When to let AI draft a reply and when to write it yourself

The short answer: let a model draft replies for routine emails, calendar confirmations, and acknowledgements with set details; write them yourself from start to finish when dealing with negotiations, apologies, difficult news, or commitments requiring personal judgement.

The value of drafts in predictable exchanges

Most emails we receive daily don't demand creativity, but operational clarity. Replying whether you can attend a call, acknowledging receipt of a document, or sharing a routine link are low-cognitive-load, high-volume tasks.

In these cases, relying on an AI-generated draft works well if three conditions are met:

• The context is fully contained in the original message: the reply doesn't rely on phone agreements or information outside the thread.
• The decision is binary or straightforward: you are either accepting or declining a proposal, or choosing between two specific dates.
• There is zero room for misunderstanding: there are no double meanings or ambiguity regarding tone or commitment.

As we discussed when exploring what an AI secretary actually does, software is brilliant at processing structure, but it lacks judgement. Letting it prepare the bones of a technical message or an acknowledgement reduces friction, provided you retain final oversight before hitting send.

When to always write from scratch

There are situations where leaning on machine-generated text doesn't save time—it actively undermines communication. As soon as empathy, diplomacy, or professional risk come into play, drafting every word yourself is the only sensible choice.

It is best to avoid automated drafts for:

• Apologies and complaint resolution: generated messages tend to rely on formulaic phrasing that feels cold or indifferent. Owning a mistake requires authenticity and taking responsibility in your own words.
• Negotiations over fees, deadlines, or contracts: an imprecise nuance could close a door or prematurely commit resources.
• Saying ‘no’ to clients or collaborators: declining a proposal gracefully without damaging the relationship demands a finely calibrated tone based on your shared history.
• Condolences or personal support: outsourcing words of comfort to an algorithm strips the message of genuine human warmth.

The risks of artificial plausibility

Language models write with flawless syntax, creating a false sense of security. The danger of delegating drafting lies in what the tool might assume or fabricate without warning.

An automated draft might promise a delivery date you haven't verified in your diary, accept implicit terms that went unnoticed in the thread, or adopt a tone that is far too stiff or overly casual for your relationship. Moreover, if you use third-party tools to process sensitive correspondence, always review how they handle that data under the GDPR and consult their privacy policy to confirm whether data is processed within or outside the European Union.

The three-second filter

If you decide to use a draft suggested by an AI email triage tool, never send it on autopilot. Run three quick checks first:

• Accuracy: are the details, dates, or links mentioned completely accurate?
• Commitment: does the closing sentence commit you to something you hadn't planned on doing today?
• Tone of voice: does it genuinely sound like how you usually write to this person?

If you find yourself rewriting more than half the draft to match your voice or fix facts, binning it and writing two straightforward lines from scratch will always be faster and safer.