Method Note
·17 June 2026·3 min readWhat AI can revise, and what still requires your judgement
AI tools can revise sentences well. Deciding whether a sentence matches the facts, certainty and responsibility of a situation is still the writer’s task.
By Casey Bawden , Creator of Neutral Authority Method
A version of the same question arrives often: “I use AI to rewrite my emails. Why would I need anything else?” It is a fair question. AI tools can do the rewriting well. Ask a current model to make an email more direct, or to remove hedging, and it will generally attempt exactly that.
What the tool cannot do on your behalf is decide whether directness is warranted in a given sentence, with this person, about this subject, and how certain you actually are. That judgement depends on facts the model does not have: your real level of certainty, your actual responsibility for the situation, and the history of the relationship.
The prompt problem
A rewriting tool acts on the instruction it is given. “Make this more confident” will generally produce something more confident, including in places where the underlying position was genuinely uncertain and the original qualification was accurate. To prompt usefully, the writer needs to have already identified which parts of the draft misrepresent the situation. That identification is a judgement call the tool is not positioned to make.
What model output can contain
Language models are trained on large volumes of professional writing, which includes the same constructions this journal examines: reflex apology, habitual hedging, negative framing. A generated draft may contain these, and in many cases appropriately, plenty of instances are accurate to the situation. The output still needs to be read against the facts rather than accepted as finished.
7 Emails, free
Seven ordinary professional emails, with the original, one possible revision and the reasoning behind it.
The PDF arrives by email. You can unsubscribe at any time. See the Privacy Policy.
Situations with no tool available
A good deal of professional communication is not drafted in advance: a comment made in a performance review, a number mentioned in a rate conversation, a two-line reply typed between meetings. A transcript can be run through a model afterwards, but the sentence cannot be checked before it is said. That is a reason to build the underlying judgement rather than relying only on after-the-fact editing.
A practical check
Take a recent email a model has rewritten. Read the output against three questions: does any sentence apologise where no responsibility or impact has been established; does any qualifier understate certainty you actually have; does a negative opening add necessary information or simply frame the message before the facts appear. Apply the same questions to the original draft. This is not an argument against AI tools. It is an argument for being able to evaluate their output, and for having the judgement available when no tool is at hand.
For the underlying definitions, see the apology that runs before you decide to send it and hedging language in professional communication.
Related Reading
- Method Note·29 May 2026·2 min read
When tone is only part of the communication problem
Tone advice addresses how something sounds. Wording, certainty, responsibility and context often need separate attention.
- Method Note·29 May 2026·2 min read
What general communication training may not address
Soft-skills training builds valuable behaviours. Sentence-level wording is a separate layer that broad training often doesn’t reach.
- Method Note·25 April 2026·3 min read
Acknowledgement and apology do different work
Acknowledging impact and accepting responsibility are not interchangeable. Each is appropriate in different circumstances.