Claude Prompts for Email: The 15 That Work, and How to Stop Needing Them

Prompts last reviewed: August 2026.

The best Claude prompts for email do one thing the generic ones never do: they tell Claude how the email should behave, not just what it is about. Where the no goes, what a follow-up must add before it earns the send, what a summary owes the person who reads it. Get that right and Claude turns a five-minute reply into a thirty-second review. Below are the 15 prompts we actually use for everyday work email, organized by situation, each group built around the mistake it exists to prevent, with a worked example showing the difference on a real negotiation. And because a prompt list should be honest about where this is all heading, the last section covers the step that ends copy-paste altogether: connecting Claude to the inbox itself.

One scope note before the prompts: this page is about everyday work email. If you want a broader template library including sales and marketing situations, our 50+ email prompts guide covers that ground, along with the full role-context-goal-constraints pattern these prompts are built on. The Claude-specific version of that pattern takes thirty seconds: paste the entire thread rather than summarizing it, because Claude handles long context well and the details are what make drafts sound informed; and describe your voice once at the start of a conversation, because Claude holds a described voice unusually consistently. Everything below assumes those two habits.

The 15 Claude email prompts, by situation

Four situations, fifteen prompts, numbered so you can find your way back to the one you need. The full list:

1. The reply that answers everything, in order
2. The yes that locks the deadline
3. The no that comes first
4. The de-escalation
5. The bad news, delivered first
6. The follow-up that earns its send
7. The second nudge, with a way out
8. The meeting recap people actually read
9. The overdue invoice chase
10. The joining-cold briefing
11. The three-line manager version
12. The real-ask detector
13. The voice clone, built from evidence
14. The reader-first announcement
15. The ruthless editor

Each situation below opens with the way emails of that kind usually go wrong, because that is exactly what these prompts are built against.

Replies, including the hard ones (prompts 1-5)

Replies fail in a predictable way: they answer the easy half of the email and bury the hard half. The pleasant questions get three sentences each, the no gets wrapped in cotton wool at the bottom, and the sender writes back to ask the thing you avoided. These five are built against that.

1. The reply that answers everything, in order

Claude

Here is an email thread [paste]. Draft a reply that answers every question asked, in the order asked, under 150 words, warm but direct. If the thread asks anything I cannot answer from the context you have, put a placeholder in [brackets] instead of guessing.

Copy

Why the brackets matter: without that instruction, the model's instinct is to be helpful and invent the missing answer, and an invented delivery date is worse than no draft at all. The brackets turn every unknown into a visible to-do instead of a hidden commitment.

2. The yes that locks the deadline

Claude

Reply to this email [paste] agreeing to the request. Confirm the deadline back to them explicitly, and ask the one clarifying question that most reduces the chance of doing the work twice.

Copy

Why the deadline echo: agreement without a restated date is where deadline drift begins. When the date appears in your own words, both sides have committed to the same one, in writing. And it asks for one clarifying question, not several, because one question gets answered and five get postponed.

3. The no that comes first

Claude

Here is the situation: [describe]. Draft an email that says no within the first two sentences, gives one honest reason without over-explaining, and offers a specific alternative. Warm, firm, under 120 words.

Copy

Why the no goes first: a decline the reader has to hunt for reads as evasive, and evasive reads as negotiable. Two sentences in, it is just an answer.

4. The de-escalation

Claude

This thread [paste] has gotten tense. Draft a reply that acknowledges their frustration in one sentence, restates the facts neutrally without conceding the substance, and proposes a short call. Under 100 words.

Copy

Why one sentence of acknowledgment: sympathy repeated becomes a debate about feelings. A single sentence closes that lane, the neutral restatement resets the facts, and the call offer moves the rest off a medium where tone misreads.

5. The bad news, delivered first

Claude

I need to deliver bad news: [describe]. Write an email that leads with the news, takes ownership in one sentence without groveling, and ends with the concrete next step we are taking. No corporate hedging.

Copy

Why the news goes first: everyone reads the first two lines and skims the rest, so bad news buried in paragraph three gets discovered later, and angrier. Leading with it costs one uncomfortable sentence and buys the credibility of the paragraph that follows.

Chasing and following up (prompts 6-9)

Follow-ups die of two diseases: the empty bump ("just floating this to the top of your inbox") and guilt. Nobody replies faster because you made them feel bad, and nobody replies at all to a message that contains nothing new. Every prompt here forces the follow-up to earn its send.

6. The follow-up that earns its send

Claude

Here is an email I sent [paste] that got no reply after [number] days. Write a follow-up that adds one new piece of information or value rather than just bumping the thread, and end with a question that can be answered from a phone in one line.

Copy

Why the phone test: most unanswered email is unanswered because replying feels like work. A question someone can answer with "yes, Tuesday" while walking between meetings gets answered.

7. The second nudge, with a way out

Claude

Write a second follow-up to this thread [paste]. Shorter than the first, zero guilt, one line of context, one clear ask, and explicit permission to decline: if the answer is not now, a quick "not now" genuinely helps me plan.

Copy

Why the permission line: it converts silence into information. People who feel trapped go quiet; people who are offered a graceful "not now" usually take it, and now you know where you stand.

8. The meeting recap people actually read

Claude

Draft a post-meeting follow-up from these notes [paste]: one line of thanks, the decisions made, who owns what by when, and the one question still open. Bullets for the middle, sentences at the ends.

Copy

Why owners and dates: a recap without names attached to deadlines is minutes, and nobody reads minutes. A recap of decisions with owners and dates is a task list wearing an email, and that is the version people act on.

9. The overdue invoice chase

Claude

Draft an email chasing the overdue invoice [number], now [days] days late. Friendly on the surface, unambiguous underneath: the amount, the original due date, a placeholder for the payment link, and the date after which it escalates. Under 100 words.

Copy

Why the escalation date: a chase without a consequence is a suggestion. Naming the date it escalates turns it back into a deadline, while the friendly surface keeps the relationship intact for every invoice after this one.

Summarizing and triage (prompts 10-12)

Ask Claude to "summarize this thread" and you get a book report: faithful, complete, and useless, because it treats every fact as equal. What you actually need is a decision brief, and a decision brief has a shape.

10. The joining-cold briefing

Claude

Summarize this thread [paste] for someone joining it cold: what is being decided, who wants what, where it stands, and what the next action is. Five bullets maximum, then one sentence telling them the first thing to do.

Copy

Why the final sentence: a summary that ends without an action leaves the new owner informed and stationary. The one-sentence "do this first" is the actual handover; the bullets are just its context.

11. The three-line manager version

Claude

Turn this long email [paste] into three lines for my manager: the situation, the risk, and the decision I need from them. Nothing else.

Copy

Why three lines: managers decide from the shape of the risk, not from the history. Situation, risk, decision-needed is that shape, and anything longer gets filed under later.

12. The real-ask detector

Claude

Read this thread [paste] and tell me what the sender actually wants, separate from what they wrote. Then draft the reply that addresses the real ask.

Copy

Why the stated-versus-real split: a surprising share of difficult email is difficult because the request on the surface ("can you resend the report?") is not the actual issue ("I don't trust the numbers"). Claude is good at surfacing that gap when you ask for it directly, and the reply that addresses the real ask ends threads that would otherwise run for a week.

Voice, and writing from scratch (prompts 13-15)

The standing problem with AI drafting is generic voice: drafts that are perfectly professional and sound like nobody. The fix is not adjectives ("write warmly!") but evidence.

13. The voice clone, built from evidence

Claude

Here are two emails I wrote that sound like me [paste both]. Study the sentence length, the greeting and sign-off, and the level of formality, then write an email to [name] about [topic] in that voice.

Copy

Why samples beat descriptions: you cannot accurately describe your own writing voice, but you can always find two emails that carry it. Claude reads the evidence and holds the register for the rest of the conversation.

14. The reader-first announcement

Claude

Write an email to [audience] announcing [thing]. Lead with what changes for the reader, not with how excited we are. One paragraph of context, bullets only if there are three or more concrete facts, one clear call to action.

Copy

Why reader-first: announcements written from the sender's excitement get archived on the subject line. The only question a reader brings is what is different for me now, and the prompt forces that answer into the opening line.

15. The ruthless editor

Claude

Rewrite this draft [paste] at half the length without losing a single commitment, date, or number it contains. Then tell me what you cut and why.

Copy

Why the cut report: halving a draft is easy, halving it without dropping a date or a promise is the skill. Asking Claude to show its cuts lets you catch the one thing it should not have removed, in seconds instead of a re-read.

The difference a good prompt makes: a real example

To show what the structure above actually buys you, here is the same email drafted twice, once from a lazy prompt and once from a proper one. The email is one most businesses eventually face: a good customer asking to renew at last year's price.

The vague prompt, "reply to this renewal email politely," produces what you would expect: a draft that thanks them twice, says pricing "reflects our continued investment in the product," commits to nothing, and quietly hopes the problem goes away. You would rewrite all of it.

The four-part version: role (account manager who values this customer), the full pasted thread (which mentions, three emails back, that they doubled their seats in March), goal (hold the new price, offer annual billing as the one concession, keep the relationship warm, invite a call), constraints (under 130 words, warm but not apologetic, no exclamation marks). First pass:

Hi Priya, thank you for being straight with us about the renewal, and for three years as a customer we genuinely value. I can't hold last year's pricing: the new rate applies to all accounts, including the features your team picked up when you doubled seats in March. What I can do is move you to annual billing, which brings the effective rate within about 4% of what you paid last year. If the budget pressure is sharper than that, I would rather find an answer on a call than lose you to a spreadsheet. Would Thursday work?

One edit needed before sending: checking the 4% figure. The vague version needed a rewrite. That gap, multiplied across every email where something is actually at stake, is the entire case for spending thirty seconds on a proper prompt. Notice also what made the draft good: the seat-doubling detail came from the pasted thread, not from the prompt. That is why you paste the whole thread.

The step this page has been building to: stop pasting

Every prompt above works by copy-paste, and copy-paste is the tax you pay at every single use. Claude's built-in Gmail connection removes it: once connected, Claude searches your real mail, reads the actual threads, and writes replies straight into your Gmail drafts folder, always as drafts, never sent on its own. The prompts change shape from "here is a thread [paste]" to "find the renewal thread from Priya and draft the reply," and the context Claude draws on is the whole conversation rather than whatever you remembered to include. Setup takes about three clicks; our connect Gmail to Claude guide covers it along with the two settings worth checking first, action confirmation and the Model Improvement privacy setting, and the multiple accounts guide covers the one-account-at-a-time boundary honestly.

One boundary to respect: the built-in connection is for your own mailbox. If the email you are answering lives at a team address like support@ or sales@, prompting from your personal login would mean the AI acts as you, invisibly to your team, on conversations the team owns. That situation deserves its own tools, which is the next section.

If you want the full system: Drag connects Gmail, your team, and the AI

For disclosure, this is where our own product lives, and it exists for the reader who got to this point thinking "I don't want to run this by hand at all." Drag turns a Gmail address like support@ into a shared inbox the whole team works from their own accounts, and its AI component does inside the inbox what the prompts above do outside it: drafting replies in the team's voice, triaging and tagging what arrives, and summarizing threads for handovers, included in the seat from $18 per user per month on plans from $12, with a 7-day trial and no card. The prompts you liked on this page have a team equivalent too: saved templates, reusable by everyone who answers from the address.

Two more pieces close the loop. Drag's Intelligence, rolling out now, takes routine conversations all the way to resolution under the team's ownership rules, with everything else escalating to a named human. And Drag ships its own MCP server, which means the same Claude you have been prompting can operate the team's queue directly, with every action attributed: the connection guide takes minutes, the hands-on guide shows the day-to-day, and if you are still weighing whether AI on a team inbox is worth it at all, the AI shared inbox guide is the deciding page.

Frequently asked questions

Is Claude good at writing emails?

Yes, with two practical strengths: it handles an entire pasted thread without losing details, and it holds a described voice consistently across a conversation. Like any assistant, it is only as good as the instruction: prompts that specify how the email should behave produce sendable drafts, while one-line prompts produce generic ones.

What is the best way to prompt Claude for an email?

Four parts: a role that sets the voice, the full pasted thread rather than a summary, the goal stated as an outcome the email must achieve, and constraints on length and tone. The worked example above shows the difference on a real negotiation: one factual check versus a full rewrite.

Can Claude read my Gmail?

Yes, if you connect it. Claude's built-in Gmail connection lets it search mail, summarize threads, and write replies straight into your drafts folder. It reads message text but not attachment contents, connects to one Google account at a time, and every draft waits for you to review and send it yourself.

Can Claude send emails for me?

No. Claude writes drafts, and sending is always yours: replies it produces through the Gmail connection appear in your drafts folder for review. Treat that boundary as a feature, because an assistant that cannot send from your address can never surprise you in front of a customer.

Do these prompts work on Claude's free plan?

Yes. Every prompt on this page works in a free Claude conversation by pasting the thread in, and the Gmail connection is available on the free tier too. Heavier daily use and very long threads fit the paid plans' larger limits better, but nothing here requires the API or any add-on.

Should I use Claude or another assistant for email?

For pasted-in prompting, the structure matters far more than the model, and our 50+ prompt library covers the same pattern across every situation including sales and marketing. Choose by where your workflow lives and which assistant is already connected to your inbox, not by prompt lists.

Where this fits

Prompting is the first rung of a ladder this blog covers end to end: connecting Gmail to Claude removes the pasting, the multiple accounts guide handles the work-plus-personal reality, the AI shared inbox guide is the decision page when the inbox belongs to a team, and the AI email assistant guide maps the tools that live in the inbox itself.