paultaki/get-cited-by-ai

This is a Claude Code skill that helps website owners figure out whether they can actually get cited by AI search engines like Bing Copilot, and if so, what to fix first. The author runs six websites and pulled together 26,767 recorded AI citations across them, then compared pages that got cited often against pages that never got picked at all. Much of the usual advice about schema markup, FAQs, and llms.txt files did not hold up against that comparison. In fact, the author's highest citing page has no JSON-LD schema at all, while some uncited pages have more schema and FAQs than the cited ones. The core idea of this skill is to check demand first. Before recommending any on-page changes, it asks whether people are even asking AI the kinds of questions your site could answer. If nobody is asking, no amount of tweaking will get you cited, and the skill says so upfront instead of wasting your time. If demand does exist, it sorts possible fixes into things worth doing anyway, and the smaller set of changes that actually correlate with getting cited. It also gives you a way to measure whether your changes worked afterward. The workflow runs in six steps: a demand probe, splitting by which AI engine camp you are targeting, a site audit, an entity check, a benchmark against a specific industry vertical, and a final measurement step. A file called RULEBOOK holds every rule along with the evidence and receipts behind it, drawing on the citation counts, a controlled regression test, an 820,000 citation study of the mortgage industry, and a public ledger of predictions made before results were known. To install it, you clone the repository into your Claude Code skills folder, either globally or for a single project. The author is upfront about the limits of the underlying data: it only covers six owned websites across five industries plus one mortgage study, focused on Bing Copilot in English, and the findings describe what was measured on those specific sites rather than universal truths. The included measurement tools are meant to let you test the rules against your own site. The project is released under the MIT license.