Many GEO case studies display visibility theatre. They show a screenshot of an AI response with a brand name, and there’s no connection made to any specific question. We believe that is the wrong way to measure success.
We are happy to endorse this version: what we intended to test, the metrics that changed, the metrics that remained stagnant, and a candid review of what we would change. If you want to gauge your position before making a commitment, our… Book a GEO Audit page is the natural next step.
The client was respected in their industry, had named clients and an actual track record — but when buyers asked ChatGPT, Perplexity or Google’s AI Overviews for a shortlist, the client did not show up. Their rivals did. There was Demand; the AI’s brokering that demand just didn’t show them.
Prior to engaging with the content, we completed a positioning check. Spending for vague positions, “More AI visibility” is a lost cause. We clarified who they were better than, what attributes actually mattered, which buyers would care, then built the GEO work on that clarity, not instead of it.
The hypothesis we tested
“Should we provide this client with the most citable and best-evidenced response to the ten buying-intent questions their prospects pose to AI engines, a measurable portion of those AI conversations — not just impressions — will convert into qualified enquiries.”
Every campaign is a hypothesis. This one included a falsifiable prediction and a metric that would have proved us wrong if we were.
Think & Plan
We charted the genuine inquiries buyers pose to generative engines at each level of intention, then reviewed how each engine responded to them and which parties were mentioned. We found twelve authoritative gaps — instances where the client should have been the most obvious cited source and was not.
The experiment design was intentionally narrow: concentrate effort on the ten highest commercial-intent questions, provide observable, organized, and sincerely valuable responses, and provide the means to measure and track each response to the origin.
Do & Review
Over nine months we created and organized material for each of those ten questions, gained third-party citations to bolster authority, and refined the on-site journey so an AI-referred visitor could self-qualify in a matter of moments.
Each month we reviewed the hypothesis, not that of the vanity metrics, and of the ten questions there were two that never moved. Rather than documenting them and dropping them, we were honest: even if a campaign appears to be a success, partially failing because there is nothing to learn from it is a genuine campaign failure.
Reporting the 31 citations as wins would have been easy. But, a citation that the buyer never acts on is decorative. The metric that was relevant was demand versus capacity — enquiries per available slot — because that’s the ratio that grants a business the power to dictate prices and select clients.
The 1.6x increase in enquiries-per-capacity is the figure the client’s finance team was truly concerned about and is the figure we would defend to the end. For a complete defense of which GEO metrics should be on a dashboard, refer to our sister piece Measuring GEO ROI: The Metrics That Actually Matter..
The GEO approach, transparently
-
Real authority, not manufactured signalsEach citation that the engines acquired was linked to real, useful, and verifiable resources and garnered legitimate outside citations. No phony statistics, no authority hijacking. If a buyer were to reverse engineer our strategy for gaining a citation, they would conclude it was legitimate.
-
Serving the buyer where they actually wereVisitors referred by AI come in with a decision rather than cold. We constructed the pathway to facilitate self-qualification and fit assessment both ways, honestly, instead of pushing to complete a form before they were ready.
-
Sustainability as a filter, not an afterthoughtKeeping the Triple Bottom Line in focus, content produced and hosted meant growth that could not be justified in regard to people and planet did not represent growth we sought for the client.
-
Reviewed under Zero BlameThe eight questions that moved were accompanied by an equal number of unexamined questions that did not move. That is how a campaign gathers wisdom rather than folklore and how the following engagements begin more intelligently.
Being honest means acknowledging what was missed. Two out of ten questions were too general — the engines favoured encyclopaedic sources rather than a specialist, and no pile of quality opened that gap. We’d de-scope those earlier and reinvest the effort into tighter, higher-intent questions where a specialist can truly be the best answer.
We’d also put in attribution from day one rather than month two; the 22 figure is an underestimate because the early AI-referred enquiries were more difficult to track cleanly. The actual figure is probably higher, but we only report what we can be confident in.