Since July 2026 SeekinWeb checks for free whether AIs can find, read and cite a website. This is what the aggregate data of the first months says. Small, self-selected sample: read it as an honest snapshot, not a census.
of domains have gaps that make AI reading harder (score under 80)
score under 50: hard to read for most assistants
do not tell the AI what they are (no valid structured data)
block some AI crawler, on purpose or unknowingly
The signals that fail most — llms.txt (60.5%), semantic structure (38.6%), structured data (30.7%) and sitemap (27.2%) — are fixed with text, not redesigns. The expensive one (content painted with JavaScript) affects 16.7%.
21.1% block GPTBot, ClaudeBot, PerplexityBot or others in robots.txt. Often it is a template or a security plugin nobody reviewed: the site is invisible to assistants it would actually want.
Software averages 91.2 out of 100; everyone else sits between 67.9 and 85.2. Those who live off the web keep it tidy; those who live off something else (services, local commerce) lose the most when the AI answers on their behalf.
We ran our own AI Mention Test on seekinweb.com: 10 questions a customer would ask when looking for a tool like ours, answered by an assistant with web search. It appeared in 0. In our place: Otterly, Peec AI, Profound, Semrush, Ahrefs. A site scoring 100/100 on readability can still not exist for the AI if nobody else talks about it. That is why the mention test is part of the report.
It is the signal that fails most (60.5%) and also the one that weighs least in our score: Google has said it does not read it and 2026 analyses show almost no crawler requests it. We generate it for free because it costs nothing, but nobody should sell it to you as ranking.
| Signal | Weight in score | Domains failing it |
|---|---|---|
| llms.txt | 8 | 60.5 % |
| Semantic structure | 12 | 38.6 % |
| Structured data (JSON-LD) | 15 | 30.7 % |
| Sitemap | 8 | 27.2 % |
| Metadata | 10 | 21.1 % |
| AI crawler access | 17 | 21.1 % |
| JS-free readability | 20 | 16.7 % |
| Performance | 10 | 11.4 % |
114 domains · average score 80/100
| Sector | Domains | Avg /100 |
|---|---|---|
| Software | 39 | 91.2 |
| Other | 44 | 67.9 |
| Services | 13 | 81.9 |
| Media | 7 | 76 |
| Commerce | 6 | 85.2 |
Sample: 819 scans of 114 unique domains between 2026-07-02 and 2026-09-05, taking each domain's latest scan. Aggregates only: we never publish individual domains or scans.
Each domain is scored on SeekinWeb's 8 public signals, weighted by evidence (JS-free readability 20, AI crawler access 17, structured data 15, semantic structure 12, metadata 10, performance 10, llms.txt 8, sitemap 8). Method documented on the public ranking.
The sample is self-selected: domains someone chose to scan, often well-known sites used to try the tool, over-weighted to software. It does not represent the Spanish-speaking web or SMBs in general. We will repeat the study every quarter with more data.
The AI Mention Test reflects one assistant's answer at one moment; it is non-deterministic and varies across assistants. It is not a ranking or a promise.
You may cite this study by linking this page. Data: SeekinWeb, 2026.
AI Visibility Score, the 8 signals with their fix, and the AI Mention Test. Free, no signup.