Methodology

How it works

Each round, we phrase a question that has no single right answer, and pose it to two responders: a human domain expert and an AI model.

The rules

1. Two answers, up to 500 characters each. Both answers are edited for uniformity of form only (length, structure, register) — the content is never changed.
2. Position randomization. Each round, a coin flip decides whether the AI's answer appears as Answer 1 or Answer 2, neutralizing position bias.
3. Fixed AI protocol. One fixed model, one standard prompt, a single attempt — no cherry-picking between generations. Model upgrades are documented in the index.
4. A separate model for phrasing. The model that helps phrase the questions is not the competing model.
5. The expert is anonymous. Experts are presented by field and experience only; their name is revealed only with their consent.
6. Voting happens on social media. Votes are cast as comments (1 or 2) on LinkedIn and Facebook. We archive screenshots before revealing results.
7. Sample threshold. A round with fewer than 100 votes is flagged as a small sample.

Known limitations

Voting is public and therefore exposed to herding bias; the voting audience is not a representative sample of the population but of the series' followers. The index measures this audience's preferences — with full transparency.