Are Calls Getting Easier to Criticize? A 20-Year Trend Across 165,000 Calls

When an AI read 165,182 earnings-call transcripts and scored each one on candor, evasion, specificity, and related traits, it flagged 145,051 calls — about 88% of the corpus — as containing at least one criticizable moment: an evasive answer, a dodged question, or a vague commitment. If you follow earnings calls closely, that near-universal rate raises a fair question: is "criticizable" a meaningful signal, or just the normal texture of how executives talk under pressure?

The flagged calls look slightly different from the rest: they score a bit higher on evasion (2.79 vs 2.70) and stress (2.58 vs 2.43), with marginally lower candor (6.83 vs 6.86). The year-by-year trend is mostly flat, hovering between 84% and 91% since 2015, so calls have not become dramatically easier to criticize over two decades. The returns gap is modest: flagged calls saw a median next-period return of -8.18% versus -7.16% for all calls, with 38.4% beating versus 39.5% overall.

These AI-read fields are noisy, and the return figures are descriptive history from a 22,449-call sample, not a trading signal — our own forward tests falsified directional prediction. Language models also partially remember famous stocks' histories, which contaminates any backtest of this kind.

Behavior profile

Guidance actions in this group

Share of calls by year

Recent examples