Research
How AI decides who to recommend. Measured, not guessed.
We run live panels across six engines and publish what we find, graded by how strong the evidence is. The vendors write listicles that rank themselves first. We write the independent version. If it's in here, we watched it happen.
The AI Citation Durability Report: what held on the next measured date
385 first-position opportunities, six engines, and the full observed fractions. What persisted on the next measured date, without a lifetime extrapolation.
Read →Field studyHow AI engines choose which local business to recommend
Six engines, six different source stacks. Why ChatGPT leans on directories, Gemini on your Google profile, and Perplexity on whatever it can quote.
Read →ResearchHow much do AI engines agree? About 12% set overlap in our archive
2,356 archived runs, separate app/API labels, and anonymous counts you can recompute: engine answer sets overlap little.
Read →Field studyField study (2026): how AI engines pick a med spa — the injector, the price sheet, and the license check
Two metros, 400 scored runs: AI recommends the named injector with the published price sheet — and now checks the state license registry before it answers.
Read →Field studyField study: the ChatGPT API and the ChatGPT app recommend different businesses
We measured it: the ChatGPT app and the OpenAI API recommend different local businesses. Bare-API overlap with the app was ~8%; browsing doubled it and still missed most picks. The numbers, per vertical.
Read →Field studyField study (2026): how AI engines pick a personal injury lawyer — the credential check nobody's optimizing for
Two states, 440 scored runs: in PI, the engines discount ads, verify board certification in the state registry, and quietly maintain a tier of firms they read but never recommend.
Read →BriefingWhen people ask AI to hire a professional
The 'nobody asks a chatbot for a lawyer' objection, answered with the engines' own usage data — and why professional services are where AI hiring lands first.
Read →Tool teardownIs a rank tracker enough to see your AI visibility?
The popular local tools added AI. We pulled one apart against primary sources: what it measures, what it can't see, and where a report tells you more.
Read →BriefingThe API is not the app: what that means for measuring AI visibility
Every engine ships two products under one name — a developer API and a consumer app — and they recommend different businesses. Why that quietly breaks most AI-visibility measurement, engine by engine, with the one exception.
Read →BriefingWhich websites AI engines cite most — and how to earn a citation
Mid-2026 analyses put Reddit, Wikipedia, YouTube, and the big directories at the top of every AI-citation list. The useful move isn't cracking that leaderboard — it's being the cited source for your category, in your city.
Read →PlaybookHow can I make my business show up in AI?
The complete playbook: what decides whether an AI names you, and the fixes in order of leverage.
Read →More field notes publish as panels complete. Every claim carries its evidence grade; when a finding dies on re-measure, we strike it and say why.