Skip to content
Solidus kodasAudit

How we assess a website's accessibility to AI search and its security

The audit measures technical conditions that can be verified from outside. This page describes exactly what we check, how we calculate the score and where the limits of this method lie.

What we can reasonably determine

  • Whether the page and its text can be retrieved from our infrastructure.
  • What robots.txt declares for specific crawlers.
  • Whether the page declares indexing and snippet restrictions (meta robots, X-Robots-Tag).
  • Whether content, internal links and structured data are reachable in the sample checked.
  • Whether additional AI-oriented files have been published and whether there are signs of integrations.
  • How encryption, security headers, cookies and protection against e-mail spoofing are configured.
  • Which software and versions the site discloses publicly and which vulnerabilities are known for those versions.

What the audit does not prove

  • That ChatGPT or any other system has already indexed the site.
  • That the site will be cited or recommended.
  • That a successful request from us means a successful request by a real crawler from its own IP addresses.
  • That the sample checked reflects every product, category or language version.
  • That the site is secure or vulnerable: we assess only publicly visible signs; we do not look for vulnerabilities and do not attempt to exploit them.

Scope of the audit

The quick audit checks the public page you entered, robots.txt, a limited part of the sitemap and up to 5 selected pages from the same domain. Pages are selected by following public links, avoiding search, basket, login and filter addresses. Every report shows the actual list of addresses checked.

The checker uses only HTTP and HTTPS; it does not log in to accounts, fill in forms, use cookies or bypass protections. We analyse crawler rules separately. When the site serves our own request and robots.txt allows it, we send one comparison request for each AI crawler User-Agent name; every such request is labelled with our real name (the X-Audit-By header).

How the score is made up

The overall score is made up of these groups:

  • Access – 25 %. Whether the page can be reached and whether crawlers are allowed to read and index it.
  • Content readability – 22 %. Whether the main content is in the HTML returned by the server and whether the topic of the page is clear.
  • Discoverability – 14 %. Whether crawlers can find the site's pages through the sitemap, canonical links and internal links.
  • Semantic clarity – 11 %. Whether structured data and language signals help to understand the content unambiguously.
  • Security – 18 %. Encryption, security headers, cookies, the software versions the site discloses and the vulnerabilities known for them. Part of the overall score; the group score is also shown on its own as the security score.
  • Speed – 10 %. How quickly the server responds and what the HTML makes the browser wait for: response time, compression, render-blocking scripts, images. This is not a Core Web Vitals measurement in a browser.

Additional agent features do not count towards the score. llms.txt, Markdown alternatives, Link headers and signs of WebMCP are shown in a separate informational section. The absence of WebMCP cannot be reliably proven from HTML alone, so we show “sign detected” or “not confirmed”.

These weights are a choice made in our methodology, not a validated model of how AI systems rank content. The score is an aid to orientation: what matters most are the specific findings and their evidence.

Coverage, critical barriers and “Insufficient data”

  • Checks that do not apply are removed from the calculation – they neither raise nor lower the score.
  • Check coverage is shown next to the score: the share of the weight of applicable rules that we were able to assess.
  • If essential data is unknown, we show “Insufficient data” instead of an overall score.
  • Critical barriers, such as a declared ban on indexing the whole page, are shown separately so that a high score on other checks does not hide them.

Check statuses

  • OK – the check was carried out and no shortcomings were found.
  • Warning – a shortcoming or contradiction was found that is worth reviewing.
  • Problem – a barrier was found that we recommend fixing.
  • Could not be determined – the data could not be obtained (e.g. a timeout occurred or protection was triggered). This is not automatically a fault of the website.
  • Not applicable – the check is not relevant to this website (e.g. hreflang on a single-language site).

Every finding has a rule ID and version, the URL checked, the time, HTTP information or a content excerpt, an explanation, a priority, a recommendation and a link to the source. We state the reliability of the evidence separately.

Crawler purposes

Search crawlers, model-training crawlers and user-initiated access are shown separately in the report because they are different decisions. Blocking training crawlers does not affect AI search crawlers, but it makes the site's content less available to AI systems, so it counts towards the score; when all training crawlers or all user-initiated access are shut out, the overall verdict is no longer "good". If the ban is the owner's deliberate decision, the finding can be treated as informational.

Limitations

  • We check from our own infrastructure. A firewall or protection service may respond to a real crawler differently than it does to us – either more favourably or more strictly.
  • The sample is small: the result for a few pages is not an assessment of the whole site.
  • We execute JavaScript only for the page you entered, and only when its initial HTML is an empty JavaScript shell. The sampled internal pages are assessed without JavaScript, so we do not see content that appears on them only in the browser.
  • A syntax check of structured data does not confirm that the data is suitable for every search system.
  • An HTTP 200 response alone is not enough: we treat a file as found only when its content matches the expected format.
  • Standards and crawler lists change. When the methodology changes, a new version is created, and the scores of old reports are not recalculated.

Rules in force

Loading the list of rules in force…