Clearer Is Not More Certain

Editors are often taught to remove words such as may, probably, and I think. Sometimes that helps. The phrase is only a warm-up, and the useful sentence is waiting underneath.

But sometimes those words identify exactly how certain the writer is, who is making the judgment, or whether an action is optional, recommended, or required. Removing them can make the sentence cleaner while making the claim less accurate.

The editor’s job is not to defend every hedge. It is to tell disposable framing from meaningful qualification.

May, probably, I think, it seems, should, might, and perhaps often get grouped under labels such as hedging, hesitation, weak language, or filler.

Yet one may mark uncertainty, another permission, another capability, and another personal judgment. A qualifier can locate responsibility, express a recommendation, describe risk, or show the limits of the evidence.

A word is not filler merely because the sentence becomes shorter without it.

May describes a possible outcome here. Can may suggest ability, authority, or permission. The sentences are nearly the same length, but they do not say the same thing.

The implementation history behind this distinction is documented in Directness Without False Certainty.

The condition remains in both sentences. The confidence level does not. Probably describes a likely delay; removing it presents the delay as definite.

A more confident sentence is not necessarily a more accurate sentence.

The first sentence identifies readiness as someone’s assessment. The second presents it as established. In workplace, technical, policy, and editorial writing, attribution can matter. Removing the speaker may also remove responsibility.

An editor should not turn a judgment into a fact merely because the fact sounds cleaner. That does not mean I think must always remain. It can go when the surrounding context shows that it is empty framing and the claim stays equivalent.

May not approve leaves the outcome uncertain. Cannot approve suggests inability, lack of authority, or prohibition. The edit changes the whole situation, not merely its tone. Broad replacement is especially risky when negation is involved.

The verb is similar, but the force changes. The first offers a suggestion; the second gives an instruction.

The announcement frame can go. We, should, and before recording remain. Revision should preserve who is acting and how strongly the action is proposed.

The conversational opening changes. The requested action moves forward. The words may not survive, so the question inside the request remains intact.

Any revision process—human or automated—can overvalue brevity, confidence, active voice, directness, or visible change. A process rewarded for producing a firmer version may remove words without determining what those words contribute.

A human editor can make the same mistake by applying a useful guideline in every context. The problem begins when a style preference is treated as permission to change the claim.

SecondDraft stopped globally deleting or replacing several qualified expressions. In tested contexts, it preserves uncertainty, recommendation strength, actors, conditions, timing, and protected values. It can still shorten supported request frames, and sometimes it deliberately leaves a sentence unchanged.

PasteLint Clean handles deterministic mechanical cleanup before editorial revision. The Text Readiness Framework places both steps in the larger Paste → Clean → Revise → Prepare workflow.

Editing quality is not measured by edit count. Leaving a meaningful qualifier intact is not timidity; it can be the only defensible choice. A revision may improve the framing while refusing to increase certainty.

The goal is not to make every sentence firmer. The goal is to make every change defensible.