12 consumer questions about paint, asked in Danish, Swedish and Norwegian. 108 answers from ChatGPT and Claude, 342 cited links. The page reports which brands, shops and sites the answers name. It says nothing about which paint is better.
Most-named paint brand per country, in the 90 answers to questions that do not name a brand.
Four findings
1. One brand is named everywhere, but each country has its own leader
Jotun is named in 52% of answers across the three countries. The leader changes by country: Flügger in Denmark (15 of 30), Beckers in Sweden (16 of 30) and Jotun in Norway (20 of 30). In Denmark the two most-named brands are almost level: Flügger 15, Jotun 14 of 30. A brand that leads at home can be close to absent next door.
2. The two assistants send people to different places
ChatGPT sends most of its links to paint makers’ own sites: 133 of 213 (62%). Claude sends most to advice sites, consumer tests and retailers: 89 of 129 (69%), and 29 to paint makers. Being found through your own site works for one assistant. For the other, the advice sites and shops that write about you decide.
3. How-to questions are open ground
Questions on outdoor wood, eco-friendly paint, cheap against expensive paint and how much to buy got 36 answers. Only 8 of them named any paint brand. The assistant explains the job and leaves the product open. That is where a brand with good, structured product data (spread rate, use, surface) has room to become the example.
4. One run is a sample
ChatGPT answered every question twice. The two answers named the same paint brands 59% of the time on average (overlap of the brand sets). Single answers vary, so the numbers here are directions, not market shares.
30 answers per country (Denmark 30, Sweden 30, Norway 30), questions 1 to 10.
Method
- Questions: 12 questions a consumer might ask before painting, written in Danish and translated to Swedish and Norwegian. Four are practical, three are about the category, three are about buying, and two name a brand on purpose. Brand numbers on this page use only the first ten.
- ChatGPT: gpt-5.6-luna through the OpenAI API with web search on and the user location set to the country. Every question asked twice: 72 answers.
- Claude: Claude with web search, run through Claude Code rather than the API, every question once: 36 answers. These answers are stored as their own run and labelled as such.
- Brands: each answer is searched for a list of paint brands and shops, whole word, any case. A brand counts once per answer. Product names (for example a wall paint range) are not counted separately.
- Links: each cited URL is reduced to its domain and matched against a rule list: paint makers, retailers, advice and test sites, public bodies, eco labels, press, forums.
Limits
- One run in October 2026, 30 answers per country. Treat the shares as directions.
- API answers with web search are close to, but not the same as, what a logged-in user sees in the app. The Claude answers were collected through a coding agent, not the consumer app.
- The translations were written by one person and not reviewed by native speakers.
- Naming a brand is not the same as recommending it. The count does not read tone.
Data
Download citations.csv (342 rows: question, language, country, assistant, sample, cited URL, domain, source type) · paint.json · how it is measured