The email that changed my mind about Claude was not a marketing blast. It was one line to a client who owed me money for six weeks, and I had rewritten the subject four times without touching the body. I described the situation, asked for a draft, and had something usable in under a minute. The only thing wrong with it was that it sounded like nobody I knew, least of all me. Fixing that took thirty seconds. Sending it took another ten.
Email is the place most people first feel AI save them real time, and it is also the place where the results go wrong fastest. A generic AI email is easy to spot from across the room: too smooth, too many transition phrases, warmth that nobody in the actual relationship uses. The fix is not a cleverer prompt. It is making Claude sound like you, and pointing it at the messages that were actually costing you time, not the ones that took twenty seconds anyway.
This is the next post in a beginner's series on Claude. The first post mapped out what Claude can do across chat, code, and the API. This one is about the place most people meet it first, the inbox.
The Draft Is the Easy Part, the Voice Is the Point
Getting Claude to produce an email draft is not the interesting part anymore, every tool does that now. What is worth thirty seconds is telling it how you actually write, because that is the difference between a message your recipient reads normally and one that makes them wonder if a person wrote it at all.
The fastest way in is evidence. Paste two or three emails you have actually sent, ordinary ones, and say something like: match this voice, short sentences, no exclamation marks, sign off with just my first name. No examples handy? Describe yourself instead. I write short sentences. I never open with "I hope this email finds you well." I say what I want in the first two lines and stop. One instruction like that changes more than any clever structure you could build around it.
Skip this step and you get the email everyone can now recognize on sight, padded with "I wanted to reach out" and "please don't hesitate," three ideas stretched into five paragraphs. That is not a flaw in Claude. That is what happens when you ask for an email and give it nothing to sound like.
Use It for the Email You Are Avoiding
The email actually worth using Claude for is not the one you would have written in two minutes anyway. It is the one sitting in your drafts folder because you keep opening it and closing it again. The no you owe a client who will not like hearing it. The apology for something you actually got wrong. The third follow-up to someone who still has not paid. The message that draws a boundary with someone you like.
For these, do not ask for "an email." Describe the situation the way you would to a friend before asking their advice: what happened, what you want the other person to walk away believing, and what you are not willing to do. Let Claude write the whole first pass, even if it lands too blunt or too soft. Reading a draft, even a wrong one, is easier than staring at an empty subject line. Then push it toward exactly how direct or how gentle you actually want to be, which is a much smaller job than writing it from nothing while your stomach is in a knot.
Reply, Do Not Just Compose
Most email is not a blank page, it is a reply, and treating it that way saves the most time of anything in this post. Paste the incoming thread, the whole back-and-forth if there is one, and ask for a reply that hits the three points that actually matter: confirm the date, push back on the price, ask for the file they forgot to attach. Claude reads the thread the way you would, picks up who said what and when, and writes something that fits the conversation instead of a generic reply bolted onto the top of it.
It also catches what a tired read-through misses, like the actual question buried in paragraph three that everyone has skipped past twice already.
Shorten, Soften, Sharpen
The mistake people make once they know the tricks above is trying to write the perfect prompt so the first draft comes out finished. That is not how this works, and chasing it wastes more time than it saves. The real workflow is one reasonable draft followed by a few small notes, the same way you would mark up a colleague's email before it goes out.
Shorter. Cut the second paragraph, they do not need the reasoning, just the decision. Too formal, we know each other, loosen it up. This line reads angry and I am not, soften it. Sharper here, they need to understand this is the last time I offer this rate. Each note takes five seconds, and Claude keeps everything else from the draft, so you are never starting over, only narrowing in. Three rounds of that gets you to something you would actually send faster than one long, careful prompt ever will.
Triage a Full Inbox
Once you trust it with single emails, the bigger win is the pile. Paste or forward a batch of unread messages, a day you ignored or a week you were traveling, and ask a plain question: what needs an answer today, what can wait until Friday, and is anything in here actually urgent even though the subject line does not say so.
Claude sorts faster than you can while dreading the unread count, and it reads every message with the same attention, so the request buried under a newsletter and two calendar invites does not slip past the way it would on your third scroll. You still write the actual replies, or draft them the way the sections above describe, but the sorting alone turns a wall of unread into a short list you can work through in order.
Keep Your Voice in a Project
Everything above works better once you stop repeating yourself. A Project on claude.ai is a workspace that holds instructions and reference files, and every conversation you start inside it already carries that context. Paid plans let you create as many as you want, so a Project called Email is worth setting up once and never touching again.
Put your voice notes in it, the same ones from the first section. Add your real signature so it never has to guess your title or your phone number. Drop in a short paragraph about your business, your role, and the handful of people you write to most, your main contacts, your team, anything Claude would otherwise need reminding of every single time. After that, every email you draft inside that Project starts from your voice and your context automatically, and the thirty seconds you used to spend re-explaining who you are is just gone.
Where It Should Not Go
None of this means you stop reading before you hit send. Claude does not know your actual availability next Tuesday, your real price if it changed last month, or whether you already promised someone a discount on a call it was never part of. Left unchecked, it fills those gaps with something plausible, and plausible is not the same as true.
Treat every number, date, and commitment in a draft as unverified until you have checked it yourself, the way you would treat a junior colleague's first pass at a contract. The words can be Claude's. The facts, the promises, and the send button stay yours. That is not a limitation of the tool. It is just what it means to put your name on something.
Where to Start This Week
Open the email you have been avoiding, the one you have rewritten in your head more times than you would admit. Describe the situation to Claude in three sentences, let it draft the whole thing, and edit it out loud the way the fourth section describes: shorter, softer, sharper, whatever it actually needs. Send that one first, before you touch anything else, because that is the email where the time savings are real and not just convenient.
If you want the fuller path through all of this, our free StudioMeyer Academy covers it lesson by lesson. Next in this series, we look at what happens when you hand Claude something harder than writing: real research, and how to get a report back that you can actually trust.
