GEO results and data: what we can show you
We don't have a client case we can publish yet. When our first clients give written permission, their results go on this page. Until then, we can show three things: our own product's numbers, a public leaderboard of websites, and our own site's score.
Our own product: after four weeks of GEO, AI became its biggest source of visitors
An AI video tool we built for overseas users. From March 21 to April 17, 2026, we did a round of GEO on its own site. These are its visitor numbers from January 1 to September 16.
What we did in those four weeks
- Let AI crawlers in: Cloudflare's AI crawler blocking had been stopping nine of them.
- Added structured data, llms.txt, and real update dates in the sitemap.
- Found one question AI gets asked a lot (which AI video tools have free credits) and wrote one page that answers it: a direct answer up top, a comparison table below.
- Around the same time, added a batch of long-tail pages such as comparisons and alternatives.
What these numbers don't show
- One site, one observation, not a controlled experiment.
- New AI models launched and a holiday promotion ran in the same weeks, so the effect of GEO alone can't be separated out.
- Visitors are not sign-ups, and sign-ups are not revenue. Direct visits were 38%, and where they came from is unclear.
- It shows the approach can work. It doesn't mean your store will see the same numbers, and it isn't a promise to you.
Source: the product's own analytics (PostHog channel grouping, visitors), January 1 to September 16, 2026.
Public leaderboard: 324 websites, median score 55
We scored 370 well-known websites with geo-score and 324 got a score. The list, every sub-score and the method are public, so anyone can re-run it.
The leaderboard doesn't yet have a separate group for independent stores. We're scoring a set of Chinese brands' overseas stores now and will publish the results here.
Measured 2026-10-02 with geo-score 1.5.0. Full leaderboard · Rubric and code
Is the score stable? Same site, tested twice
On September 9, 2026, with the version of the time (1.1.x), we scored 229 sites twice on the same day. 95.6% landed within 5 points of their first result, 45.0% scored exactly the same, and five sites flipped in or out of the access cap (reproducibility report). The tool's stability has improved since and hasn't been re-measured, so treat this as that run's reading.
So we set a rule: a before-and-after change of less than 5 points is not a result, and contract acceptance works the same way.
We also want to test something nobody has shown yet: whether higher-scoring sites get cited by AI more often. We published how we will measure and analyse it before collecting the data (a preregistration), so we can't pick a favourable method afterwards. Until then, we don't claim a higher score gets you recommended more.
$ python3 geo_score.py www.citespot.si --brief AIV READINESS https://www.citespot.si 90 / 100 Leading Reachable 15/15 ✓✓✓ Understandable 22/22 ✓✓✓✓✓ Content Citability 31/35 ✓✓✓◐✓ Brand Credibility 3/10 ⊘⊘✗✓✗ Answer Fit 5/6 ✓⊘◐
Real output (excerpt): the website you are reading, geo-score 1.8.0, measured on launch day, October 9, 2026. The missing 10 points are in three places: no knowledge-graph entry for CiteSpot yet (a new brand, and we won't create one ourselves); most figures on the 8 sampled pages carry no source, and two of those pages are our own price lists; no video yet. Scoring rubric
What would your store score?
Run it yourself, or send us the URL.