ones that rank.

If you would rather not do that by hand, the free AI visibility check asks the questions for your business across four assistants and emails you what came back, per assistant, with no call attached.