The person receiving your message does not experience it as an AI draft. They experience it as a promise from you.

If it says a project is on schedule, they may plan around that. If it suggests you understand their concern, they expect your next action to reflect that understanding. A polished sentence does not reduce your responsibility for what happens after someone reads it.

I believe AI-assisted writing belongs inside a human standard of care. You decide what should be said, what should remain private and which commitments you are ready to make. That judgment cannot be handed off just because the draft reads smoothly.

Start with the decision, not the wording

Before drafting, ask: What does this person need to understand or decide?

A client update might need to explain an unresolved issue. A proposal might need to clarify scope. An apology might need to acknowledge a mistake and identify the next responsible action.

Write that purpose in one plain sentence. Then identify what you know, what remains uncertain and what action you can reasonably ask the reader to take.

Without those boundaries, it is easy to approve a message that sounds professional while leaving the important question unanswered. Your job is to make the communication useful, not simply finished.

Protect the context before you share it

Useful context does not mean unrestricted access to someone else's information.

Before putting material into an AI tool, follow your organization's rules and check that you have permission to use that information there. If the rules or permissions are unclear, pause and ask the appropriate owner. Do not make the decision by convenience.

Remove names, contact details, account information and confidential commercial details unless their use is authorized and necessary. A description such as "a client needs a revised timeline" may be enough to work on structure without sharing the client's identity or contract.

Removing a name alone does not settle the privacy question. Other details may still identify the person or reveal sensitive information. Share only what the task requires, through an approved process.

Verify the facts before polishing the tone

Read the draft once for meaning, then again for claims.

Check dates, prices, names, scope, project status and commitments against an approved source. That might be a signed agreement, a confirmed project update or a decision from the person who owns delivery.

Pay particular attention to words such as "confirmed," "included," "completed" and "will." Each one can carry an expectation. If you cannot verify the statement, remove it, qualify it honestly or get confirmation before sending.

Keep uncertainty visible. "We are checking the schedule" means something different from "Your schedule is confirmed." Editing should make that distinction clearer.

Make the message sound like someone accountable

Here is a hypothetical example, not a client story.

Suppose two planning questions remain open, no rollout date has been approved and the project owner has agreed to a Friday progress update. An AI-assisted draft awaiting review reads:

Great news! Everything is on track for Monday. We've resolved the issues and will send the final plan tomorrow.

The human reviewer should reject it. The enthusiasm hides open questions, and the dates and completion claims have no support in the approved facts.

A responsible revision could read:

Two planning questions are still open, so we have not confirmed a rollout date. I'll send you a progress update Friday, including what has been decided and what still needs attention. Is there a deadline on your side we should account for?

The revision gives the reader usable information without inventing certainty. It also sounds like a person willing to own the next step.

Voice is more than replacing formal words with casual ones. Ask whether the message reflects how you actually listen, explain and follow through. Always Be Caring includes telling the truth when the truth is unfinished.

Use the Five-Question Send Gate

Before any AI-assisted business message goes out, have a named person answer these questions:

  1. Purpose: Does this help the reader understand or decide something specific?
  2. Permission: Was the information used appropriately, and is the recipient entitled to receive it?
  3. Facts: Can we verify every material claim and commitment?
  4. Voice: Does this sound like us, with care appropriate to the relationship?
  5. Ownership: Who approves this message and owns the next action?

If an answer is unclear, hold the message. For sensitive or high-stakes communication, involve the responsible specialist or leader before approval.

Try the gate on one draft today. Mark unsupported statements, remove unnecessary private details and read the revision aloud. Approve it only when you can stand behind it.

For a starting point, explore the free AI Prompt Collection. Keep this review standard beside it. Better drafting still needs your judgment.

Better drafting still needs your judgment.

Get the free AI Prompt Collection