Not sure where to start? Take a guided tour.
AI Site Grader
AI Query Scanner
llms.txt GeneratorBest-of prompts carry a high-divergence tail
Best-of prompts show a 14.8 percent high-divergence rate in the model divergence study.
The study Same Question, Different AI, Different Answers evaluated model consistency and found that best-of prompts frequently split models.
The numbers behind it.
- Best-of high divergence
- Best-of prompts frequently split models.
- 14.8%
The evidence.
| Measure | Result |
|---|---|
| Best-of high divergence Best-of prompts frequently split models. | 14.8% |
Population and method
Built from 797,644 valid comparisons across 44,088 reports and 8 models, covering 6,439,133 model responses in the observed window.
Where the evidence stops
Agreement is measured across recommendation outputs, not across hidden reasoning or retrieval context.
Average agreement can hide large differences between query classes and model pairs.
The study measures overlap, not which answer was objectively “right”.
What to do with this
Strategists should treat best-of content as a high-risk visibility category that requires diversified optimization tactics to account for inconsistent model outputs.
Cite this fact.
Source study: Model Divergence“Best-of prompts show a 14.8% high-divergence rate in the model divergence study.”
Same Question, Different AI, Different Answers - Trakkr Research · Best-of prompts carry a high-divergence tail
Same Question, Different AI, Different Answers · Trakkr Research · Updated 11 March 2026