In this guide
  1. Start with the reader and the relationship
  2. Give the draft something real to say
  3. Create a small voice card
  4. Draft for structure before style
  5. Remove the language that gives AI away
  6. Read it aloud and restore your judgement
  7. Keep facts, confidentiality and accountability human
01

Start with the reader and the relationship

Before drafting, name the reader and your relationship to them. A colleague who knows the project needs different context from an executive seeing it for the first time. A customer message needs a different level of explanation from an internal handover. Decide what the person should understand, feel or do after reading.

Tone is not a collection of adjectives. ‘Professional, warm and engaging’ can still produce empty language. Give the model concrete signals instead: direct but respectful; use ordinary workplace language; lead with the request; explain the reason in one sentence; avoid slogans and exaggerated praise.

02

Give the draft something real to say

Generic input produces generic prose. Supply the facts, your conclusion, the trade-off and the next action. Separate verified information from your interpretation. If you have not decided what you think, ask AI to organise the material or surface questions before asking it to speak for you.

A human voice contains judgement: what matters, what can wait, what you recommend and where you remain uncertain. Add those decisions yourself. Do not use fluent wording to disguise missing evidence or an unresolved position.

03

Create a small voice card

Write five instructions that describe how you communicate at work. Keep them observable. For example: use short openings; prefer specific verbs; acknowledge uncertainty directly; avoid motivational language; end with an owner, decision or next step. Add two phrases you naturally use and three expressions you never want.

Use the card as guidance, not a costume. Your tone should still change with the audience and the seriousness of the message. Review it every few months using messages you wrote yourself and still consider effective.

04

Draft for structure before style

Ask first for a plain outline or a short factual draft. Check whether the order makes sense: purpose, essential context, evidence, request or next step. Once the logic is right, ask for a tighter or more conversational version. Styling a weak argument only makes the weakness harder to see.

When several versions are useful, request meaningful alternatives: one that leads with the decision, one that leads with the risk, and one suitable for a brief verbal update. Compare how the emphasis changes. Do not ask for endless variations when you already know the message.

05

Remove the language that gives AI away

Delete throat-clearing openings, repeated conclusions and phrases that sound impressive without adding meaning. Replace abstract nouns with people and actions. ‘The implementation of process optimisation was undertaken’ becomes ‘We changed the handover process.’ Use the terms your team actually uses, but spell out jargon for readers outside the team.

Watch for false warmth as well as stiffness. Excessive enthusiasm, decorative adjectives and automatic compliments can sound less human than a calm, specific sentence. Respect the reader by getting to the point.

06

Read it aloud and restore your judgement

Read the draft aloud once. Mark any sentence you would not say, any claim you could not defend and any transition that feels too polished for the situation. Shorten long sentences, vary the rhythm and bring the key request forward. Add one relevant detail only you would know, provided it is appropriate to share.

Then compare the draft with your original purpose. Has the wording made you sound more certain, enthusiastic or critical than you intended? Has it removed a necessary disagreement? Put your position back. A recognisable voice is not casual writing; it is clear ownership of the message.

07

Keep facts, confidentiality and accountability human

Check names, figures, dates, commitments and quotations against the source. Confirm that the message does not reveal confidential or personal information through an unapproved tool. For sensitive conversations, use AI only within your organisation’s rules and consider keeping the task manual.

You remain responsible for the communication. If the message could affect someone’s job, reputation, rights or access to an opportunity, review it with extra care and the appropriate people. Good editing is more than making prose sound natural; it protects the meaning and the people affected.

Before you move on

  • The reader, relationship and desired action are clear.
  • The draft contains verified facts and my actual judgement.
  • The wording follows a concrete voice card rather than vague tone labels.
  • Generic openings, filler and exaggerated claims have been removed.
  • I have read the message aloud and would say it this way.
  • Confidentiality, fairness and accountability have been checked.

Sources and further reading

These official resources informed the guardrails and practical method in this guide.

  1. Digital.gov Plain Language Guide ↗
  2. OpenAI Academy: Prompting ↗
  3. CISA: Stay Safe Online When Using AI ↗
Use the free Workweek Starter →

This preview uses fictional examples where indicated. Read our editorial approach.