Content strategy is wasted on a site an AI crawler cannot read. Run these twelve checks before writing anything new.
1. Robots rules per agent
List every AI agent your robots file mentions: GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot, Google-Extended, Bingbot, CCBot. Each should be allowed or blocked on purpose. Inherited plugin defaults are the most common cause of accidental invisibility.
2. Raw HTML content
Fetch a key page with JavaScript disabled. If the main content is missing, retrieval systems may be storing an empty shell. Static generation or server rendering removes the risk permanently.
3. One H1, logical headings
Exactly one H1 per page, then H2s and H3s in order. Heading structure is how a parser segments your page into retrievable chunks.
4. Self-contained paragraphs
Short paragraphs that make sense in isolation. Long paragraphs that depend on three preceding ones extract badly and get misquoted.
5. Visible dates and authors
Published and updated dates in the HTML, plus a named author with real credentials. Undated content is treated as less reliable by both engines and models.
6. Validated structured data
Organization, WebSite, Article, Service, FAQPage, HowTo and Breadcrumb where relevant, connected and validated. Broken JSON-LD is worse than none because it wastes eligibility.
7. Consistent entity facts
Your name, category, services, locations and people described identically across the site and every external profile, connected via `sameAs`.
8. Clean canonicals
One canonical URL per piece of content. No duplicate paths with and without trailing slashes serving the same page, no parameter variants competing with the original.
9. A real sitemap
An accurate XML sitemap listing canonical URLs with sensible last-modified dates, referenced from robots.txt. Remove redirected, dead and noindexed URLs.
10. Speed on real devices
Page weight and JavaScript execution affect crawl budget and human conversion. Target a fast first paint on mid-range mobile, not on your laptop. Static pages with minimal JavaScript are the cheapest way to hit this.
11. Accessible markup
Semantic HTML, alt text, labelled form fields, sensible focus order. Accessible pages are more machine-readable pages; the same structure that helps a screen reader helps a parser.
12. Optional: llms.txt
A flat markdown file at your root pointing to the pages you want models to read. Support is uneven and it is not a standard, so treat it as a small bonus after the other eleven checks pass — never as a strategy.
Run it as a pass/fail
Score each item pass or fail, fix the failures, then re-check with the same method. Most sites we audit fail on two, six and seven — rendering, schema and entity consistency — and those three fixes typically unlock more than a quarter of new content would.
If you would rather have it done and verified, that is exactly the scope of our LLM & ChatGPT SEO service and our AI Websites build.