Rehearsal Season: The Earnings Calls That Sound a Little Too Polished
We tracked 63,830 earnings calls (out of 165,182) where the model answered YES to "Calls That Read Rehearsed" — 38.64% of the corpus, with a 95% CI of 38.41% to 38.88%. The share climbed from 38.04% of calls in 2015 to 45.01% in 2025. Rehearsed calls score higher on promotion (+0.47) and evasion (+0.11) but lower on candor (-0.34) and specificity (-0.18) than the rest of the corpus. They also lift classic tells: Scale-Dependent Advantage Claims appear 1.7x more often, and The Finished-Story Tell 1.37x more often. Their beat rate was 37.23% vs 39.47% for other calls.
- 38.64% of all 165,182 calls were flagged as reading rehearsed (95% CI 38.41%-38.88%).
- The rehearsed share rose from 38.04% of calls in 2015 to 45.01% in 2025, peaking across 2022-2025.
- Rehearsed calls score higher on promotion (+0.47) and evasion (+0.11) but lower on candor (-0.34) and specificity (-0.18) than other calls.
- Rehearsed calls beat guidance expectations 37.23% of the time versus 39.47% for other calls, with a median one-day return of -0.0893% versus -0.0716%.
1Introduction
Anyone who listens to hundreds of earnings calls develops an ear for the difference between a management team thinking out loud and one reciting a script. The scripted version sounds smoother, but smoothness can be a strategy: prepared language tends to emphasize wins, blur specifics, and close off follow-up questions. Whether that style is becoming more common — and what it coincides with on the call itself — is worth measuring rather than guessing. This study examines all calls from 1990 to 2026 where the model answered YES to "Calls That Read Rehearsed": 63,830 calls, or 38.64% of the 165,182-call corpus, tracked by year.
2Data & methodology
The corpus comprises 165,182 earnings-call transcripts published between 1990 and 2026, each scored independently by a large language model on an identical 37-field battery: seven categorical business verdicts, eight 0–9 behavioral meters, and twenty yes/no judgments. The study group is defined as calls where the model answered YES to "Calls That Read Rehearsed", tracked by year (n = 63,830; 38.6% of the reference set, 95% Wilson interval 38.4%–38.9%). Baseline figures use all scored calls. Market outcomes join a fixed sample of 22,449 calls with twelve-month total returns in excess of SPY, measured from the first close after each call; this sample skews toward liquid U.S. names and is reported as descriptive history only.
3Results
Rehearsed calls read differently in every behavioral dimension: they score higher on promotion (5.52 vs 5.05) and evasion (2.81 vs 2.70), and lower on candor (6.52 vs 6.86) and specificity (7.38 vs 7.56). The tells align — Scale-Dependent Advantage Claims appear 1.7x more often and The Finished-Story Tell 1.37x more often than in other calls, while "Pricing Recovering" appears 0.68x as often. The trend is steadily upward: 38.04% of calls in 2015, dipping to 32.37% in 2020, then rising every year since to 45.01% in 2025. On outcomes, rehearsed calls were associated with guidance lowered 9.49% of the time versus 11.56% for others, and beat 37.23% versus 39.47%.
| Meter | Study group | Baseline | Δ |
|---|---|---|---|
| Candor | 6.52 | 6.86 | -0.34 |
| Evasion | 2.81 | 2.70 | +0.11 |
| Specificity | 7.38 | 7.56 | -0.18 |
| Stress | 2.49 | 2.43 | +0.06 |
| Promotion | 5.52 | 5.05 | +0.47 |
| Confidence | 7.26 | 7.21 | +0.04 |
| Action | Study group | Baseline |
|---|---|---|
| Raised | 18.5% | 21.1% |
| Maintained | 48.6% | 48.8% |
| Lowered | 9.5% | 11.6% |
| Withdrawn | 1.8% | 2.7% |
| Signal | Lift | In group | Baseline |
|---|---|---|---|
| Scale-Dependent Advantage Claims | 1.70× | 18.8% | 11.1% |
| The Finished-Story Tell | 1.37× | 6.0% | 4.4% |
| When the CFO Dominates | 1.34× | 18.9% | 14.1% |
| A Tiny Fraction of the Market | 1.32× | 39.5% | 30.0% |
| Pricing Recovering | 0.68× | 14.7% | 21.5% |
| Statistic | Study group | Returns sample |
|---|---|---|
| Median excess return | -8.9% | -7.2% |
| Interquartile range | -28.3% to +10.7% | — |
| Share beating SPY | 37.2% (95% CI 36%–38%) | 39.5% |
| Observations | 7,895 | 22,449 |
| Ticker | Quarter | Call date | Call grade |
|---|---|---|---|
| USCB | Q2 2025 | 2025-07-25 | B+ |
| HCA | Q2 2025 | 2025-07-25 | C |
| AON | Q2 2025 | 2025-07-25 | C |
| BFH | Q2 2025 | 2025-07-25 | B |
| FFIC | Q2 2025 | 2025-07-25 | B+ |
| OMF | Q2 2025 | 2025-07-25 | A |
| GBCI | Q2 2025 | 2025-07-25 | A |
| LARK | Q2 2025 | 2025-07-25 | B |
4Discussion
The honest reading is descriptive: calls the model labels rehearsed tend to feature more promotional, less candid, less specific language, and that labeling has become more common over the past decade. A careful reader should not conclude that rehearsal causes worse outcomes, that the style predicts stock returns, or that the model reliably detects intent — a polished speaker may simply be disciplined. The returns and guidance gaps are associations measured after the fact, not evidence of any edge, and the differences are modest in absolute terms.
5Limitations
The "rehearsed" flag comes from AI-read judgments and is inherently noisy; it may partly reflect disclosure style or sector conventions rather than rehearsal. The returns sample covers only 22,449 calls with non-null outcomes, skewed toward liquid names, so the beat-rate and return comparisons may not generalize. Our own forward tests falsified directional prediction from these signals, and LLMs partially remember famous stocks' histories, contaminating any backtest. Nothing here supports a trading rule. See the full methodology, including the C1 pattern’s forward-test failure and the LLM-memorization finding.