For press
Figures, quotes and images from the October 2026 measurement. Everything may be reproduced freely with attribution.
The short version
Reko AI Index put 38 ordinary purchase questions to 4 AI models, 3 times each, and counted which Swedish brands were named. That produced 456 analysed answers.
The result: it matters a great deal which AI the consumer asks.
This month's findings
1. ChatGPT often names no brand at all
In 47 % of answers, ChatGPT does not name a single supplier. The model reasons about the question, points to comparison services and asks follow-up questions, but recommends nobody. Compare with Gemini, which does the same in 1 % of answers.
2. The gap in how many brands get named is more than double
Gemini names 6.15 brands per answer on average. ChatGPT names 2.48.
| Model | Brands per answer | Answers naming no brand |
|---|---|---|
| Claude | 4.33 | 6 % |
| ChatGPT | 2.48 | 47 % |
| Gemini | 6.15 | 1 % |
| Perplexity | 3.4 | 7 % |
3. The same brand can be invisible to one model and dominant with another
Fortum in electricity differs by 80 percentage points between the model that names it most often and the one that names it least.
| Model | Share of answers naming Fortum |
|---|---|
| Claude | 57 % |
| ChatGPT | 7 % |
| Gemini | 87 % |
| Perplexity | 53 % |
Quotes you may use
We are used to a search returning roughly the same result no matter who is searching. AI answers do not work that way. Fortum is named in 87 % of Gemini answers and in 7 % of ChatGPT answers, to exactly the same questions.
The most unexpected finding is the silence. In 47 % of answers ChatGPT names no supplier at all. For a brand that built its visibility on appearing at the top, that cannot be optimised away, because there is no position to take.
The quotes are generated from the measurement and therefore always match the tables above.
Images and data
- Overview image
- Claude
- ChatGPT
- Gemini
- Perplexity
- Broadband
- Electricity
- Insurance
- Mobile plans
- Raw data
Images and figures may be published freely, including commercially, with attribution to Reko AI Index and a link to https://rekomarketing.se. The data sits under CC BY 4.0.
How the measurement was made
The same 38 open questions were put to Claude, ChatGPT, Gemini, Perplexity, 3 times each, with web search enabled. We never name a brand in the question, because we are measuring which brands the model puts forward on its own.
Brands are extracted by deterministic name matching, not by a model. That means anyone can download the raw data and arrive at exactly the same figures. The full method is published, including known limitations.
Contact
Questions, additional cuts of the data or an interview: hello@rekomarketing.com