Found, but not understood
We measured about 100,000 local business websites in 34 countries. The gap is in understanding and enquiring, not in finding.
Four stages, one gap
An AI assistant has to find a website, understand it and be able to quote it. Only then can it enquire or book for its user. This is how many websites meet each stage:
Find
99.6%block no AI reader that gives answers
76,885 of 77,187 websites
Understand: who
16.0%tell a machine their company, address and phone
12,339 of 77,187 websites
Understand: what
7.5%tell a machine an offer or a price
5,827 of 77,187 websites
Quote
66.7%have a page description
66,046 of 98,981 websites
Enquire or book
2.1%offer booking, ordering or an enquiry in machine-readable form (minimum)
2,030 of 98,981 websites
Almost nobody is blocked: 99.6% of websites let in every AI reader that gives answers.
But only 16.0% tell a machine who they are. That means company, address and phone in the format machines read (schema.org).
2.1% offer a booking, an order or an enquiry in a way an assistant can trigger. This is a minimum: we count only the machine-readable entry, not every form.
In addition, 16.2% have an llms.txt, a file with notes for AI readers (16,047 of 98,981 websites). On 8.4%, almost no text arrives without JavaScript.
Measure your own website
The free AI access check tests your website on nine points and shows what an AI reader finds there.
By country
The twelve countries with the most measured websites. “Find” is above 99% in all of them; the differences begin with understanding. In the United States, 3,429 of 13,592 websites state “Who”; in Germany, 2,271 of 19,641.
| Country | Websites | Find | Who | What | Quote | Enquire |
|---|---|---|---|---|---|---|
| Germany | 24,833 | 99.8% | 11.6% | 5.4% | 63.6% | 1.1% |
| United States | 17,447 | 99.7% | 25.2% | 11.9% | 69.0% | 3.6% |
| France | 7,557 | 99.6% | 18.2% | 9.9% | 74.6% | 3.5% |
| United Kingdom | 5,770 | 99.5% | 18.0% | 7.1% | 68.6% | 2.8% |
| Netherlands | 4,810 | 99.5% | 12.6% | 6.4% | 62.3% | 1.7% |
| Poland | 3,216 | 99.7% | 14.5% | 8.8% | 61.8% | 2.3% |
| Canada | 3,157 | 99.7% | 21.2% | 9.5% | 70.0% | 2.3% |
| Italy | 3,048 | 99.6% | 14.9% | 5.5% | 68.3% | 1.7% |
| Austria | 2,725 | 99.5% | 15.9% | 6.2% | 69.7% | 1.0% |
| Spain | 2,642 | 99.3% | 12.5% | 8.0% | 69.2% | 1.4% |
| Switzerland | 2,547 | 99.3% | 14.0% | 5.4% | 68.1% | 1.8% |
| Belgium | 2,220 | 99.2% | 15.0% | 5.8% | 64.0% | 2.0% |
By sector
Food and drink offers booking or ordering in machine-readable form most often. In every other sector the minimum is 1.0% or less.
| Sector | Websites | Find | Who | What | Quote | Enquire |
|---|---|---|---|---|---|---|
| Retail | 22,175 | 99.5% | 14.5% | 6.9% | 69.1% | 0.4% |
| Food and drink | 21,569 | 99.8% | 19.3% | 9.9% | 63.2% | 8.0% |
| Offices and advice | 12,034 | 99.5% | 15.8% | 7.6% | 70.3% | 0.2% |
| Leisure and culture | 10,899 | 99.4% | 9.2% | 3.8% | 60.3% | 0.2% |
| Accommodation | 9,886 | 99.7% | 16.0% | 7.2% | 67.8% | 1.0% |
| Health | 8,855 | 99.7% | 19.7% | 9.0% | 69.7% | 0.5% |
| Trades and production | 6,166 | 99.9% | 15.3% | 6.8% | 67.3% | 0.3% |
| Local services | 6,132 | 99.6% | 18.8% | 8.0% | 68.3% | 0.3% |
| Other | 1,265 | 98.9% | 10.0% | 4.3% | 68.1% | 0.2% |
By website system
The system a website is built with matters more than country or sector. Depending on the system, the share for “Who” ranges from 0.4% to 52.9%, the share with an llms.txt from 0.4% to 94.0%. We name no vendor and rate no system.
| System | Websites | Find | Who | What | Quote | Enquire |
|---|---|---|---|---|---|---|
| Other or custom-built | 45,336 | 99.4% | 16.1% | 9.8% | 69.4% | 2.7% |
| System A | 38,531 | 99.8% | 9.5% | 5.9% | 60.6% | 0.6% |
| System B | 5,096 | 100.0% | 52.9% | 2.5% | 79.7% | 10.1% |
| System C | 3,412 | 100.0% | 39.6% | 2.8% | 59.2% | 0.7% |
| System D | 2,929 | 100.0% | 8.5% | 10.2% | 83.5% | 0.3% |
| System E | 1,596 | 99.9% | 17.0% | 3.9% | 67.4% | 0.8% |
| System F | 1,050 | 100.0% | 0.4% | 0.7% | 63.2% | 0.0% |
| System G | 1,031 | 99.8% | 22.2% | 17.4% | 90.0% | 2.7% |
Blocking
At least 0.4% of websites block an AI reader that gives answers. About four in ten blocks visibly come from a template.
Method
The addresses come from OpenStreetMap and were drawn at random: 150,000 addresses, of which 116,940 were read.
Schools, associations and public offices are not counted. Several websites of the same operator count as one. That leaves 98,981 websites.
We measured from 6 October 2026, 23:30 UTC, to 7 October 2026, 05:13 UTC.
For each website we read the public home page, robots.txt and llms.txt.
The shares for “Quote”, “Enquire or book” and llms.txt are based on all 98,981 websites.
The shares for “Find”, “Who”, “What”, for blocking and for text without JavaScript are based on 77,187 websites. We have measured these features only since an extension during the run; websites read before it are not counted there.
Limits
- We do not report groups of fewer than 300 websites.
- Blocking figures are minimums: we did not measure Google’s crawler.
- We measured the home page, not every subpage.
- The report states counts. It rates no business and no website system.
Source: Metatron Intelligence, measurement of 98,981 local business websites, 6 and 7 October 2026.