Patrick O'Donnell

Patrick O'Donnell is cofounder and CTO of Gumshoe.ai. He co-led the AI recommendation consistency study with Rand Fishkin, the volunteer-driven project that ran thousands of recommendation prompts across major AI surfaces.

The study's low repeat rate for identical brand lists forced a measurement reset. Visibility in AI answers had to be discussed as a distribution across runs, not a single stable rank.

O'Donnell's engineering role on the project matters. Methodology notes and raw data do not publish themselves; someone has to care enough about reproducibility to ship them with the narrative.

For Gumshoe customers and competitors alike, the study became a yardstick. If a tracking product cannot show distribution across runs, it is selling a simpler story than O'Donnell helped prove is false.

He sits near Fishkin on this list for a reason: the consistency study is a joint artifact, and O'Donnell's engineering and publishing choices are why the inconvenient findings stayed attached to usable data.

The work drew extra attention because Gumshoe sells AI tracking tools, and the findings undercut naive rank-style tracking. Publishing methodology notes and raw data made the study a standard that other visibility tools get judged against. SparkToro's public write-up remains the narrative entry for most marketers.

As an arbiter on this list, O'Donnell represents measurement honesty: show the variance even when it complicates the product story. That posture builds trust with buyers who have learned to distrust category marketing.

Teams evaluating AI visibility software now ask whether a vendor accounts for inconsistency the way the SparkToro and Gumshoe work documented. That question is part of O'Donnell's legacy on the list.

The volunteer design also made the work harder to dismiss as an in-house lab artifact. Hundreds of people ran the prompts, which is part of why the inconsistency finding stuck in industry memory.

That yardstick helps buyers ask better demos. Show me the distribution across runs, or explain why your product still speaks as if AI answers were stable rankings.

Start with the study's methodology notes, published alongside the raw data. They explain how the volunteer design worked and why the inconsistency claim is hard to dismiss, even for readers who skip the marketing summary.

Arbiter status here is also about tooling norms. O'Donnell helped make public methodology a competitive expectation, which is a healthier equilibrium than a market of black-box AI rank scores.