Rand Fishkin

Rand Fishkin, cofounder and CEO of SparkToro, co-led one of the most consequential measurement studies in generative engine optimization. With Patrick O'Donnell of Gumshoe.ai, he recruited 600 volunteers to run 12 recommendation prompts through ChatGPT, Claude and Google's AI systems a combined 2,961 times.

The headline result was stark. The same brand list came back in any two of 100 responses less than one time in 100. That finding undercut the comfort of treating AI answers like a stable SERP you can rank-track once a week — and it is the core of SparkToro's public report.

For a field that was rapidly inventing dashboards, the study arrived as a cold shower. If answers reshuffle constantly, then a single screenshot is close to useless as proof that a GEO program worked.

Fishkin has spent years arguing that marketers should measure what audiences actually do, not what dashboards imply. The AI inconsistency study fits that career pattern: it replaces a comforting fake precision with a messier truth that is still actionable if you sample enough runs.

After the report, serious teams shifted how they talk about AI visibility. Instead of asking who ranks first in a single capture, they ask how often a brand appears across many runs, prompts and sessions. SparkToro published raw data alongside the write-up, which made the study a reference point rather than a marketing claim.

Fishkin's broader career in SEO research and audience intelligence gives the work extra weight. When he says measurement must change, buyers already know his habit of publishing methods in public. That reputation helped the inconsistency finding travel outside the usual SEO Twitter circle and into strategy meetings.

The study also raised the bar for vendors. Tools that still sell a single "AI rank" without showing variance now look incomplete next to a public dataset that documented how rare perfect repeats are.

Publishing the raw data invited replication and secondary analysis. That openness is why the study still gets cited when newer tools claim they have solved AI rank tracking.

Read SparkToro's report on AI recommendation inconsistency, including the public raw data. It is still one of the clearest explanations of why single-run AI rank tracking misleads, and why multi-run sampling belongs in any GEO measurement plan.