AgentScore benchmarks
How agent-ready is the web?
Benchmarks from real AgentScore scans: how well AI agents can get into, read and find their way around typical sites, and where your own site stands. We publish aggregates only and never name a site.
Live · updated every Monday
Every site AgentScore scanned in the last 90 days
Latest benchmark · October 2026
Online stores
255 sites scanned on October 20, 2026 with AgentScore 2026.10-v8. Read the full study.
- 0% scored 0–49
- 9% scored 50–64
- 40% scored 65–79
- 51% scored 80–100
Points earned, by category
The median site's share of the points available in each part of AgentScore.
- Access92%
- Readability89%
- Navigability72%
What the leaders do differently
How often the top quarter of sites passed each check, against the bottom quarter.
| Check | Top quarter | Bottom quarter |
|---|---|---|
| Form fields are labelled | 77% | 18% |
| Buttons and links have names | 71% | 16% |
| llms.txt guide for AI | 78% | 27% |
| Page has main and navigation landmarks | 81% | 41% |
| Clickable things are real buttons and links | 57% | 19% |
| Product pages give price and stock in a form agents can read | 84% | 51% |
The most common problems
Share of sites falling short (a fail or a warning) on each check, among the sites where it ran. Each links to the guide that explains the fix.
How this benchmark was made
600 websites drawn at random from the online shops listed in Wikidata, so the sample leans toward established retailers from many countries. 467 could be scanned, and 255 were stores, with products or a cart. Every site got the same automated AgentScore scan, which reads public pages and never fills in a form or places an order. We publish results only in aggregate and never name a site. Scores from a different scoring version aren't directly comparable.
Beat the benchmark
Get your AgentScore and a ranked fix list in a minute, or start a 30-day trial to watch it as you ship.