Methodology: how Geoa measures

The GEO Score engine: 15 dimensions, weights summing to 100%

GEO Score is a weighted average of 15 technical and content dimensions. The weights below are the actual weights in the engine — the table is generated directly from the code, so it cannot diverge from what we actually measure:

The same engine powers the free GEO test, the platform's ongoing scans and the GEO Benchmark — one measurement standard, everywhere. The score is 0-100 with grades A-F.

Our own research numbers: n, date and caveats — always

When we publish statistics from our own scans (for example llms.txt adoption in Denmark), three fixed rules apply:

• The sample is stated. For example: 465 search-visible Danish business websites, scanned July 21, 2026.
• The caveat is stated. Our samples consist of search-visible businesses — not a random sample of the entire industry. Each article says so explicitly, because it affects what the numbers can be used for.
• The method can be verified. The scans use the same engine as the public GEO test — anyone can run it on any domain and see the same dimensions.

Cite our numbers as "Geoa, [month year], n=[sample]" — and please link to the analysis so your readers can see the method and caveats.

GEO Benchmark: editorial rules

The benchmark live-scores a curated list of the Nordic region's 40 largest companies (by revenue, public sources) plus 10 well-known SEO tools. The rules: only curated, public companies — never user-tested domains; same engine, same weights as the free test; and geoa.app is shown out of competition when we appear ourselves. The list is hardcoded and versioned in the codebase.

AI citation measurement: 5 surfaces, measured — not guessed

We measure whether AI answers actually mention a business on five surfaces: ChatGPT, Claude, Gemini and Perplexity via their answers, plus Google AI Overviews via the search results. We are happy to mention other engines (for example Microsoft Copilot, which builds on the Bing index) as market context — but we only publish measurement numbers for surfaces we actually measure.

Because AI answers are non-deterministic (the same question can produce different answers on the same day), we measure our core content with multiple samples per question per engine and report frequencies over time rather than single measurements — small week-to-week fluctuations are noise, and we say so openly.

Ongoing fact-checking — of ourselves too

• Competitor claims are machine fact-checked every week. Prices, existence and identity on our comparison pages are automatically verified against the sources and date-stamped. If a claim cannot be verified, it is flagged — it is never "corrected" automatically.
• Our own copy is held to the same standard. The codebase contains automated guards that block any release with known error classes: invented ratings, unmeasured case numbers, unsourced amounts and measurement claims about engines we do not measure. The guards run on every single change to the site.
• Removed numbers do not come back through the back door. When a number turns out to be impossible to substantiate, it is removed everywhere — and the pattern is added to the guards' blocklist.

The honesty contract

Geoa's product is factual credibility — that is exactly what AI search engines reward our customers for. So the same standard applies to us: we only publish numbers with a named source or our own documented measurement. No invented customer numbers, no purchased badges, no outcome promises we have not measured. If you find a number on our pages that does not live up to this, write to us — it will be corrected or removed.

See also the Trust Center for data security and data processing, and the complete GEO guide for the discipline behind the measurements.