A fintech company's reported AI share of voice increased from 36% to 46% during a three-month content pilot. Pages in its content hub received 5.643 times the average citation count of other pages.
I worked on making the company's content clearer, better connected, and easier to use as a source. This case study covers what I changed and what the results tell us. The company is anonymous for confidentiality.
The challenge
The company operates in digital mortgage brokerage. It wanted to understand how often its website appeared in AI answers to questions about its products and market, and whether those appearances led to visits or business results.
Unlike a search ranking, an AI answer can change when someone repeats the same question. The wording, the AI tool, and updates to its underlying model can all affect which sources appear.
I focused on two questions: was the company appearing more often in the monitored answers, and were the pages I worked on receiving more citations?
What I changed
I reviewed the existing pages against relevant topics and customer questions, then selected the questions to monitor.
I built a content hub, a set of connected pages covering related topics. I added links between the hub and supporting pages so readers could move between them more easily.
I also rewrote pages with clearer headings and direct answers, and added accurate structured data: machine-readable information describing what each page contains.
This work falls under generative engine optimization (GEO) and answer engine optimization (AEO), which aim to improve how content appears in AI-generated answers.
The aim was to make useful answers easier to find, understand, and reference.
How I measured visibility
I used a monitoring provider's results across a fixed set of questions and competitors over three months.
The provider's AI share of voice metric tracked the company's appearances in sampled answers, with some questions weighted more heavily than others. It describes that monitored sample, not all AI searches.
I also compared citations to content-hub pages with citations to other pages. A citation means an AI answer used or linked to a page as a source.
Website analytics tracked visits from known AI referrers separately from citations.
The results
AI share of voice: 36% to 46%
The reported increase was 10 percentage points, or 27.8% relative growth from the starting figure.
The company appeared more often across the monitored answers by the end of the pilot.
Content-hub pages: 5.643 times the average citations
Content-hub pages received 5.643 times the average citation count of other pages.
Most of that difference came from a small number of heavily cited pages, rather than an even increase across the hub.
A separate before-and-after comparison
A separate before-and-after analysis also recorded an increase. It measures a different result from the provider's AI share of voice and is presented separately.
Measure | Result |
|---|---|
Matched observations | n = 30 |
Before | 82.27% |
After | 83.73% |
Paired t-test | t(29) = -1.408, p = .170 |
Measurement note
These are observed results from a three-month pilot without a control group, not an estimate of growth caused solely by the work. Neither the page-group comparison using Welch's t-test nor the paired test above reached statistical significance. The provider's share-of-voice figure covers its monitored sample and comes without an uncertainty range.
Business figures for sessions, ROI, customer acquisition cost, and branded clicks remain unverified. The citation counts 4,027 and 4,095 have unresolved data scopes and are excluded from the headline findings. A sparse 4×2 table requires an exact or simulation-based test. The separate paired analysis also depends on comparable before-and-after observations.
Conclusion
For founders, the distinction is straightforward: appearing in more AI answers, getting more website visits, and winning more customers are separate outcomes. Each needs its own evidence.
The main result was a 10-percentage-point increase in reported AI share of voice, alongside a higher average citation count for content-hub pages. Those findings give me a clear focus for the next phase: track which pages earn citations consistently and whether that visibility brings qualified visitors.
