The inbox problem is not a new problem. But it has a new shape for a lot of people right now: the volume of email hasn't decreased, the expectation to respond quickly hasn't let up, and somewhere along the way the inbox became a place that induces a low-grade dread rather than something you can just work through.
AI won't fix the underlying conditions that created email overload. But it is genuinely useful for specific, concrete parts of the inbox problem — and understanding which parts those are can make a meaningful difference in how much time and mental energy you spend on email in a given day. This is not a guide to "inbox zero" philosophy. It is a practical look at where AI tools actually help, based on what I've found actually works.
The Part AI Is Genuinely Good At: Drafting the Replies You Keep Putting Off
If you look at what actually bogs people down in email, it isn't usually the quick responses — the "thanks, confirmed" or "see you then." Those take ten seconds. What creates the pile-up is the emails that require a reply you're not sure how to word: a decline that needs to be diplomatic, a follow-up on something awkward, a response to a complaint, an explanation of a decision that someone isn't going to like.
These are the emails that sit in the inbox for three days because every time you open them, you close them again. And this is exactly the category where ChatGPT or Claude is most useful.
The approach that works: paste the email you received into Claude or ChatGPT, explain briefly what you want the reply to accomplish, and ask for a draft. You're not looking for something you'll send verbatim — you're looking for a starting point that gets the blank page out of your way. Here's a simple version of how to frame it:
"I received this email from a client who's frustrated about a delayed delivery. I need to respond apologetically but without promising something I can't deliver. Can you draft a reply? [paste email]"
What comes back won't be perfect. It'll probably be slightly too formal, or it'll include a phrase you'd never actually say. But it will be a draft — something to edit rather than something to create from nothing. For the emails you've been avoiding, that draft is often enough to break the paralysis.
Claude tends to handle tone well in these situations, particularly when the email requires some delicacy. ChatGPT is solid too. Try both on a few of your recurring email types and see which one produces drafts you need to edit less.
Using AI to Understand What an Email Is Actually Asking
Some emails are long. Some are written by people who bury the actual request under six paragraphs of context. Some arrive in a chain of twenty previous messages and you genuinely can't tell, scanning quickly, what anyone needs you to do right now.
For these, a simple prompt to ChatGPT or Claude is surprisingly effective:
"Here's an email thread. What is the most recent sender actually asking me to do, and is there anything time-sensitive? [paste thread]"
This sounds almost too simple, but it works. The AI reads the thread, identifies the actual request buried in it, and flags anything that looks urgent. You still make the judgment call about what to do — you're just not spending five minutes rereading the same paragraph to figure out what the person wants.
One honest caveat here: if an email thread involves sensitive context that you wouldn't want going through a third-party server, be thoughtful about what you paste. For personal or confidential business communication, some people use Claude on Anthropic's platform (which has clear data handling policies) or check whether their employer has a company-licensed AI tool with appropriate privacy terms. This is worth a moment's thought before pasting anything sensitive.
Where AI Struggles With Email
It is worth being specific about where AI falls short in inbox management, because the limitations are real and ignoring them leads to frustration.
AI doesn't know your relationships. A reply that's appropriate to send to a long-term colleague who you have an easy shorthand with is different from a reply to someone you've never met. AI doesn't know those distinctions unless you tell it — and even when you explain, it can produce something that technically matches your brief but misses the relational texture. Always read the draft with the specific recipient in mind before sending.
AI-generated email is recognizable to people who know you. If you have a distinctive writing style, or if you communicate regularly with someone who knows how you write, a draft that came straight from an AI without much editing will feel off to them. This matters more in some contexts than others. For routine transactional email, it probably doesn't matter. For emails to people you have real relationships with, your voice matters — use AI to draft, then rewrite enough that it actually sounds like you.
AI can confidently get facts wrong. If you ask Claude to draft a reply that includes specific information — a date, a number, a detail from a previous conversation — and you haven't explicitly provided that information in your prompt, it may fill in something plausible but incorrect. AI tools sometimes generate confident-sounding details that aren't real; this is sometimes called "hallucination" in the technical literature, and in email context it means you need to verify any specific claim before sending. Never let an AI draft go out without checking that every factual detail is actually accurate.
A Simple System That Combines All of This
The way I've found this most useful in practice is not as a constant tool for every email, but as a specific intervention for the stuck ones. Here's the rough system:
When you open your inbox, sort mentally into three buckets: emails that need a quick reply (handle directly, no AI needed), emails that need a substantial reply you know how to write (handle directly), and emails you keep opening and closing without responding. That third bucket is where AI earns its keep.
For each email in the third bucket, open ChatGPT or Claude in a browser tab alongside your email. Paste the email. Give a one-sentence brief about what you need the reply to do. Get the draft. Edit it until it sounds like you. Send.
For long threads you can't parse quickly: paste the thread, ask what's actually being asked of you. Read the summary. Decide what to do with that information.
This isn't a productivity system overhaul. It's just two specific uses of a readily available tool applied to the exact part of email that most reliably wastes time. The whole thing takes about the same time it would take to write the reply yourself — sometimes faster — and it removes the activation energy cost of the emails you'd otherwise avoid for days.
What About AI Tools Built Into Email Apps?
A number of email clients now have AI features built directly in. Gmail's "Help me write," Microsoft Copilot in Outlook, and various third-party plugins all promise to help you write and manage email without leaving your inbox.
These tools are genuinely convenient when they work well, because the friction of switching to a separate browser tab is gone. Gmail's AI writing feature, in particular, has become reasonably capable for short replies. Copilot in Outlook is useful if you're in a Microsoft 365 environment and your organization has it enabled.
The tradeoff is that the built-in tools tend to be less flexible than going directly to Claude or ChatGPT, where you can give detailed context and shape the output more precisely. For straightforward replies, the built-in tools are fine. For the complicated ones — the diplomatic no, the tense follow-up, the explanation of a decision — I've found that going to a dedicated AI tool with a more detailed prompt gets better results.
The practical answer: try what's already in your email app first. If it handles your stuck emails well enough, that's all you need. If the results feel generic or you keep having to rewrite them significantly, try Claude or ChatGPT with more explicit instructions. There's no wrong answer here — use whatever gets the email sent.
The Part Only You Can Do
The inbox problem has an AI-assisted component and a human component, and it's worth being clear about which is which.
AI can help you draft a reply. It cannot decide whether to send it. AI can summarize a thread. It cannot decide how much this relationship matters to you and what tone is appropriate given history you carry and the AI doesn't. AI can remove the blank-page problem. It cannot remove the judgment calls — which emails actually need a detailed response, which ones can be answered in a sentence, which ones don't need a reply at all.
That last category is worth mentioning. One of the more useful things you can ask an AI to help you think through is whether a given email actually requires a response. Paste it in, describe your relationship to the sender, and ask: does this actually need a reply from me? Sometimes the honest answer is no, and having that confirmed helps.
The goal isn't to automate your inbox. It's to spend less of your working day on the parts of email that drain energy without requiring your actual judgment — and to get through the stuck ones faster so you can get back to the work that does. Used that way, AI and your inbox can reach a reasonable working arrangement.