How Gemini picks which companies to name

Gemini names the companies that appear in the pages it retrieved through Google for that question, favouring the ones whose page says clearly that they fit the category as asked. The retrieval step is where the decision is really made, and that step is a search.

What Gemini does differently

Because retrieval runs through Google, Google's ranking shapes the candidate set. Whoever is on the pages Google surfaces is in the running. Whoever is not is usually not considered, whatever the product is like.

It tends to show its links, so its working is visible. It also responds to the qualifiers in a question more visibly than some assistants: add a city or a customer size and both the links and the names change.

Freshness plays a part. A page updated recently, with a date on it, competes better on questions where the buyer clearly wants a current answer, and most buying questions are of that kind.

What is not public is how it chooses among the retrieved pages, or how much weight the model's own training contributes when the retrieved pages are thin. So read the links as evidence and everything else as inference.

What that means for whether you are named

There are two routes into the candidate set, and you need both.

Your own pages ranking. If your page is among the results for the question, your facts are available directly and you have some control over the sentence written about you.

Other people's pages naming you. When the pages Google surfaces are comparison articles and directory listings, your presence depends on being in them. This is the route most smaller companies actually win on, because outranking an established list page is harder than being added to it.

The uncomfortable version: if you rank nowhere for the list style queries in your category and appear on none of the pages that do, there is no path for Gemini to name you. That is a specific, fixable diagnosis rather than a mystery.

The five checks that explain most Gemini results

Run these in order on one question and you will usually know why the answer looks the way it does.

One, are your pages indexed. Search Console, not assumption. An unindexed page cannot be retrieved.

Two, what does Google return for the exact question. Those pages are the likely candidate set.

Three, are you named on any of them. Open each one and search for your name.

Four, do the links Gemini showed match the Google results. High overlap means your problem is a ranking and presence problem. Low overlap means it reached something else, which is worth watching over months.

Five, does the question carry a qualifier you can claim. A city, a company size, an industry. Those are the questions where a smaller company actually gets named.

A worked example

The clearest case we have of an ordering decided by what a page publishes came from a Claude run on 17 September 2026, in the category of clawlaw.in, our sister company.

On a question about finding every case filed against a company, two enterprise vendors were placed above it, specifically because they publish explicit lists of the courts and tribunals they cover. The assistant noted that its ordering reflected price transparency and source authority rather than product quality.

That is the retrieval step doing what this page describes. Whoever is on the pages that surface is in the candidate set, and among those candidates the one whose page states its fit plainly, with a coverage list and a price, is the one that gets named. We recorded it on Claude, so read it as the mechanism behind a retrieved answer rather than as a Gemini result.

Two routes follow, at different prices. Publish the checkable version of your own fit: the list of what you cover, the price, and the date each was last reviewed. And get named on the pages already surfacing for the question, which is a week of factual emails rather than a quarter of content.

What this does not explain

Four gaps, stated so that nothing here reads as more certain than it is.

The grounding logic is not published. Nobody outside Google can say how the retrieved pages are weighed.

Why one answer differs from the next. Answers vary between runs, so a single result is an example rather than evidence.

How much comes from training rather than retrieval. When the retrieved pages are thin, the model's own knowledge appears to fill in, and there is no way to measure the split from outside.

How this surface relates to Google's AI answers in search. They are related and they do not always agree. Measure them separately rather than assuming one predicts the other.

And the standing limit: no guarantee of a position is available from us or anybody else.

Common questions

Does Gemini only use Google results?

Its grounding leans on Google's index, and that is the most useful working assumption. It is not a documented rule about every answer, and you will occasionally see a link that Google does not rank highly. Note those and watch them rather than building a plan on them.

Is it easier to be added to a list page than to outrank it?

Usually yes, and much faster. An established comparison article took years to earn its position. Being added to it takes an accurate email. Do that first and treat outranking it as the longer project.

Why does adding a city to the question change everything?

Because the search behind the answer changes, so the candidate pages change. This is why a question set should carry the qualifiers your buyers use, and why a general question tells you less about your position than a specific one.

Do I need to rank first?

No. You need to be on the pages that are retrieved, which can be somebody else's page. Ranking your own page is one route and it is not the only one.

How do I know whether my page was used at all?

Look at the links shown. If your page is not among them, it was probably not part of the answer, whatever else is on your site. That is a cleaner signal than most assistants give you and it is worth logging every month.

What if a retrieved page describes us wrongly?

Treat that as urgent. Write to the page owner with the correct facts and a link to the page on your site that states them, and make sure your own pages say the same thing in the same words.

What to do first

For each tracked question, log three things: whether you were named, the order of the names, and the links Gemini displayed. Then open Google, run the same question, and compare the two lists of pages.

Where the lists overlap and you are absent from those pages, you have found your work. Start with getting added to the retrieved pages you are missing from, which is the cheapest move available, and check Search Console for any of your own pages that are not indexed.

Read direction over several readings rather than the jump between two, because the same question can answer differently on the same day.

How we know this

We read the assistants rather than guessing what they do. In our own audit of 1,260 saved answers from ChatGPT and Gemini across the United States, the United Kingdom and India, 82.0% of every citation landed on a site the business itself owned, and review directories took 0.7%. Counted as one domain per reading, 16,516 of 20,143.

That is the whole asymmetry: the pages you own are the pages an assistant quotes back to a buyer. See the dated readings and the method on our proof page.

Find out whether assistants name your business, on your own domain and your own buying questions.

By Aditya Goel, Founder, AI Knows Us. Reviewed 1 November 2026.