Every major inbox now has some version of it — a little sparkle icon that promises to write your email for you. Gmail calls it Help Me Write. Outlook has its Copilot draft. Superhuman, Spark, even some CRMs bolted one on last year. Tap it, type a rough intent, and out comes a full paragraph ready to send.
I've been using these buttons for months now, across different apps, for everything from scheduling requests to actual customer complaints. And the pattern that's emerged surprised me: the feature is genuinely excellent for a narrow band of messages, and actively worse than typing it yourself for almost everything else. Nobody selling you the feature tells you which is which. So let's sort that out.
Photo by cottonbro studio on Pexels
The messages where the AI draft earns its spot
Think about the emails you write on autopilot — the ones where the content barely varies and the stakes are near zero. That's the sweet spot.
- Confirming a meeting time someone already proposed
- Asking for a document or file you know exists
- A polite "following up on this" nudge
- Declining a sales pitch or cold outreach
These messages have almost no information content. You're not persuading anyone or navigating a delicate situation — you're just performing the social ritual of writing a full sentence instead of replying "yes" in all caps. An AI draft handles the ritual part fine, because the ritual part is exactly what it's trained on.
Say a recruiter emails asking if Tuesday at 2pm works. You genuinely don't care, Tuesday works, and normally you'd spend ninety seconds crafting three sentences that say "yes" with some padding around it. That's the exact task the draft button was built for. You tap, skim it for accuracy, hit send. Ninety seconds becomes fifteen.
Where it quietly starts costing you time instead of saving it
The moment a message has actual stakes, the math flips. Not because the AI writes badly — it writes fluently, which is part of the problem. Fluent, generic-sounding text is exactly what you don't want when the content matters.
Take a message where you're pushing back on an invoice, asking a manager for a schedule change, or writing a condolence note. These all require something the model doesn't have: the specific relationship history between you and the recipient, and the specific thing you actually want to happen. The draft it hands you will be polished and will also miss the one detail that makes the email work — the fact that you already agreed to a discount last time, or that your manager already knows why you need Friday off and doesn't need the backstory repeated.
What happens in practice is you end up editing more than you'd have written from scratch. You're not drafting anymore, you're proofreading someone else's guess at your intent, and untangling generic phrasing takes longer than generating your own from the specific facts you already have in your head. I've caught myself doing this more than once — spending four minutes "fixing" a draft that would've taken two minutes to just write.
There's a sneakier cost too: tone drift. AI drafts trend toward a bland, over-polite register that reads fine in isolation but feels off if you've been emailing someone casually for two years. Recipients notice this before they can articulate why. A message that suddenly sounds like corporate boilerplate is a small trust hit, even if nobody says anything about it.
A rough way to decide, before you tap the button
Here's the filter that's worked for me: ask whether the email requires information only you have. If the answer is no — if a stranger with your calendar could write the same message — the AI draft is a shortcut worth taking. If the answer is yes, you're better off writing the first sentence yourself and, if you want help, asking the AI to expand or polish what you already drafted rather than generate it from a one-line prompt.
A few other signals that push a message into "write it yourself" territory:
- It involves money, a complaint, or correcting someone
- The relationship has history the model can't see
- Getting the tone slightly wrong would actually matter
- You'd be embarrassed if the recipient guessed it was AI-written
That last one is worth sitting with. Some of these drafts have a texture to them — the sentence rhythm and hedging language settle into a shape you start recognizing after enough exposure, the same way you can spot a form letter. For a low-stakes scheduling reply, nobody cares. For an apology to a client, it can read as exactly the kind of thing you were supposed to write yourself.
The version of this that actually works well
Photo by Andrea Piacquadio on Pexels
The best use I've found isn't "write this email for me" — it's "clean up what I already wrote." Type your reply badly, in fragments, the way you'd talk to a friend, and then ask the tool to tighten grammar and flow without changing the content. You keep the specific facts and the actual intent, and the model handles the part it's genuinely good at: smoothing rough phrasing into readable sentences.
This flips the failure mode. Instead of the AI guessing at your intent and you correcting it, you supply the intent and the AI just handles the mechanical polish. It's a smaller ask, and small asks are where these tools are reliable.
FAQ
Does using an AI-drafted email look unprofessional if someone finds out?
Not inherently — plenty of professional writing gets touched up by tools now. The risk isn't the tool, it's sending something generic when the situation called for specificity. Read every draft before sending, and if it feels like it could've been sent to anyone, it probably needs a rewrite, not a polish.
Which is better for this — the built-in inbox feature or a separate chatbot like ChatGPT or Claude?
The built-in features are faster for quick replies since they already see the thread you're replying to. A separate chatbot is better for anything longer or more sensitive, because you can give it more context and iterate on tone before you ever touch your actual inbox.
Is it worth paying extra for an email AI feature if my free plan doesn't include it?
Only if you send a genuinely high volume of the low-stakes, templated kind of email described above — customer support, recruiting, scheduling. If most of your email is one-off and personal, you won't get enough use out of it to justify the cost.
The short version
The AI draft button isn't a universal writing assistant, and treating it like one is where people get burned. It's a very good tool for exactly one job — turning a routine, low-content message into a polished one fast — and a mediocre tool for anything that needs judgment, history, or nuance. Use it for the emails you'd write the same way every time anyway. For the ones that actually matter, write the substance yourself and let the AI handle the polish, not the thinking. That's the boring answer, but it's the one that's held up for me across months of actual use.
Comments
Post a Comment