Delv
Guide
29 March 20269 min read

How to Use AI for Email (And Save Yourself Hours Every Week)

Stop copy-pasting emails into ChatGPT one at a time. Here are the actual workflows and prompt templates that save real time.

DV

Delv Editorial

Delv Team

The wrong way to use AI for email

Most people use AI for email like this: copy an email, paste it into ChatGPT, type "write a reply," copy the response, paste it back into their email client, then spend five minutes editing out all the "I hope this message finds you well" nonsense.

That's not saving time. That's adding steps.

The right way to use AI for email is to build specific workflows for specific types of emails, with prompts that produce output you barely need to edit. I've been refining these for about a year, and the ones below are the ones that actually stuck.

Workflow 1: The quick reply

This is the one I use most. Someone sends you an email that needs a response but doesn't need a lot of thought. Status updates, meeting confirmations, simple questions.

The prompt template: 
Reply to this email. Keep it under 3 sentences. Friendly but professional tone. British English. Don't use "I hope this finds you well" or any filler phrases. Just answer directly.

[paste email]

That "don't use filler phrases" instruction is crucial. Without it, every AI-generated email starts with some variation of "Thank you for reaching out" or "I appreciate you sharing this." Nobody talks like that. Just answer the question.

ChatGPT handles these well. Claude also does a good job but tends to be slightly more verbose. For quick replies, speed matters more than quality, so I usually stick with whichever one responds faster.

Workflow 2: Summarising long email threads

This one genuinely saves hours. You know those email threads with 47 replies where someone changed the subject halfway through and three different sub-conversations are happening simultaneously? Yeah.

The prompt template: 
Summarise this email thread. I need:

  1. The key decisions that were made
    2. 

  2. Any action items (and who's responsible)
    3. 

  3. Any unresolved questions
    4. 

  4. The overall current status
    5. 


Be concise. Bullet points preferred.

[paste full thread]

Both ChatGPT and Claude are excellent at this. Claude is slightly better at handling very long threads because of its larger context window. If the thread is over about 20 emails, Claude is the better choice.

The trick is pasting the entire thread, not just the latest reply. AI can't summarise what it can't see, and the most important context is often buried 15 replies deep where someone casually mentioned a deadline that everyone subsequently forgot about.

Workflow 3: Cold outreach that doesn't sound desperate

Cold emails are where AI helps and hurts in equal measure. The help: it can produce grammatically correct, well-structured outreach. The hurt: if you don't guide it carefully, every cold email sounds like it was written by the same LinkedIn recruiter bot.

The prompt template: 
Write a cold outreach email. Details:

  • Who I am: [your role and company]
    • 

  • Who I'm emailing: [their role and company]
    • 

  • What I want: [specific ask]
    • 

  • Why they should care: [genuine value for them]
    • 

  • Mutual connection or context: [if any]
    • 


Rules:

  • Under 150 words total
    • 

  • No buzzwords or marketing speak
    • 

  • Don't say "I'd love to" or "I was wondering if"
    • 

  • Sound like a real person, not a sales bot
    • 

  • One clear call to action at the end
    • 

  • British English
The word limit is the most important part. AI left to its own devices will write a 400-word cold email. Nobody reads a 400-word cold email. The best cold emails are short, specific, and honest about what you want. Critical warning: Never send AI-generated cold emails without reading them first. AI occasionally hallucinates details, and sending someone an email that mentions something that isn't true about their company is worse than not emailing them at all.

Workflow 4: Handling angry customer emails

This is where AI genuinely shines. When you're reading an angry email, your instinct is to be defensive. AI doesn't have that instinct. It writes calm, empathetic responses that acknowledge the problem without getting drawn into arguments.

The prompt template: 
I received this angry customer email. Write a response that:

  1. Acknowledges their frustration without being patronising
    2. 

  2. Takes responsibility where appropriate (don't admit fault if we're not at fault, but empathise)
    3. 

  3. Offers a specific next step
    4. 

  4. Keeps a warm, human tone
    5. 

  5. British English, under 200 words
    6. 


The customer's issue is: [brief context]

Their email:
[paste email]

Both ChatGPT and Claude handle this well, but I'd give the edge to Claude here. Its responses tend to feel more genuinely empathetic and less formulaic. ChatGPT sometimes falls into customer service speak ("We value your feedback and take all concerns seriously...") which sounds corporate and insincere.

Important: Always add the context line. Without it, the AI is guessing at the situation and might write something that misrepresents your position. With context, it can calibrate the response correctly.

Workflow 5: Weekly update emails

If you send regular update emails to clients, stakeholders, or your team, AI can cut the drafting time from 30 minutes to about 5.

The prompt template: 
Write a weekly update email based on these bullet points:
[paste your rough notes/bullet points]

Format:
  • Brief friendly opening (one sentence, no filler)

  • "Completed this week" section

  • "In progress" section

  • "Next week" section

  • Any blockers or things that need input

  • Brief closing


Tone: professional but not stiff. British English.

This workflow is about turning messy notes into a structured email. You still need to provide the substance. The AI just handles the formatting and polishing. That's the right division of labour.

Workflow 6: The diplomatic "no"

Saying no to things via email is an art form. Say it too bluntly and you burn bridges. Say it too softly and people don't realise you've said no.

The prompt template: 
Write a polite but clear email declining this request:
[what they asked for]

Reason for declining: [honest reason, or "I don't want to give a specific reason"]
Relationship: [colleague/client/stranger/boss]
Offer alternative: [yes/no, and what if yes]


Rules: Be genuinely kind but make sure the "no" is unmistakable. British English. Under 100 words.

The "make sure the no is unmistakable" instruction is essential. AI's natural tendency is to soften the refusal until it sounds like a maybe. You want the recipient to read it once and know where they stand, not spend ten minutes trying to decode whether you're saying no or asking for more time.

The tools

ChatGPT is my default for most email tasks. It's fast, the output is consistently decent, and if you're using the Chrome extension, you can access it without leaving your inbox. Claude is better for longer, more nuanced emails. Anything that requires emotional intelligence, like the angry customer workflow or the diplomatic no, Claude handles with more subtlety. Grammarly sits underneath everything. It catches the small errors that AI introduces (yes, AI makes grammar mistakes) and flags anything that sounds too robotic. The tone detection feature is useful for checking that your "diplomatic no" actually reads as diplomatic and not passive-aggressive.

What not to do

Don't automate sending. AI should draft emails. You should read and send them. The moment you set up a system that sends AI-generated emails without human review, you're one hallucination away from a PR disaster. Don't use AI for sensitive topics. Redundancies, legal matters, personal issues. These need a human touch that AI can approximate but never quite nail. The risk of getting the tone wrong is too high. Don't tell people it's AI-generated. Not because it's dishonest, but because it frames the conversation oddly. "I used AI to write this email" sounds like "I couldn't be bothered to write to you myself." Just use AI as a drafting tool and own the final result. Don't skip the editing step. Every AI-generated email needs at least one read-through. Check for accuracy, check the tone matches your relationship with the recipient, and for the love of everything, check that it hasn't invented any facts.

The real time savings

I tracked my email time for a month before and after adopting these workflows. Before: about 2.5 hours per day on email. After: about 1.5 hours. That's a genuine hour saved every day, which over a month is roughly 20 hours.

Most of the savings come from the thread summaries (I was spending ages re-reading long threads) and the quick replies (those three-sentence responses that used to take five minutes of overthinking now take 30 seconds).

Is it perfect? No. I still edit most AI drafts before sending. But editing a decent first draft is massively faster than staring at a blank compose window trying to figure out how to phrase something diplomatically.

DV

Delv Editorial

Delv Team

The Delv editorial team reviews AI tools, MCP servers, Agent Skills, and autonomous agents. Reviews are drafted with AI assistance and human oversight. Every install command and config snippet is verified against the source. We're independent, we don't sell tools, and we say when something isn't worth it.

AI ToolsMCPSkillsAgents

How to Use AI for Email (And Save Yourself Hours Every Week)

Stop copy-pasting emails into ChatGPT one at a time. Here are the actual workflows and prompt templates that save real time.

By Delv Editorial9 min read

The wrong way to use AI for email

Most people use AI for email like this: copy an email, paste it into ChatGPT, type "write a reply," copy the response, paste it back into their email client, then spend five minutes editing out all the "I hope this message finds you well" nonsense.

That's not saving time. That's adding steps.

The right way to use AI for email is to build specific workflows for specific types of emails, with prompts that produce output you barely need to edit. I've been refining these for about a year, and the ones below are the ones that actually stuck.

Workflow 1: The quick reply

This is the one I use most. Someone sends you an email that needs a response but doesn't need a lot of thought. Status updates, meeting confirmations, simple questions.

The prompt template:

That "don't use filler phrases" instruction is crucial. Without it, every AI-generated email starts with some variation of "Thank you for reaching out" or "I appreciate you sharing this." Nobody talks like that. Just answer the question.

ChatGPT handles these well. Claude also does a good job but tends to be slightly more verbose. For quick replies, speed matters more than quality, so I usually stick with whichever one responds faster.

Workflow 2: Summarising long email threads

This one genuinely saves hours. You know those email threads with 47 replies where someone changed the subject halfway through and three different sub-conversations are happening simultaneously? Yeah.

The prompt template:

Both ChatGPT and Claude are excellent at this. Claude is slightly better at handling very long threads because of its larger context window. If the thread is over about 20 emails, Claude is the better choice.

The trick is pasting the entire thread, not just the latest reply. AI can't summarise what it can't see, and the most important context is often buried 15 replies deep where someone casually mentioned a deadline that everyone subsequently forgot about.

Workflow 3: Cold outreach that doesn't sound desperate

Cold emails are where AI helps and hurts in equal measure. The help: it can produce grammatically correct, well-structured outreach. The hurt: if you don't guide it carefully, every cold email sounds like it was written by the same LinkedIn recruiter bot.

The prompt template:

The word limit is the most important part. AI left to its own devices will write a 400-word cold email. Nobody reads a 400-word cold email. The best cold emails are short, specific, and honest about what you want.

Critical warning: Never send AI-generated cold emails without reading them first. AI occasionally hallucinates details, and sending someone an email that mentions something that isn't true about their company is worse than not emailing them at all.

Workflow 4: Handling angry customer emails

This is where AI genuinely shines. When you're reading an angry email, your instinct is to be defensive. AI doesn't have that instinct. It writes calm, empathetic responses that acknowledge the problem without getting drawn into arguments.

The prompt template:

Both ChatGPT and Claude handle this well, but I'd give the edge to Claude here. Its responses tend to feel more genuinely empathetic and less formulaic. ChatGPT sometimes falls into customer service speak ("We value your feedback and take all concerns seriously...") which sounds corporate and insincere.

Important: Always add the context line. Without it, the AI is guessing at the situation and might write something that misrepresents your position. With context, it can calibrate the response correctly.

Workflow 5: Weekly update emails

If you send regular update emails to clients, stakeholders, or your team, AI can cut the drafting time from 30 minutes to about 5.

The prompt template:

This workflow is about turning messy notes into a structured email. You still need to provide the substance. The AI just handles the formatting and polishing. That's the right division of labour.

Workflow 6: The diplomatic "no"

Saying no to things via email is an art form. Say it too bluntly and you burn bridges. Say it too softly and people don't realise you've said no.

The prompt template:

The "make sure the no is unmistakable" instruction is essential. AI's natural tendency is to soften the refusal until it sounds like a maybe. You want the recipient to read it once and know where they stand, not spend ten minutes trying to decode whether you're saying no or asking for more time.

The tools

ChatGPT is my default for most email tasks. It's fast, the output is consistently decent, and if you're using the Chrome extension, you can access it without leaving your inbox.

Claude is better for longer, more nuanced emails. Anything that requires emotional intelligence, like the angry customer workflow or the diplomatic no, Claude handles with more subtlety.

Grammarly sits underneath everything. It catches the small errors that AI introduces (yes, AI makes grammar mistakes) and flags anything that sounds too robotic. The tone detection feature is useful for checking that your "diplomatic no" actually reads as diplomatic and not passive-aggressive.

What not to do

Don't automate sending. AI should draft emails. You should read and send them. The moment you set up a system that sends AI-generated emails without human review, you're one hallucination away from a PR disaster.

Don't use AI for sensitive topics. Redundancies, legal matters, personal issues. These need a human touch that AI can approximate but never quite nail. The risk of getting the tone wrong is too high.

Don't tell people it's AI-generated. Not because it's dishonest, but because it frames the conversation oddly. "I used AI to write this email" sounds like "I couldn't be bothered to write to you myself." Just use AI as a drafting tool and own the final result.

Don't skip the editing step. Every AI-generated email needs at least one read-through. Check for accuracy, check the tone matches your relationship with the recipient, and for the love of everything, check that it hasn't invented any facts.

The real time savings

I tracked my email time for a month before and after adopting these workflows. Before: about 2.5 hours per day on email. After: about 1.5 hours. That's a genuine hour saved every day, which over a month is roughly 20 hours.

Most of the savings come from the thread summaries (I was spending ages re-reading long threads) and the quick replies (those three-sentence responses that used to take five minutes of overthinking now take 30 seconds).

Is it perfect? No. I still edit most AI drafts before sending. But editing a decent first draft is massively faster than staring at a blank compose window trying to figure out how to phrase something diplomatically.

Delv Editorial - Delv Team

The Delv editorial team reviews AI tools, MCP servers, Agent Skills, and autonomous agents. Reviews are drafted with AI assistance and human oversight. Every install command and config snippet is verified against the source. We're independent, we don't sell tools, and we say when something isn't worth it.

Related Articles

It felt sudden. It wasn't. A short history of how the iceberg surfaced.

8 min read

Karpathy's actual CLAUDE.md is boring. The viral one is something else entirely.

5 min read

I installed Osaurus on my Mac this week. Here's what it actually changes.

5 min read