The honest cost of email
Research on this varies, but the number that gets cited most often is that knowledge workers spend somewhere between two and three hours a day on email. I think that estimate is low, because it does not count the cognitive overhead: the context-switching, the half-finished replies you drafted in your head during a meeting, the low-grade anxiety of knowing there are forty-seven unread messages waiting for you.
The problem is not that email exists. It is that most people’s relationship with their inbox is completely reactive. Something arrives, you respond. Something else arrives, you respond to that. There is no triage, no prioritization, no system, just an endless queue that refills the moment you clear it.
AI does not fix email. But it can change your relationship with it in a few specific ways that are genuinely worth knowing about. Let me walk through them from the simplest to the most involved.
Level 1: Using AI to draft replies faster
This is the lowest-friction starting point and the one that delivers results the same day you try it. The idea is simple: instead of composing a reply from scratch, you paste the email into Claude or ChatGPT, tell it what you want to say, and let it produce a draft you can edit and send.
The key phrase there is “tell it what you want to say.” Not “write a reply.” The vaguer the instruction, the more generic the output. The more specific, the more useful. Here is the difference in practice:
Vague prompt:
Write a reply to this email.Specific prompt:
Write a professional but warm reply to this email. The sender
is a long-time client who is frustrated that their order was
delayed. I want to acknowledge the frustration genuinely, give
them the honest reason for the delay without making excuses,
and let them know the new expected date is December 18th.
Keep it under 150 words and do not use the word "unfortunately."The second prompt takes thirty extra seconds to write. The reply it produces will take you thirty seconds to review and send rather than five minutes to compose. That trade is worth making every time for emails that carry any weight.
A few types of email where this approach pays off most quickly: client complaints, requests you need to decline gracefully, internal messages where tone matters, and anything you have been putting off because you could not figure out how to start.
Level 2: Using AI to triage what actually needs your attention
Not every email deserves the same amount of your brain. Some require a real decision. Some require a reply. Some are informational and need to be filed. Some are noise. The problem is that most inboxes treat all of them identically: everything sits in the same pile until you get to it.
A practical habit that works well, especially if you manage a high-volume inbox: once or twice a day, copy your unread subject lines into Claude or ChatGPT and ask it to help you sort them. Something like:
Here are my unread emails. Categorize each one as:
(A) Needs my decision or reply today
(B) Needs a reply but not urgent
(C) Informational only — just needs to be read and filed
(D) Likely noise or not relevant to me
[paste subject lines and senders here]You are not asking AI to read your email. You are asking it to help you decide where to focus. The categorization will not be perfect, it does not know your relationships or your context the way you do, but it is usually good enough to get your A pile separated from your C and D pile in about two minutes, which is faster than reading through everything yourself.
Over time, you can refine the categories to match how you actually think about your work. The goal is not a perfect system. It is a system you actually use.
Level 3: Summarizing long threads before you respond
Long email threads are their own category of pain. You have been out of office, or cc’d on something that moved fast, or just let a thread go for a few days and now there are twelve replies and you need to respond intelligently without reading every word of every message.
This is something AI does genuinely well. Paste the full thread, ask for a summary of what has been decided, what is still open, and what, if anything, is waiting on you. For a thread that would take eight minutes to read carefully, the summary takes thirty seconds and gives you everything you need to respond.
One note on this: paste threads into a business-tier AI tool with proper data handling, not a free consumer account, if the thread contains anything sensitive. This applies to everything in this post. Client names, financial information, internal disputes, these do not belong in a tool that may use your input to improve its model. The free tools are fine for generic tasks. Sensitive business content needs a tool with a signed data processing agreement. I covered this in more detail in post one if you want the full picture.
Level 4: Building a shared inbox triage system for your team
If your business has a shared inbox: support@, info@, sales@, billing@, hello@, the volume problem is usually worse than in a personal inbox, and the stakes of a slow or missed reply are higher. This is where AI automation, rather than AI assistance, starts to make sense.
The basic version of a shared inbox triage system works like this:
- An email arrives in the shared inbox.
- An automation reads the email and classifies it: new inquiry, existing customer question, billing issue, vendor outreach, spam, based on rules you define.
- Each category routes to the right person or team, with a draft reply generated for the common ones.
- A human reviews the draft, adjusts if needed, and sends. Nothing goes out automatically.
The tools that make this work together are usually something like Zapier or Make for the automation layer, plus whichever AI assistant fits your stack for the classification and drafting step. Setup takes a few hours. The time it returns, for a business handling even fifty to a hundred shared inbox emails a week, is significant and compounds every week thereafter.
I want to be clear about one thing: the draft-and-review step is not optional. The fastest way to damage a customer relationship is to send an AI-generated reply that misread the tone of the original message. AI is fast and usually directionally correct. It is not infallible. A human checkpoint on every outgoing message is worth the thirty seconds it takes.
Level 5: Template and signature intelligence
This one is smaller but worth mentioning because it is often overlooked. If you send a lot of similar emails: proposals, follow-ups, onboarding messages, check-ins, meeting confirmations, you probably have some version of templates already. Maybe saved drafts. Maybe a folder of old emails you copy from. Maybe just the same phrases you type from memory every time.
AI is very good at helping you build a proper template library and then personalizing from it. You give it the template and the details of the specific situation: the person’s name, what you know about them, what you want to emphasize in this particular send, and it produces a version that reads like it was written for this person, not mail-merged at them.
The time this saves is not huge per email. Across a week of outreach, it adds up. Across a year, it is a meaningful reduction in the low-level mental effort that drains you without you noticing.
What AI cannot do for your inbox
It cannot fix the underlying problem if the underlying problem is that too many people have your email address and too many of them use it for things that should be a Slack message, a ticket, or a phone call. AI makes handling email faster. It does not make email the right tool for everything it gets used for.
It also cannot make judgment calls about relationships. Whether to push back on a client, how much to apologize, whether a situation warrants a phone call instead of a reply, those decisions belong to you. AI can draft the message once you have made the call. It cannot make the call for you, and you should not want it to.
The goal is not an inbox managed by AI. The goal is an inbox that takes less of your time and attention so the important stuff: the calls, the decisions, the relationships, gets more of both.
Where to start this week
Pick one type of email you write repeatedly: the same kind of reply, week after week, and build a reusable prompt for it. Something that captures the tone you want, the information you always need to include, and the things you never want to say. Save it somewhere you can find it. Use it for the next ten emails of that type and notice what changes.
That single habit, a saved prompt for your most common email type, is the entry point for everything else in this post. Everything more sophisticated builds from it.
This is post four of a two-year series on AI for real people doing real work. Post one covers what AI actually is. Post two is a look at how I use these tools day to day. Post three covers the five tools worth trying before you spend anything. January brings the prompt engineering post I promised last month: how to ask AI better questions and get more consistent results. Questions or something specific on your inbox you want me to address? Send a note.
