Get Cited By AI — AI skill for Claude Code

get-cited-by-ai

**Get cited by AI, or find out you can't.**

I run seven sites. Bing has logged 100,000+ AI citations across them (102,133 through September 9, 2026, still adding close to two thousand a day), and I compared the pages that get cited constantly against the hundreds that never get picked once. Most of the standard advice did not survive that comparison. My highest-citation page earned its first 5,000 citations with zero JSON-LD on the page (the markup came later). My uncited pages have MORE schema, tables, and FAQs than my cited ones. The most polished llms.txt in the whole portfolio belonged to the site with zero live chat-engine citations.

This skill is what survived. It runs one honest check before anything else: is anyone actually asking AI the questions your site could answer? If nobody is, no amount of on-page work will get you cited, and the skill tells you so before you spend a week on it. If they are, it sorts your fixes into floors (worth doing, but they won't get you cited on their own) and the few levers that actually correlate with citations. Then it gives you a free way to measure whether any of it worked.

Built for anyone running Claude Code on their own site or a client's. This work goes by a few names: GEO (generative engine optimization), AEO (answer engine optimization), AI SEO. Same problem, different labels. Every recommendation here traces to an evidence-tagged rule instead of industry folklore.

What this is

A demand-gated workflow. Probe for grounding demand first, split by engine camp, then audit, entity-check, and benchmark a vertical. No craft recommendation before the demand gate. Every rule cites a measured dataset: 26,767 first-party AI citations across six owned sites, a 751-page same-template controlled regression (observational, see RULEBOOK "Honest limits"), an 820K-citation vertical archive study (mortgage), and a registered-predictions ledger with scored outcomes. `RULEBOOK.md` holds the full rule set, evide