Common agent-readiness gaps in a measured sample
A prospecting sample of 567 company sites revealed recurring technical gaps. This guide explains the patterns while keeping the sample and scanner limitations visible.
Marketing sites are often strong for people and weak for agents, and the gaps are predictable. The evidence here is a scan of 567 company sites with one independent scanner between 28 June and 3 September 2026, which grew out of an earlier write-up of sixteen Finnish sites. The sample is a prospecting list and not a random draw, so read it as what recurred in the sites reviewed rather than as a count of the whole web. Of the 567 sites, 485 read Level 1 of 5 and 74 read Level 0, below it. The write-up reads Level 1 as the floor a default CMS ships, but the study did not record which CMS each site runs.
Rendering
A site that builds its content with client-side JavaScript, without server rendering or prerendering, returns an empty shell to any agent that does not run a browser, so for those clients the content never arrives in the first response. It is the gap to check first, because a site with no content in the first response has nothing else to fall back on.
Discovery
No llms.txt and a thin or missing sitemap leave an agent to follow page links, unless the site points to a Markdown version another way, such as an alternate link or an HTTP Link header. Among the 74 Level 0 sites, 68 have a first-fix note, and robots.txt appears in 45 of those notes and the sitemap in 38, with 29 naming both: a robots.txt the CMS shipped by default and a sitemap that is missing or never announced in it. Among the 439 Level 1 sites with a note, the sitemap appears in 102 and llms.txt in 99.
Content
The site offers only HTML, with no markdown form, so an agent spends part of its budget on markup, and a long page can exceed the reading tool's content limit and arrive cut short. Whether that happens depends on the tool's text extraction, the page's size and the limit it applies. This is a cost problem rather than a discovery problem: the agent found the page, but reading it can take more of its budget than the content is worth.
Data correctness
Capability is usually undeclared. The site may have an API or a useful action, but with no server card, OAuth discovery or a published API catalog such as the well-known endpoint RFC 9727 defines, an agent has nothing to find it from on its own. An integration configured by hand with the endpoint and its credentials can still work, so what the site loses is automatic discovery rather than the API itself. Among the 439 Level 1 sites with a note, robots.txt appears in 72, structured data in 49 and an MCP server in 35. Structured data is often missing too, so prices and facts are left for the agent to infer from layout instead of reading them from a declared source.
Verification
The discovery and content gaps rarely need a rebuild. That work is mostly at the edge and in a few small files, and the result shows up in a scanner on the next run against the version of the site that is live then. Rendering and data correctness can reach into the application or the system the data comes from, so their effort is estimated site by site. For discovery, a site does not have to change its structure. It has to publish what agents already look for and verify with a fresh scan that the fix landed.
turva.dev runs this exact review and reports each gap with a concrete fix. For an audit, contact info@turva.dev.
Frequently asked
What are the most common agent-readiness gaps on marketing sites?
In the first-fix notes of the measured sample, robots.txt and the sitemap lead among the Level 0 sites, and the sitemap and llms.txt among the Level 1 sites. HTML-only delivery with no markdown form is another gap to check, and so is client-side rendering that returns an empty shell to non-browser agents, which the sample did not record. Capability is usually undeclared and structured data is often missing, so prices and facts are left for the agent to infer.
Are agent-readiness gaps hard to fix?
It depends on the gap. Discovery and content fixes are mostly at the edge and in a few small files, and the result shows up in a scanner on the next run. Rendering and data correctness can reach into the application or the system the data comes from, so their effort is estimated site by site.
Which gap costs a marketing site the most?
In my reading, the costliest is client-side rendering that returns an empty shell to non-browser agents. The measured sample did not record rendering, but nothing else on the page can be read if the first response carries no content, and prices and facts are then left for the agent to infer.