Original research · September 2026
Which Brands Do AI
Assistants Recommend?
We analysed 22,139 AI answers from ChatGPT, Claude and Gemini and recorded every brand named in them — 212,758 brand mentions across 30,462 distinct companies. The result is the opposite of what most people expect.
of brand mentions go to the ten biggest names
distinct brands recommended at least once
more brands named by Claude than ChatGPT
difference in your odds of being named
Methodology
What We Measured — And What We Didn't
Ranklytics runs live visibility checks for customers: we send a real prompt to each AI assistant, capture the full response, and record every domain named in it. This study aggregates 22,139 such checks run between February and September 2026 across 619 websites and 7,366 keywords.
This is a study of brand recommendations, not source citations. The distinction matters and is easy to blur. When ChatGPT answers “best SMS marketing tools” by naming Twilio, TextMagic and EZ Texting, those are products it is recommending — not sources it consulted to ground a factual claim.
Excellent research already exists on citations — which sources AI systems draw on to build an answer — and it reports far higher concentration than we do here. That is not a contradiction. It is a different question. Citation studies ask where did the answer come from; this study asks whose product got named in it.
Perplexity is deliberately excluded from this report. For ChatGPT, Claude and Gemini we extract brands from the answer text itself. Perplexity instead returns a separate list of sources, and only 6.7% of those URLs appear anywhere in its answer. Counting them together would compare a citation list against in-prose brand mentions and inflate every cross-assistant figure. Perplexity is covered in a separate citation study.
What we counted
- Every domain named in the response body
- Deduplicated within a single answer
- Reference and platform domains separated out (10.9% of URLs) — Wikipedia, Reddit, app stores, government and social sites are not brand recommendations
- Google's vertexaisearch redirect wrapper excluded as an artifact
- Raw counts, no weighting
What this is not
- Not a citation study — see above
- Not a random sample of AI queries; keywords are customer-selected and skew commercial and SMB
- 94.6% of checks ran in August 2026, so we claim no trend over time
- Assistants were queried at different volumes
- Answers vary run to run; this is a snapshot
Finding 01
AI Recommends the Long Tail, Not Just the Leaders
Across 212,758 brand mentions we counted 30,462 distinct companies named at least once.
The ten most-recommended brands account for just 2.9% of all mentions. It takes around 10,000 brands to reach half.
The common fear is that AI answers will collapse demand onto a handful of dominant brands. In recommendation behaviour we see the opposite: assistants name an enormous variety of companies, because most questions are specific and the useful answer is a specific product rather than a famous one.
For a smaller company this is the encouraging read. Being named in an AI answer for a specific, well-covered topic is a more realistic goal than a competitive first-page ranking — though a considerably less stable one.
Finding 02
The Four Assistants Behave Completely Differently
Treating “AI search” as a single channel is the most common mistake we see. On every measure we tracked, the four assistants diverge sharply.
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.
| Assistant | Answers analysed | Brands per answer | Distinct brands named | Named the tracked site |
|---|---|---|---|---|
| Claude | 7,194 | 14.0 | 14,428 | 2.0% |
| Gemini | 8,930 | 9.5 | 17,832 | 10.3% |
| ChatGPT | 6,015 | 4.5 | 4,761 | 2.5% |
The last column matters most if you want to appear in AI answers. A site in our sample was named in 10.3% of Gemini answers but only 2.0% of Claude answers — a 5.2x spread on identical prompts.
Measuring “AI visibility” as one blended number averages systems that share little behaviour, and produces a figure that moves for reasons you cannot act on.
Finding 03
The Most-Recommended Brands
Share of answers naming each brand, with the per-assistant split. Read the columns rather than the blended figure — the assistants rarely agree.
| # | Brand | All | Claude | Gemini | ChatGPT |
|---|---|---|---|---|---|
| 1 | mayoclinic.org | 4.2% | 6.4% | 1.8% | 5.1% |
| 2 | semrush.com | 3.8% | 6.2% | 2.9% | 2.3% |
| 3 | ahrefs.com | 3.2% | 5.3% | 2.3% | 2.2% |
| 4 | healthline.com | 3.0% | 4.1% | 1.6% | 3.6% |
| 5 | moz.com | 2.9% | 4.7% | 1.4% | 3.1% |
| 6 | canva.com | 2.7% | 5.0% | 2.0% | 0.9% |
| 7 | hubspot.com | 2.4% | 3.6% | 1.7% | 2.0% |
| 8 | angi.com | 1.9% | 3.1% | 1.1% | 1.7% |
| 9 | my.clevelandclinic.org | 1.9% | 2.0% | 2.2% | 1.3% |
| 10 | adobe.com | 1.8% | 2.9% | 1.8% | 0.6% |
| 11 | webmd.com | 1.6% | 2.0% | 1.5% | 1.4% |
| 12 | homedepot.com | 1.5% | 2.1% | 1.4% | 1.0% |
| 13 | clutch.co | 1.5% | 3.0% | 0.8% | 0.6% |
| 14 | psychologytoday.com | 1.4% | 2.3% | 0.6% | 1.4% |
| 15 | investopedia.com | 1.4% | 2.8% | 0.2% | 1.5% |
Health and SEO brands dominate this table because our customers track health and SEO keywords more than the general population searches them. This is a property of our sample, not a finding about which industries AI favours, and we flag it rather than bury it.
So What
What Follows From This
Three things, in rough order of how confident we are in them.
Measure per assistant, never blended
A single AI visibility score averages four systems with almost nothing in common. Track each separately or the number will not tell you where to act.
The long tail is genuinely open
With 30,462 brands named and the top ten taking 2.9%, being specific beats being big. Depth on a narrow topic is the realistic route in.
Match the assistant to your buyers
On Gemini a tracked site was named in 10.3% of answers; on Claude 2.0%. Audience research now precedes channel strategy.
Caveats
What This Study Does Not Show
Published in full, because a study that hides its limits should not be cited.
Scope
- It does not measure source citations — a separate question, and the reason Perplexity is excluded
- It does not measure traffic; many AI answers resolve without a click
- It does not show causation — we cannot say why a brand was named
- Google AI Overviews and Perplexity are not included in this dataset
Sampling and stability
- Keywords are customer-selected, skewing commercial, SMB, health and SEO
- 619 sites is a sample of our customers, not of the web
- 94.6% of checks ran in August 2026 — no claim is made about change over time
- Assistant behaviour shifts with model updates and may already differ
Every figure behind every chart is stated in the text. We are happy to answer methodology questions from anyone writing about this.
FAQ