Saying exactly how sure you are — the language that makes an expert trustworthy.
Lesson 3 of The Art of Being Understood · see all five lessons →
Read each part, then pause for the questions. The whole idea turns on a single discipline: matching your words to your evidence.
When the Intergovernmental Panel on Climate Change sits down to summarise the work of thousands of scientists, it faces a peculiar problem. The science is uncertain — usefully, honestly uncertain — and ordinary English is hopeless at conveying degrees of uncertainty. One reader's "likely" is another reader's coin-toss; one person's "very probably" is another's near-certainty. So the Panel did something quietly radical: it built a calibrated vocabulary in which words carry exact numbers.
In its guidance, "likely" does not mean "I have a hunch". It means a probability of at least 66 per cent. "Very likely" means at least 90 per cent. "Virtually certain" is reserved for the rarefied band of 99 to 100 per cent. The terms are defined in a published table, and every author is meant to reach for the rung that the evidence actually supports — no higher, no lower.
The point was never to sound more confident. It was the opposite: to state uncertainty with such precision that a careful reader could tell exactly how much weight a sentence was asking them to bear. The doubt is not hidden. It is measured, labelled, and handed over honestly.
Now watch what a single word does to a finding as it travels. A research team publishes a study; their paper says the data "suggest" a link between two things. By the time the result reaches a headline, "suggest" has quietly become "proves". Nobody lied, exactly. A reporting verb was simply swapped — and the claim's whole standing changed with it.
Reporting verbs sit on a ladder of strength. To say a study "hints at" something is to make a small, hedged claim. To say it "suggests" or "indicates" is to climb a rung. "Shows" and "demonstrates" climb higher still. "Proves" sits at the very top — and in most empirical work it is a rung that almost nothing genuinely reaches. Choose a verb above your evidence and you commit certainty-laundering: a single, tentative study is passed off as settled fact, and the doubt that was honestly there at the start is quietly washed out.
This is why credibility, at expert level, lives in calibration — in matching the strength of your language to the strength of your evidence. Overclaiming destroys trust faster than admitting doubt ever could, because the day the overclaim is exposed, every other thing you have said is re-priced downward. The expert who says "proves" once too often is not believed more; they are believed less, about everything.
There are two ways to miscalibrate, and they are mirror images. The first is the overclaim — the false "proves", the certainty-laundered headline, the flat assertion that outruns its evidence. The second is subtler and is committed mostly by the cautious: the limp over-hedge, the answer so wrapped in "it may possibly perhaps to some extent appear that" that it commits to nothing at all. The over-hedger imagines they are being rigorous. In fact they are being cowardly — declining to say anything a listener could actually use or hold them to.
So calibration is a tightrope, and you can fall off either side. The instruments that keep you balanced are two opposed families of words. Hedges — downtoners — lower the temperature of a claim: tends to, broadly, it appears that, to some extent, arguably. Boosters — intensifiers — raise it: clearly, undeniably, the evidence is overwhelming. The skilled communicator reaches for whichever the evidence demands, and is equally willing to do both. Neither family is virtuous in itself. Used to match the truth, they are precision instruments. Used to flee it, they are how you lie while sounding careful.
The vocabulary of saying exactly how sure you are. This is the metalanguage of calibration — the words you need to talk about the strength of a claim. Tap each card for a precise definition and a usable example.
The same finding, told three ways. Notice that only the calibration changes; the underlying evidence is identical — and only the middle row is honest about it.
| Calibration | The same finding (a single, promising study) |
|---|---|
| Overstated | "This proves that the new method works — the case is closed." |
| Properly calibrated | "An early study indicates that the method may help, though it needs replication before we rely on it." |
| Over-hedged | "It could perhaps be argued that there might, to some limited extent, possibly be some kind of effect of some sort." |
These are the precision instruments of a C2 communicator: the means of saying exactly how sure you are, no more and no less. Try each before you reveal the model answers.
Reporting verbs form a ladder of certainty: hints → suggests → indicates → shows → demonstrates → proves. Each rung commits you to more. The skill is choosing the rung your evidence can actually carry — and "proves" is almost always too high. Match each situation to an honest verb.
1. One small study, not yet replicated: "This a possible link."
2. A large, repeatedly replicated body of work: "Decades of research that…"
3. A consistent pattern across several solid studies, but mechanism unclear: "The data that…"
4. A formal, closed mathematical argument: "The theorem that…"
1. "This hints at / suggests a possible link." (weak, single-study evidence → bottom rungs)
2. "Decades of research show / demonstrate that…" (converging replication earns a high rung)
3. "The data indicate that…" (solid but incomplete — the honest middle)
4. "The theorem proves that…" (the only place "proves" truly belongs — a closed deductive system)
The move: before you choose the verb, ask what your evidence could survive being cross-examined on — and climb no higher than that.
Epistemic modals encode certainty: may / might / could flag a genuine possibility; should flags a confident expectation; must flags near-certain inference. Swapping one for another quietly changes how sure you sound. Choose the modal that fits the certainty described.
1. A 50-50 possibility, no more: "It be the cache."
2. A reasonable expectation, all going well: "The fix hold until Friday."
3. An inference so strong it's almost forced: "The lights are off, so they have left."
1. "It may / might / could be the cache." (open possibility, low commitment)
2. "The fix should hold until Friday." (confident but not guaranteed — note the built-in escape hatch)
3. "…so they must have left." (epistemic "must": near-certain inference from evidence, not an order)
The move: "should" is the most useful and most misread of these — it promises a likelihood, not a certainty. Use it when you mean "I expect, but I'm not staking my life on it".
Hedging devices lower the temperature of a claim to match weaker or partial evidence: it appears that, tends to, broadly, on the whole, to some extent, arguably. Used once and deliberately, a hedge is rigour. Stacked three-deep, it becomes the limp over-hedge. Rewrite each overclaim into a single, well-placed hedge — without burying it.
1. "Remote workers are more productive." →
2. "This policy will fix the housing crisis." →
1. "Remote workers tend to be more productive, at least for focused tasks." (one downtoner + one scope limit — not five)
2. "This policy could go some way towards easing the housing crisis." (commits to a direction, not a cure)
The move: a good hedge narrows the claim to exactly what you can defend, then stops. If a listener can't tell what you're actually claiming, you've over-hedged.
Boosting devices raise a claim: clearly, undeniably, the evidence is overwhelming, beyond serious doubt. Used where the evidence is genuinely strong, they are honest and bracing. Used to bully a weak claim past scrutiny, they are the overclaim — and the first one exposed re-prices everything else you've said. Decide whether each booster is earned or reckless.
1. Smoking and lung cancer, after seventy years of evidence: "clearly linked." →
2. A single preliminary trial of a new drug: "the evidence is overwhelming." →
1. Earned. Decades of converging, replicated evidence can carry "clearly". The booster matches the weight.
2. Reckless. One preliminary trial cannot bear "overwhelming". This is certainty-laundering — and when it fails to replicate, the speaker's every other claim is quietly downgraded too.
The move: spend boosters like rare currency. Each one you can't back draws down your credibility on everything else.
Concessive honesty is counter-intuitively persuasive: by openly naming a claim's limit, you signal that you have one eye on the weaknesses, which makes the strength you do assert more believable. The phrasings — "with one important caveat…", "that said…", "to be fair…", "the one thing that gives me pause…" — let you defend a claim and concede its limit in the same breath. Keep a bank of these ready and reach for them out loud.
Here is a single finding — a new teaching method tested in one promising study — stated at three levels of certainty. Read the overstated version, then watch calibration earn back the trust the overclaim spent.
"We've proven that this method works. The evidence is overwhelming and the debate is over: any school still teaching the old way is simply failing its students. There is no longer any serious question about this."
Every verb outruns the evidence. "Proven", "overwhelming", "the debate is over" — from a single study. The day a replication fails, none of it can be walked back, and the speaker's next claim starts from a deficit.
"An early study indicates that this method may help, and the effect looks meaningful — with one important caveat: it's a single trial, and we won't know how far it generalises until it's been replicated. That said, the result is promising enough to be worth a careful pilot."
"Indicates", "may", "the effect looks" — each rung matched to the evidence. The caveat is named openly, and the closing "that said" still commits to a real, defensible recommendation. It hedges without going limp.
Same finding. Only the calibration changed — and only the second version could survive a hard question.
Now move to the last tab. Everything above is about to be tested out loud.
Do all four, in order. The first warms up the idea; the rest put the toolkit under pressure. The timer is there to stop you escaping into vagueness — fill the time with precisely calibrated claims instead.
Explain calibration in your own words, as if to an intelligent colleague who missed today's session. Cover what calibrated certainty is, why both overclaiming and over-hedging fail, and how the two failures mirror each other.
Use at least five of the Key Language terms (to hedge, a booster, epistemic, to overstate, tentative, categorical, a caveat, to qualify, stance, plausible…) — and notice whether your own claims about the idea are themselves well calibrated.
Choose a finding or claim you genuinely know well — from your field, the news, your own experience. State it three times:
First overstated — "proves", "overwhelming", debate closed. Then properly calibrated — the right reporting verb, the right modal, one honest caveat. Then deliberately over-hedged — wrapped in so many downtoners it says nothing at all.
The goal is to feel the difference in your own mouth — and to notice which version you'd actually believe coming from someone else.
Take a claim you actually hold and defend it out loud — while openly conceding its limit. Build the whole thing around concessive honesty: "the evidence is strong, with one caveat…", "that said…", "the one thing that gives me pause…".
Make the concession real, not decorative. The aim is to sound more credible for having named the weakness, not less.