009 Agency case studies are reported the way we report to clients: baseline first, then the work, then measured runs with dates. AI answers are volatile, so every result below is a point in a tracked series, not a peak we screenshotted once.
casestudies #1, #3 and #4 are named, with the clients’ approval. casestudy #2 is published under NDA: the figures are real, dated and logged, the identifiers are not shown. every number comes from our logged measurement panels.
casestudy #1 · named, with the client’s approval
An EU crypto-law firm with 1,004 published articles the engines had read and never once recommended. Twenty buyer prompts, four engines, every run dated and logged: zero recall in june, all twenty by august, #1 in ChatGPT, #2 in Europe on Perplexity - cited from the firm’s own page.
casestudy #2 · published under NDA
A licensed casino that every engine warned German players away from - not because it was a scam, but because its own site never said who ran it. Five months of dated runs: the scam verdicts gone in six weeks, then the recommendations.
casestudy #3 · named, with the client’s approval
A Kyiv and Lisbon AI studio whose brand name resolved to an IEEE conference and a UN committee before the company itself, while the engines quoted pages that had stopped existing. Health score 27 → 53 in seven weeks, all four engines naming it inside the first month, and by review 4 first place on the programme’s main buyer question - with the firm’s own article as a source.
casestudy #4 · named, with the client’s approval
A tokenized real estate platform that had just migrated from Webflow to Astro: hundreds of URLs returning errors, a blog Bing had never indexed, and the brand named in 3% of AI answers about its category. Five months in four languages, every change measured against the pages and languages that did not get it.
An enterprise client under NDA, measured on the same frozen-panel method as the three cases above. Re-audited eight weeks after the baseline; first on its buyer questions in the USA and China with Claude and Perplexity. The case is being written and publishes once the client approves.
Published aug 21, 2026, updated sep 17, 2026 with the fourth case and a fifth in progress. Every figure on this page is a snapshot of a dated, logged run; nothing updates by itself. Prompts are run live against the engines, never from cache, and fired ten separate times on each engine per pass.
manual GEO audit · complimentary
We run the audit by hand before we reply, so what comes back is about your domain — not a brochure.