135 Norwegian sites, scanned end to end
64% of Norwegian commercial sites have a blocking issue that stops an AI assistant from completing a customer's task.
Most of them do not know. The failure happens on the assistant's side of the conversation — the customer is told it could not filter your products, and you never see the visit at all.
87 of 135 sites across 23 sectors. Median score 40/100.
The sites with the most to lose are failing hardest
Sectors whose entire business is a browsable catalogue sit at 83–91%. Sites with little to filter score well largely because those checks do not apply to them.
| Sector | Sites with blockers |
|---|---|
| DIY / home / garden | 91% |
| Fashion | 88% |
| Beauty / pharmacy | 86% |
| Big retail / electronics | 83% |
| Travel / transport | 80% |
| Health clinics, logistics | 0% |
Health clinics and logistics: 0%.
What actually breaks
The six most common blocking issues across the corpus, by number of sites.
| Blocker | Sites |
|---|---|
| Cannot filter, sort or paginate a listing end to end | 37 |
| Cannot narrow by a published attribute and confirm buyability | 35 |
| Search returns nothing readable without JavaScript | 21 |
| Refuses an honestly identified agent outright | 20 |
| A deep-landed page is a dead end | 16 |
| robots.txt forbids live and search assistants | 8 |
How it works
We fetch every sampled page twice — once as an ordinary client, once as an honestly declared agent. The difference between the two responses is the finding. We never spoof a vendor's user agent: bot verdicts are identity-based, so pretending to be Googlebot would prove nothing about your real agent traffic.
No JavaScript is executed, because the major AI crawlers do not execute it either. Anthropic states it plainly: the web fetch tool does not support sites rendered dynamically with JavaScript. An emulation with a headless browser would systematically over-report what an assistant can actually see.
Then we run the scenarios. Not "does this site have a filter" but "can an agent narrow this catalogue by a published attribute and confirm the result is buyable". Every finding carries the page it came from and a command to reproduce it.
What we measure — and what we do not
Measured
Whether an assistant that reaches your site can navigate and use it: search, filters, sort, pagination, deep-link entry, product data, forms and actions.
Not measured
Not whether assistants reach you at all. Merchant-feed enrolment bypasses the website entirely, index membership is only partly observable, and because CDN bot decisions are identity-based we never claim that a refusal for our scanner proves a refusal for a named assistant. We verify that data is readable, not that it is correct.
Caveats
Sampling, not audit: roughly 8–25 pages per site, from a single vantage point and locale.