Bermee Index · The independent, published method for tracking AEO
The index of what AI names.
Bermee is an open and independent measurement standard, published monthly. When people ask AI what to choose — a product, a provider, a service — it names a few brands. Bermee shows which ones, how often, and how sure we are.
Issue 1 · Running shoes — US · Published 27 October 2026 · Method v0.9
What the Bermee Index is for
The Bermee Index is an independent, published method for measuring how brands are positioned in AI answers — answer engine optimisation, or AEO — across 5 AI systems: OpenAI, Perplexity, Gemini, Anthropic and Google AI Overviews.
It helps brands, agencies and marketing professionals follow trends and changes month by month, against a published method — anyone can check how each number was made.
Issue 1 covers running shoes in the US. Four more categories follow, with their first issues in November and December. Any category can be commissioned: $699 founding rate for the first 25 commissions or until 31 January 2027, then $2,000.
Pricing →Join the waiting list →Log in to see the full dataset — free →
The leaderboard
Who AI names, brand by brand
| Rank | Brand | 95% range | |||
|---|---|---|---|---|---|
| 1 | Brooks | 888 | |||
| 2 | Asics | 888 | |||
| 3 | Hoka | 888 | |||
| 4 | New Balance | 888 | |||
| 5 | Saucony | 888 | |||
| 6 | Nike | 888 | |||
| 7 | Adidas | 888 | |||
| 8 | Altra | 888 | |||
| 9 | On | 888 | |||
| 10 | Puma | 888 | |||
| 11 | Mizuno | 888 | |||
| 12 | Topo Athletic | 888 | |||
| 13 | Salomon | 888 | |||
| 14 | Skechers | 888 | |||
| 15 | Under Armour | 888 | |||
| 16 | Orthofeet | 888 | |||
| 17= | Carson Footwear | 888 | |||
| 17= | Softstar Shoes | 888 | |||
| 19 | La Sportiva | 888 | |||
| 20= | Merrell | 888 | |||
| 20= | SOM Footwear | 888 |
Statistically tied at the top: Brooks, Asics and Hoka. Their 95% ranges overlap, so the index orders none of them above the others.
Within margin: the two brands’ 95% ranges overlap, so the index does not call one ahead of the other.
Systems measured (5)
| System | Model / version | Surface | How it was collected | Answers |
|---|---|---|---|---|
| OpenAI | gpt-4o | Answer returned by the model’s API | API request with the issue’s questions; 100% of answers used web search | 200 answers counted |
| Perplexity | sonar-pro | Answer returned by the model’s API | API request with the issue’s questions; 100% of answers used web search | 200 answers counted |
| Gemini | gemini-3.6-flash | Answer returned by the model’s API | API request with the issue’s questions; 100% of answers used web search | 200 answers counted |
| Anthropic | claude-sonnet-4-5 | Answer returned by the model’s API | API request with the issue’s questions; 69% of answers used web search | 200 answers counted |
| Google AI Overviews | google-aio-dataforseo-2026-10-19 (engine of record); google-aio-dataforseo-2026-10-26 (engine of record); google-aio-serpapi-2026-10-19 (cross-check only, in no metric); google-aio-serpapi-2026-10-26 (cross-check only, in no metric) | Google AI Overviews in Google search results | Search results collected through dataforseo; 100% of answers used web search | 88 answers counted |
Issue details
Issue 1 · Running shoes — US · Published 27 October 2026 · Method v0.9 · Issue window: 19–26 October 2026 · Collected: 19 October 2026, 26 October 2026 (2 weekly runs) · 21 brands ranked
40 questions × 5 runs × 5 AI systems = 1,000 answers · 112 excluded (63 system errors, 49 searches showed no AI Overview) · 888 counted
What is published, what stays private — and why
Published CC BY 4.0
- Metric definitions and formulas
- Prompt construction rules
- Sampling rules and executions
- Source taxonomy
- Sample prompts and every retired prompt
- Disclosure block per issue
- Headline tables as CSV
Kept private
- The live prompt set
- Brand alias dictionaries
- The source classifier
- Raw answers (licensable)
Published: the method, the rules, each issue’s disclosure and the headline data (CSV, CC BY 4.0). Kept private: the live questions and the raw answers — licensable — so no one can tune their content to the test.