The 27 things AI looks at on your site.

Every frontier model uses a named user-agent. If your robots.txt blocks them — or worse, allows only Googlebot — you are invisible to ChatGPT, Claude, Perplexity, and Gemini. The default Next.js and Vercel configs lean cautious; check yours.

Models follow sitemaps to discover canonical URLs. A sitemap that 404s, contains stale URLs, or omits half your content is worse than no sitemap — it teaches the model your site is broken.

Some models render JavaScript, most do not. If your pricing, product detail, or About content only appears after hydration, expect to be misquoted. The page in view-source is what models see.

Cloudflare's "Bot Fight Mode" challenges most AI crawlers by default. Vercel's firewall settings are equally aggressive. If a crawler hits a CAPTCHA, it never returns.

Models cite canonicals. If a page lacks one — or worse, points to a different URL — models may cite the wrong URL or none at all. Common with marketing landing pages and UTM-laden links.

Several crawlers HEAD before GET. If your CDN returns 405 on HEAD — common with static-only hosts — you lose those crawlers.

Crawlers timeout faster than browsers. If your time-to-first-byte exceeds 1 second on a cold cache, you're a coin flip. Models that time out twice often stop trying.

The brand-level manifest file frontier models check before parsing your HTML. Sites with a clean llms.txt get cited more accurately; sites without one get reconstructed from scratch every query.

The atomic identity card. Name, URL, logo, founding date, contact, social profiles. Read by every model. Without it, models reconstruct your identity from page titles and footer copy.

Q&A pairs in machine-readable form. Pulls directly into Google AI Overview and Perplexity. The fastest way to seed answers to known prospect questions.

Critical for AI shopping agents and pricing queries. Without it, models cite "no pricing information available."

OG tags travel with your links wherever they go. Models read them as a secondary identity source — and use the OG title and description when summarizing.

Tells models where this page sits in your site. Used to disambiguate similar pages and improve citation accuracy. Bonus: improves Google sitelinks.

Schema-stuffing or broken JSON-LD is worse than no schema at all. Models that hit a parse error on one block stop trusting the rest.

Two Organization blocks on the same page — common when a CMS injects one and your template injects another — confuses models. Pick one source of truth.

"Starting at $X" buried in marketing prose. "Contact sales for pricing." Both produce "no pricing information available" in model responses. Plain dollar amounts at the top of a /pricing route work.

"Founded in 2019" beats "established for over half a decade." Models extract facts, not vibes. Vague language gets paraphrased into hallucination.

"Platform" on the homepage, "tool" in the docs, "service" in the FAQ — models pick one and stick with it. The one they pick is rarely yours.

Brand identity questions often include "who founded" or "who runs." If your About page hides leadership, the model invents an answer or refuses to engage.

A working email or contact form, indexable. Hidden behind login? Models say "no contact information available." That answer travels.

Models lean on heading structure to segment context. Skipped levels (H1 → H4) and multiple H1s on one page reduce parsing accuracy.

Models prefer recent sources. Undated pages lose to dated competitors on the same query. Even a "last updated" footer helps.

Models weight Wikipedia heavily for entity disambiguation and founding facts. You don't need a full article — a mention on a related page can be enough.

Cross-referencing entity identity. sameAs in your Organization JSON-LD pointing to LinkedIn, X, GitHub, YouTube — whichever you actually use.

Models lean on a small set of high-authority publications for trust. A TechCrunch or Wired mention is worth ten product directories. Specialty trades count if they're indexed.

Spelling variations between your site, Crunchbase, LinkedIn, and your Twitter bio create entity-resolution noise. Models hedge when they can't reconcile.

Models cite press coverage. A consolidated page that links to coverage with quotes and dates makes you easier to cite — and signals that you're a real entity worth citing.