Research on AI search and technical SEO
I publish what I measure. The audits and recommendations on this site are built on the research below and on Raptor, the open source crawler I wrote in Rust. The client in the field report is a veterinary practice, not a software company: the method carries over to software, and the numbers are theirs.
Generative engine optimization in 2026
How ChatGPT and Google AI Mode choose the sources they cite, measured on a live client, a regional veterinary practice, with SEMrush Brand Performance data, query fan out captured from four AI engines, and traditional ranking data for the same queries.
Read the field reportKey findings
AI search and Google reward different things
The client ranked first on more than 30 ChatGPT prompts about its specialty while not ranking at all on Google for the matching typed search. Natural language prompts favor rich, specific pages; typed search still favors tight exact match optimization.
Citations concentrate on a few pages
On the client domain, six pages accounted for over 90 percent of all AI citations: a service page, the About page, one blog post, two FAQ pages, and the homepage.
Each engine cites differently
Google AI Mode leaned on aggregators first, while ChatGPT cited the client's own domain more than any other source. Off page work has to be planned per engine.
The AI feedback loop is faster
New content typically appeared in AI citation sources within 2 to 4 weeks. Significant Google movement on competitive terms took 90 to 180 days.
How I run the research
The same discipline goes into every client audit and AI visibility report.
Fixed prompt sets
The same questions run on every engine, every cycle, so a change in the numbers reflects a change in the world rather than a change in the questions.
Trends over snapshots
AI answers are not deterministic. I read patterns across 4 to 12 weeks of runs and only report what holds across several of them.
Real client data, anonymized
Findings come from live engagements, with the client whitelabeled. Nothing is simulated.
Limitations stated up front
Each report says what the data cannot show, such as changes to a tool's query panel between runs.
The tooling behind the research
I do not rent the audit. Raptor is the open source SEO crawler I wrote in Rust, with 128 checks covering rendering, indexation, structured data, and performance. My field report on generative engine optimization uses data from a live client, a regional veterinary practice, to show how ChatGPT and Google AI Mode choose the sources they cite.
Checks in Raptor, the open source crawler I wrote in Rust.
AI search engines on every prompt set: ChatGPT, Perplexity, Gemini, Claude, and Google, where I check both AI Overviews and AI Mode.
Senior person on your account. Me, from the first crawl to the monthly report.
Put the research to work on your site
Book the free 30 minute intro call. I will tell you what the research suggests for your site and whether there is a fit for an audit or retainer.