“AI-readable” should mean that an intended, permitted system can retrieve the public page and interpret its main content without contradictory signals. It does not mean that every model can understand the page perfectly or that an answer product will cite it.
1. Fetch the canonical URL
Check the final status, redirect chain, response headers and content type. The primary page should normally return 200 directly. Verify that HTTP, www and trailing-slash variants resolve in one hop.
2. Evaluate the exact crawler policy
Read robots.txt for the documented user agent. Check page-level robots meta and X-Robots-Tag. A noindex directive generally must be crawlable to be seen. Do not use robots.txt as access control for private data. See Google's robots meta guidance.
3. Inspect initial and rendered HTML
Confirm that title, H1, canonical, primary answer and important links exist in robust HTML. Test client rendering where used, but do not make essential facts depend on a fragile interaction. Check mobile and assistive-technology usability too.
4. Resolve canonical conflicts
The canonical tag, internal links, sitemap and redirects should identify the same preferred URL. Multiple indexable copies with conflicting dates or claims make source review harder. Merge overlapping legacy pages when the intent is genuinely the same.
5. Review the answer structure
State the direct answer, then evidence, constraints, exceptions and next steps. Use headings that describe the section. Tables are useful for genuine comparisons; FAQ blocks should exist only when the page answers real recurring questions.
6. Verify material claims
Link time-sensitive platform behavior to primary documentation. For original data, publish the sample, period, method and limitations. Remove claims that cannot be supported. A clean HTML outline cannot rescue unreliable facts.
7. Separate search from training
OpenAI distinguishes OAI-SearchBot and GPTBot, and Perplexity distinguishes PerplexityBot and Perplexity-User. Configure and test the intended purpose. A live citation does not prove training-data inclusion, and a training crawler decision does not necessarily control search discovery.
8. Test outcomes
Use logs and search reporting for discovery, a stable prompt set for answer representation, exposed-source review for citations, and analytics for visits. Do not infer one layer from another.
Audit result
Report each check as pass, fail, not applicable or unknown, with evidence. Prioritize broken access and contradictory canonical signals before stylistic rewrites. Use the crawler-layer guide for detailed diagnostics.
Interpretation note
This article is educational material, not a product commitment or a guarantee of rankings, citations, traffic or commercial outcomes.