The brand AI names first, and what put it there

On the morning of October 20 our system asked five AI platforms the same question within a minute. The question was "best merino wool sneakers under $150 for standing on your feet all day."

ChatGPT put Allbirds first. "My first pick for everyday long shifts," it said. It was also the only one of the five that cited allbirds.com.

Gemini put Allbirds first too.

Claude didn't. Its answer opened with "Best bet under $150: Giesswein Merino Runners (~$150)" and followed with "Runner-up: Allbirds Wool Runners (~$110)." Then came the line about Allbirds that we keep rereading. "These are the most popular pick, but they're less ideal for your use case."

Perplexity sided with Claude. Giesswein came first because "they have a structured heel and replaceable footbed." Allbirds was listed as another option.

Google AI Mode ranked nobody. It asked a follow-up question instead. "Would you prefer a shoe with thick, high-cushion foam (like Allbirds) or a flat, minimalist/zero-drop sole (like Softstar)?"

Allbirds was named in all five answers and came first in two.

The citations held one more detail. A review site, ecothes.com, was cited by three of the five platforms. Claude named it as the source of its reason for choosing Giesswein, which it summarised as "unlike a lot of wool sneakers, they come with a proper structured footbed and heel."

That one question shows most of what we found when we looked at thousands of them.

What we measured

Pharos tracks how ChatGPT, Gemini, Google AI Mode, Perplexity and Claude answer the questions a brand's buyers ask. For this piece we pulled every stored answer to a non-branded question, meaning one that doesn't name the brand, asked between July 22 and October 20, 2026.

That came to 27,125 answers to 1,086 questions, from audits of 40 companies. They include shoe and luggage brands, B2B software, dental and eye clinics, restaurants, a plumber in Canada, a dairy company and a property developer, in the US, the UK, Canada and the Gulf.

We audited these brands, so the sample leans toward companies that wanted to know where they stood. It's large enough to show patterns. Where a number rests on a small group, we say so.

Being named is the easy part

The brand was named in 7,595 of the 27,125 answers, about 28%. Most AI visibility reports lead with that number.

Then we looked at position. Of the 7,483 answers that named the brand and placed it in a list of recommendations, it came first in 2,529. That's 34%. It came second 29% of the time, third 17% and fourth or later 21%.

In two of every three answers that name a brand, a different business gets the first slot.

We expected the platforms to differ on this. They barely do. The share of named answers with the brand in first place ran from 32% on ChatGPT to 35% on Claude and Perplexity, with Gemini and Google AI Mode in between.

The business in first place surprised us more. Every audit carries a list of competitors the brand chose to track. When the brand was named but not first, a competitor from that list held first place in 47% of answers. In the other 53%, the business named first wasn't on the brand's list at all.

So more than half the time a brand loses first place, it loses to a business it wasn't tracking.

Five platforms, five opinions

The sneaker question split two, two and one. We wanted to know whether that was unusual.

It's the norm. We grouped answers by question and day, so each group is one question asked on every platform on the same day. In 330 groups, all five platforms put a tracked business first. They chose the same business in 116 of them, or 35%. In 86 groups, three or more different businesses took first place.

Look at a single brand and the spread gets wider. On 1,175 question-days, the brand was first on at least one platform. It was first on exactly one platform in 624 of those. It was first on all five 25 times, about 2%.

This could be random noise. The data says it mostly isn't. When we compared consecutive runs of the same question on the same platform, a few days apart, the same business came first 79% of the time. A brand that was first stayed first 84% of the time.

Each platform holds a fairly steady view of who should be first. The five views just differ.

For a marketing team, that means a screenshot of ChatGPT is a reading of ChatGPT. Perplexity may give the same buyer a different first pick on the same day.

What the first pick wins on

To get at the reasons, we classified answers that named businesses, one answer at a time, by the main thing each one used to choose between them. That covers 9,608 answers from 28 companies, using the most recent 30 days of answers for each.

Expertise or specialisation was the main basis in 45% of them. Reviews and reputation came next at 16%, then price or value at 13% and range or selection at 11%. The other four bases we tracked shared the remaining 15%.

Expertise led for 20 of the 34 brand audits with enough answers to judge, including every clinic and every software company. The exceptions match how those markets sell. In all three home-services audits, answers chose mainly on availability, meaning how soon someone could come out. Three large online retailers were mostly judged on price, and two UAE fitness studios on reviews.

"Specialisation" sounds abstract until you read an answer. On the sneaker question, Claude moved Allbirds to second place because the buyer would stand all day, and it said so in one sentence. The shoe it put first was the one a review site recommended for people who spend all day on their feet. Claude's whole case for Giesswein was about the footbed and the heel.

We can't count this next part, so treat it as our read from the answers we went through by hand. First place tends to go to the business described in terms closest to the condition in the question. In the questions we track, those conditions look like "under $150", "for men with wide feet", "for expat couples", "open Sunday" and "same day". A good reputation gets a brand onto the list. A specific claim that matches the buyer's condition, written somewhere the platform can read it, is what usually earns first place.

Some questions have no winner

On English questions phrased as "best", "top", "recommended", "cheapest" or "near me", the brand was named in 36% of answers. On "how do I choose" and "what should I look for" questions it was named in 14%. On other question forms, 9%.

Across all the answers we classified, 26% named no business at all. The platform explained the decision and left the choosing to the reader.

Check this before you read your own score. If many of your tracked questions are explainers, a low number may say more about your question list than about your brand.

Citing your own site

This is the finding we most wanted to hold, and it only partly did.

When an answer cited the brand's own website, the brand came first 41% of the time. When it didn't, 30%. The gap appeared on ChatGPT, Gemini, Claude and Perplexity. Google AI Mode showed no gap.

Then we compared the same question on the same platform, with and without the brand's site among the citations. Across 130 such pairs the difference was under one point. In 33 pairs the brand did better when cited, in 35 it did worse and in 62 nothing changed.

A citation of your site mostly marks questions where you already lead. We can't show that earning one moves you up where you're behind. And about 60% of the first places in our data came in answers that didn't cite the brand's own site at all.

What we'd do with this

• Count how often you come first, per question. A mention count treats every answer where someone else is first as a win.
• Check each platform on its own. In our data, platforms that named a first pick disagreed on it more often than they agreed.
• List the businesses that beat you to first place and compare that list with your competitor list. Expect names you've never benchmarked.
• For each question that matters, find the condition in it, such as the price cap or the use case, and make sure a page says in plain words that you meet it, and why. If your shoe is built for standing all day, write that sentence.
• Get described by people who aren't you. The reason Claude gave for its first pick on the sneaker question came from a review site.
• Sort your questions into "best X" and "how to choose" and judge them separately.
• Don't chase citations of your own site for their own sake.

Our free audit at heypharos.com runs your questions on all five platforms and shows who comes first on each. Start with the question your best customer would type the day before they buy.