Data Story

Our first GEO project started in May 2025. By the middle of 2026 the work was running across more than 20 clients. Before July 2026, though, datastory.nz still had no pages about generative engine optimisation. An AI engine looking for evidence of who did this work in New Zealand had little reason to mention Data Story, even while we were busy doing it.

Webinars and speaking engagements, including the iMedia Future of Marketing Summit New Zealand, were generating plenty of interest in GEO and AEO. Dave Hockly was talking about the subject on stage and continuing the conversation on LinkedIn. Data Story had become increasingly vocal about the work. Its website, meanwhile, remained conspicuously silent.

Call it the builder's-house problem. We were helping clients become visible in AI search while our own digital footprint had been left half-finished.

There was another gap. For a business that sells measurement, we had surprisingly little data on our own visibility in the category. We could point to client work. We could explain the methodology. What we could not do was answer the obvious question, does this work for you, with evidence from our own site.

This was not a technical problem. There was nothing on the site to optimise, because there were no GEO pages to optimise. When a business sells a service its own website never mentions, the fix is to write the pages, not to tune the ones already there.

We published four pages. A service page for someone evaluating a GEO agency. An engagement overview for what happens after they buy. And two articles on the question that comes before either decision: can AI-search visibility be measured, and how you would measure your own brand in AI search.

Aside from telling people what we are doing in the GEO space, we had a second motive: to treat the new pages as a test of what happens when a website enters a category it previously had no content for. Generative engine optimisation is a new specialty, and on our own site we can be explicit and honest in a way we cannot be with a client's. So we are the perfect test dummy.

We used our own measurement framework. That involved building a prompt library for Data Story: ten topics across four clusters, mapped to the stages of a prospect's evaluation journey. Rather than following search volume, the library followed the questions a buyer might ask, from understanding the category to evaluating methodology and fit.

We tested 40 prompts across eight AI engines, using the same prompts and the same engine set for every reading, and checked each result against a second, independent measurement tool.

We also measured across five consecutive days to establish how much the number moves when nothing has happened. That gave us a noise floor, so later movement could be judged against ordinary fluctuation instead of being declared a breakthrough.

There was no attempt to flood the site with content. No volume play, no ranking trick, no sprawling content calendar. Four pages went live, each answering a specific buyer question. Then we watched what happened.

Between 7 August and 30 August 2026, our citation rate across the four tracked questions rose from 14.7% to 23.1%. The largest movements appeared in the questions most directly addressed by the new content. "Who can set up AI-search measurement and reporting for a New Zealand business" went from 6.2% to 23.8%. "Who offers generative engine optimisation services in New Zealand" went from 5.0% to 13.8%. "What should you look for in a GEO or AEO agency" went from 3.8% to 7.5%.

The fourth question acts as a control. "Best data-led marketing agencies in New Zealand for analytics and measurement" does not mention GEO at all, it is brand-adjacent, and we were already doing well on it. It moved only slightly, from 43.8% to 47.5%. That matters, because if every topic had jumped at once it would be harder to tell whether the movement had any relationship to the new subject matter or simply reflected a wider change in the engines.

The new GEO service page showed the clearest pattern: flat across the first two readings, then cited about three times as often by the end of the month.

We also analysed the citation rate delay. Citation rate stayed flat through the first 11 days of measurement, and the larger movement appeared about six weeks after publication. Entering a category was not instant. Part of the movement also coincided with a broader change in what the engines were citing: nine of the fourteen most-cited domains in the study fell over the same window while ours rose.

A second measurement tool placed the result within 2.8 percentage points of the first, and the position held across five consecutive daily readings. That makes it less likely the increase was a one-day fluctuation or an artefact of a single platform.

At 9.1%, there is obviously still a lot of work to do. We think we punch above our weight on GEO as an offering and a service. Ironically, we just need AI to work that out.