AI Prompts for Email and Support: The Drag Prompt Library (2026)

The short version: this page is the index of the Drag prompt library. Seven pages, more than a hundred prompts, all built the same way: the prompt itself, a note on why it is written like that, and a worked example of what it produces. Nothing here is a scraped list. Every prompt was written against real inbox situations, and a small set of them goes further than copy-paste: connect an inbox once and they run the work themselves.

The four collections below each solve a different problem. Pick by the problem, not the tool: everything works in ChatGPT, Claude, and Gemini unless a page says otherwise.

ChatGPT prompts for email: the general collection

The biggest page in the library, and the place to start if the problem is "I need to write this email and I keep rewriting the first line." Cold outreach, follow-ups, saying no politely, delivering bad news, asking for things. The prompts work in any assistant even though the page says ChatGPT on the tin.

Read it: 50+ ChatGPT Prompts for Writing Perfect Emails in Any Situation

Claude prompts for email: the opinionated fifteen

The sharpest page for people who already use Claude. Fifteen prompts built around the failure modes of real replies, like answering the easy half of an email and burying the hard half, plus a section on when you should stop prompting by hand entirely.

Read it: Claude Prompts for Email: The 15 That Work

Shared inbox prompts: for support@ and sales@

Written for team addresses, where the expensive problems are not about writing. Three people drafting the same answer, a thread nobody owns, handoffs that lose their context: these fifteen cover triage, ownership, team voice, and weekly reporting, and three of them can run a connected inbox directly.

Read it: AI Prompts for a Shared Inbox: 15 That Run a Team's Email

Customer service prompts: the hard tickets

The page for the messages nobody wants to answer. Angry customers, refund refusals you have to hold, troubleshooting loops, the holding reply that buys time without lying. Eighteen prompts, and three of these can also run a connected inbox directly.

Read it: AI Prompts for Customer Service: 18 That Handle Real Tickets

Email triage prompts: getting to zero and staying there

The sorting page. Fifteen prompts for the inbox itself rather than the replies: the morning sort, digging out of a thousand-unread backlog, the routine that keeps the count at zero, and the connected versions that read and sort the real queue for you.

Read it: AI Prompts for Email Triage and Inbox Zero: 15 That Clear the Queue

Support QA and tone prompts: fix replies before they send

The review page. Instead of asking the AI to write a reply, these fifteen ask it to find what is wrong with the one you wrote: rewrite passes that keep the message, pre-send checks for completeness and accidental promises, a scoring rubric for team voice, and three connected prompts that QA the replies your team actually sent.

Read it: AI Prompts for Support QA and Tone: 15 That Fix Replies Before They Send

Support reporting prompts: numbers into decisions

The analysis page, and the clearest demonstration of the connected tier: weekly reviews that separate trends from bad Tuesdays, reply-time diagnosis, reports written for founders and boards, and four prompts that pull the real response times and volumes from a connected inbox instead of working from your export.

Read it: AI Prompts for Support Reporting and Weekly Reviews: 15 That Pull Real Numbers

The prompts that run your inbox for you

Most prompt pages on the internet assume one workflow: copy the prompt, paste your email, copy the answer back. That works, and most of the library works that way. But it has a ceiling, because you are the transport layer, and the assistant only ever sees what you paste.

A small set of prompts across the library is written for the other workflow. Connect an inbox to an assistant once, through an MCP server, and the same request stops being a writing exercise: the assistant searches the actual queue, drafts the actual replies, applies the actual labels, and tells you what it did. Here is the shape of one, from the shared inbox page:

Works in Claude, ChatGPT, or Gemini

Look at our support inbox for the last 7 days. Give me Monday's briefing: volume in and out, median first reply time, the three threads that took longest to resolve, and anything still unanswered older than 48 hours.

Runs on a connected inbox→

Copy

Pasted into a bare chat window, that prompt can only ask you for the data. Run against a connected inbox, it pulls the real numbers and the real threads, because the assistant can search and measure the queue itself. These are the prompts we mark as MCP-runnable on their pages. The prompts are written verbatim against Drag's own MCP server and its real tools, and each one is published with the result shape it produces, so you know what to expect before you run it. The connection guide covers the team setup in about ten minutes, and the hands-on guide shows the day-to-day. For a personal inbox, the per-assistant walkthroughs cover ChatGPT, Claude, and Gemini.

One rule holds across every page, runnable or not: the AI drafts, sorts, and summarizes, and a human reviews and sends. Nothing in this library puts a bot in front of your customers.

How these prompts are written

Three things distinguish the pages in this library from the fifty-prompt lists that rank alongside them, and we hold every new page to them.

First, every prompt carries its reasoning. A prompt you understand is a prompt you can adapt; a prompt you pasted blind breaks the first time your situation differs from the author's.

Second, the pages show their work. Worked examples and result cards show what a prompt produces before you spend a real email finding out, and every connected prompt states exactly what happens when it runs.

Third, the collections are category-native. We publish prompts for email and support work because that is the work our product does all day, and it is the work we can test prompts against honestly. There is no page of prompts for essay writing or trip planning here, and there will not be.

For disclosure: Drag is our product, a shared inbox that turns Gmail into a team workspace, from $12 per seat per month with Drag AI in the seat from $18 and the MCP server on the Pro plan. The library exists because prompt-driven email work is what our users do, but every copy-paste prompt on these pages works with no Drag account at all.

FAQ

What are AI prompts for email?

Instructions you give an AI assistant such as ChatGPT, Claude, or Gemini to draft, rewrite, summarize, or sort email. A good email prompt carries the context the model cannot see: who the reply is for, what tone the situation needs, and what a correct answer must include.

Which AI assistant works best for email prompts?

The prompts in this library are written to work in ChatGPT, Claude, and Gemini alike, and each page notes the differences where they matter. The bigger gap is not between assistants but between pasting threads into a chat window and connecting the inbox itself, which is what the MCP-runnable prompts do.

What is an MCP-runnable prompt?

A prompt written to execute against a connected inbox through an MCP server rather than against pasted text. Instead of you copying threads in and answers out, the assistant searches the inbox, drafts replies, and reports what it did. Drag ships its own MCP server for this, on its Pro plan.