The common fear about AI search is that it will collapse demand onto a handful of giant brands — that assistants will recommend the same five companies forever and everyone else disappears. We had the data to check, so we did. Across 22,139 AI answers from ChatGPT, Claude and Gemini, we counted 212,758 brand mentions. They were spread across 30,462 different companies.
Table of Contents
The top ten get 2.9%
That is the headline. The ten most-recommended brands in our entire sample account for just 2.9% of all brand mentions. It takes roughly 10,000 companies to reach 71%.
| Brands | Share of all mentions |
|---|---|
| Top 10 | 2.9% |
| Top 100 | 10.9% |
| Top 1,000 | 30.5% |
| Top 10,000 | 71.0% |
Compare that to a Google results page, where ten positions absorb effectively all the attention and the top three take most of it. On recommendation behaviour, AI assistants are dramatically less winner-take-all than the search results they are replacing.
The reason is mundane once you see it. Most questions people ask are specific, and the genuinely useful answer to a specific question is a specific product — not a famous one. Asked for “SMS tools for a dental practice,” an assistant that names Twilio and stops is less useful than one that names something built for dental practices.
Why Perplexity is not in these numbers
This is a study of brand recommendations — which companies get named inside an answer, read directly from the answer text. Perplexity works differently: it returns a separate list of sources, and only 6.7% of those URLs appear anywhere in its answer. Counting them alongside the others would compare a citation list against in-prose mentions and inflate every figure above.
So Perplexity gets its own report, and it produced a finding we did not expect: it is the only one of the four that cites sources at all. Perplexity marks its claims in 95.1% of answers. ChatGPT does so in 0.0% — two answers out of 6,015. That is covered in the citation study.
Keeping the two apart matters. Citation research consistently finds high concentration among a small set of reference sites; recommendation behaviour spreads across tens of thousands of brands. Both are true. Mixing them produces numbers that look like they contradict each other when they simply measure different things.
The three assistants are not one channel
Claude names 14.0 brands per answer. ChatGPT names 4.5 — a 3.1x difference in how many companies get any visibility at all, on identical prompts.
That gap carries straight through to you. A site we tracked was named in 10.3% of Gemini answers and only 2.0% of Claude answers — a 5.2x spread. If you are measuring “AI visibility” as a single blended score, you are averaging systems that share little behaviour, and the resulting number will move for reasons you cannot act on.
What we would actually do with this
Find out which assistant your buyers use before optimising for any of them. The 5.2x spread means this decision dominates everything downstream.
Compete on specificity rather than size. With 30,462 brands named and the top ten taking 2.9%, depth on a narrow topic is a realistic route in. Being the obvious answer to a precise question beats being a famous answer to a broad one.
Treat it as brand reach, not traffic. Many AI answers resolve the question without a click. Being named is worth having, but it will not show up in your organic click numbers, and measuring it there will make it look like a failure.
Read the full studies
Both reports cover methodology in full, including what the data does not show — why we make no claim about change over time, and how our keyword sample skews the results.
Which Brands Do AI Assistants Recommend? — ChatGPT, Claude and Gemini, 22,139 answers.
Only One AI Assistant Shows Its Sources — Perplexity, 362,179 citations.
The figures are free to cite with a link. If you need a specific cut of the data for something you are writing, get in touch and we will run it.
Written by
ranklytics