“Probably” is doing a lot of work in that sentence.
Decades of research on estimative language shows the same phrase lands as wildly different probabilities in different heads. Paste a sentence — an email, a forecast, a doctor’s note — and see the approximate ranges readers actually assign to your hedge words. Then say the number.
The lexicon
| Phrase | Approx. range | Median ≈ |
|---|---|---|
| almost certainly | 90–99% | 95% |
| highly likely | 80–95% | 90% |
| very likely | 75–90% | 85% |
| likely | 60–80% | 70% |
| probably (probable) | 60–80% | 70% |
| we believe | 60–85% | 75% |
| better than even | 55–65% | 60% |
| toss-up (coin flip) | 45–55% | 50% |
| real possibility | 20–80% | 45% |
| serious possibility | 20–80% | 50% |
| possibly (possible, might happen, may happen) | 20–60% | 40% |
| cannot rule out (can't rule out) | 5–40% | 20% |
| unlikely | 15–35% | 25% |
| probably not | 15–35% | 25% |
| highly unlikely | 5–15% | 10% |
| little chance (slight chance, chances are slight) | 5–15% | 10% |
| almost no chance (virtually impossible) | 1–5% | 3% |
Honesty note (read this): the ranges above are approximate syntheses of published estimative-language research — Sherman Kent, “Words of Estimative Probability” (CIA Studies in Intelligence, 1964, declassified); Wallsten et al. on the vagueness of probability phrases (1986); and the survey popularized by Mauboussin & Mauboussin, “If You Say Something Is ‘Likely,’ How Likely Do People Think It Is?” (Harvard Business Review, 2018). Values are rounded to 5s and vary by study, context, and population — treat them as calibrated ballparks, not decimals, and check the originals. That imprecision is, of course, the point of the tool.
Related: lock a numeric call with VERD Receipts · measure your own calibration on the Scorecard · audit an AI’s hedging with Second Opinion.