Every brand page on this site shows one number: an AI Visibility Score out of 100. This article explains where that number comes from, step by step, so you can judge it, reproduce it, or argue with it.
The short version: we ask three AI models the questions real buyers ask, in every language of the region, many times over. Then we score what comes back on five components and publish the result for every brand we measured.
The corpus
The Nordic-Baltic edition covers 66 brands across 11 markets. Each brand was queried in 12 languages (Lithuanian, Latvian, Estonian, Finnish, Polish, Czech, Slovak, Swedish, Norwegian, Danish, German, English) on three AI models with live web grounding: GPT, Gemini, and Perplexity Sonar. That produced 35,640 grounded responses carrying 131,667 citations from 21,077 unique domains.
The Poland edition covers 46 consumer brands with Polish buyer prompts on the same three models, yielding 17,891 citations from 1,907 domains.
Buyer-intent prompts drive everything. We ask what a customer would ask ("best music streaming service in Sweden", "which bank should a small business in Slovakia use"), never "tell me about brand X". A brand earns its score by showing up unprompted.
The five components
Each brand's composite score is a weighted sum of five components, all normalized to 0-100:
- Sentiment (25%): how positively the models describe the brand when it appears.
- Recommendation (25%): the share of buyer-intent prompts where the models name the brand. This is the component that maps most directly to pipeline.
- Source quality (20%): what kind of domains ground the answers. Tier-1 media and industry reports score higher than anonymous long-tail pages.
- Consistency (20%): whether the story stays the same across languages and phrasings.
- Stability (10%): whether the same question asked again returns the same answer. We measure it with the same repeated-prompt protocol as our free Dice Roller.
The composite maps to named bands: 85 and above reads as Highly recommended, 70 to 84 Recommended, 40 to 69 Emerging, below 40 AI-invisible. The 2026 leaderboard's top score is 67, which tells you something about how much headroom even the winners have.
The metrics you will not find anywhere else
Three measurements in the Index exist because multilingual data exposes things single-language studies cannot see.
Bilingual penalty. We compare each brand's AI Reputation Index in its home language against English. Latvijas Gāze scores 60.1 in Latvian and 42.9 in English, a 17.2-point home advantage that disappears the moment a buyer switches language. Tatra Banka shows the same pattern at +13.6. Global brands like Spotify sit within a point across all twelve languages. If your expansion plan assumes your home reputation travels, this number says whether it does.
Citation concentration (source HHI). A Herfindahl-Hirschman index over the domains that ground a brand's answers. Above 0.72 means a few domains control the narrative, which cuts both ways: fragile, but fixable through two or three domain partnerships. Below 0.50 means the story is spread wide and resilient.
Blind languages. The count of study languages where a brand's score falls below the regional 45-point floor. A brand can lead its home market and be functionally invisible in eight of twelve languages.
What we publish
Every measured brand gets a public page at /ai-visibility/{brand} with its score, rank, per-model sentiment, language divergence chart, and citation sources: Spotify, Wise, Amica, and 108 more. The full ranking, country cuts, and the downloadable report live on the Index hub, with reuse terms in the press kit.
We measure visibility, narrative, and recommendability in AI answers. We do not rate company or product quality. A low score means AI does not tell your story, and that is a fixable marketing problem, not a verdict on the business.
Dates and versions
Nordic-Baltic data was collected and published in April 2026 and last updated in May 2026. The Poland edition was published in June 2026. Model versions are pinned per study; when we re-run the Index, prior scores stay archived so movement is measurable.
Questions about the method, or want your category measured? Talk to the team.