Research
Original studies on how ChatGPT, Gemini, Perplexity, and Claude decide which brands to recommend. Each one is published with its method, its limits, and its data, so you can check the work.
The AI Verdict Study
September 2026 · 576 buyer conversations · 18 brands · ChatGPT, Gemini, Perplexity, and Claude
I put 18 brands through 576 simulated buyer conversations on ChatGPT, Gemini, Perplexity, and Claude, same wording every time, with each buyer written to fit the challenger brand. The category leader got turned down about twice as often as the challenger, and in 21 of 72 brand and engine pairings the same buyer got a yes on some runs and a no on others. Every chart, the method, and the data are public.
How I run these studies
Same buyer, same words, every run. Each study fixes the buyer's wording and repeats it across engines and across days, so the differences come from the engines, not the prompts.
An automated classifier labels every verdict, the labels are checked for contradictions, and two human raters score a blind sample. Each study reports how often they agreed, the confidence intervals, and the limits of the method. The full data is published alongside the charts.
Want the next one when it's out? Sign up below. Want your own brand run through the same test? See the AI Visibility Audit →