Claude Prompt Library

Claude Prompts for Gmail

35 copy-paste prompts

Claude does not plug into Gmail, and it does not need to. Paste a thread, a screenful of subject lines, or an export, and these 35 prompts hand back a triage list, a finished reply, a follow-up sequence, or a filter rule you copy straight into Gmail settings.

In short: This page contains 35 copy-paste ready prompts, organized into 5 categories with a description and pro tip for each. The first 5 prompts are free instantly, no signup needed. Hand-curated and tested by the AI Academy team.

Louis Corneloup
By Louis Corneloup · Founder, Techpresso
Last updated ·Hand-curated & tested by the AI Academy team

Inbox Triage & Summaries

7 prompts

Morning Inbox Triage

1/35

✨ What it does

Claude sorts the pasted Gmail dump into a morning triage and names the five messages [ROLE] should clear first against [PRIORITIES]. You handle those five, then batch or ignore the rest.

You are a chief of staff triaging a busy inbox before the day starts. I will paste raw text copied out of Gmail. <context>[PASTE 20 to 60 emails: sender, subject, date and the preview snippet of each]. My role: [ROLE]. My three priorities this week: [PRIORITIES]. People I can never leave waiting: [NAMES].</context> <task>Sort every message into Reply now, Reply today, Delegate, Read later, or Archive. For each Reply now item, state the actual ask in one line and name who is blocked until I answer.</task> <constraints>Judge by consequence of delay, not by sender seniority. If a message has no ask in it, it is not a Reply now. Say when you are guessing.</constraints> <format>Five grouped lists, one line per message as sender / subject / the ask. Then a numbered list of the five things to clear before 10am.</format>

💡

Pro tip: Select the message list in Gmail and copy it. Sender, subject and snippet paste through as plain text, which is everything Claude needs to sort them.

Backlog Catch-Up After Time Off

2/35

✨ What it does

Claude rebuilds what happened between [START DATE] and [END DATE] into decided, waiting, overdue, and safe to ignore. You answer the overdue threads and star the waiting ones before you reopen projects.

You are rebuilding a week I missed from my inbox alone. I will paste everything that landed while I was away. <context>[PASTE the sender, subject and snippet of every email received between [START DATE] and [END DATE]]. I was out [NUMBER] days. Live projects: [PROJECTS]. My role: [ROLE].</context> <task>Reconstruct what happened: decisions taken without me, threads that stalled waiting on me, anything now past its date, and everything I can safely archive unread.</task> <constraints>Quote the line that proves each decision. Do not invent outcomes for threads that simply went quiet, label those as unresolved.</constraints> <format>Four sections: Decided without me, Waiting on me, Now overdue, Archive unread. Finish with the first three replies I should send today.</format>

💡

Pro tip: Paste in date order, oldest first. Claude reads a backlog far more accurately when the sequence of events is intact.

Where This Thread Stands

3/35

✨ What it does

Claude states where the pasted thread stands, which decisions are locked, what is still open, and who owns each item, plus a paragraph you can forward. You send the catch-up and act only on your open items.

You are summarizing one long email thread so a busy reader can act on it. I will paste the whole thread. <context>[PASTE the full thread, oldest message first, including quoted replies]. My role in it: [ROLE]. What I need out of it: [OUTCOME].</context> <task>State where things stand right now, which decisions are locked, which questions are still open, who owes what to whom with dates, and the single next step that unblocks it.</task> <constraints>Ignore pleasantries and repeated quoted text. If two people contradict each other, show both positions rather than picking one.</constraints> <format>Five short blocks with those headings, then a forwardable paragraph under 90 words for someone joining the thread late.</format>

💡

Pro tip: Use Gmail's "Show trimmed content" arrow before copying, otherwise the oldest and most useful part of the thread never makes it into your paste.

Sender Priority Map

4/35

✨ What it does

Claude groups a month of senders into four tiers: people who need a fast reply and people you should batch, based on [OUTCOMES] and [TEAM]. You star the top tier and build filters for the bottom one.

You are auditing who actually deserves my attention in Gmail. I will paste a month of inbox metadata. <context>[PASTE 100+ lines of sender name, sender address and subject from the last 30 days]. My job: [ROLE]. What I am measured on: [OUTCOMES]. Team I sit in: [TEAM].</context> <task>Group senders into four tiers: always read, batch once a day, batch once a week, and never needs me. Base it on whether their mail ever requires a decision from me.</task> <constraints>Count volume per sender. Flag any sender sending more than ten messages a month that has never needed a reply.</constraints> <format>A table of sender, volume, tier and a one-line reason. Then a shortlist of the addresses worth filtering out of the inbox entirely.</format>

💡

Pro tip: Search "in:inbox newer_than:30d" in Gmail, set the page size to 100 in settings, then copy the list. That is the fastest way to get a month of metadata out.

Decision Log From a Thread

5/35

✨ What it does

Claude pulls a decision log from the [PROJECT NAME] thread with who decided, when, and what, and lists guesses that were never confirmed. You paste the confirmed lines into the project doc and flag the guesses in the next reply.

You are extracting a decision log from email so it stops living in one person's memory. I will paste the thread. <context>[PASTE the full thread or several related threads]. Project: [PROJECT NAME]. Stakeholders and their roles: [NAMES AND ROLES].</context> <task>Pull out every decision made, who made it, when, and what it changed. Separately flag decisions that were implied or assumed but never explicitly confirmed by the person accountable.</task> <constraints>Quote the exact sentence that supports each entry. Never merge two half-decisions into one confident one.</constraints> <format>A table with Decision, Owner, Date, Evidence quote, Status of confirmed or implied. Then a short list of the implied ones and the one-line email that would confirm each.</format>

💡

Pro tip: The implied-decision list is the valuable half. Those are the items that resurface as "I never agreed to that" three weeks later.

Weekly Digest for My Manager

6/35

✨ What it does

Claude writes a short weekly update from the threads you pasted, aimed at what [WHAT THEY CARE ABOUT] and in their [DIRECT / DETAILED / NUMBERS FIRST] style. You add the one number they will ask for, then send it.

You are writing my weekly status update using only what is in my sent mail and inbox. I will paste the week. <context>[PASTE the threads that mattered this week: subjects, key messages, outcomes]. My manager cares about: [WHAT THEY CARE ABOUT]. Tone they prefer: [DIRECT / DETAILED / NUMBERS FIRST].</context> <task>Compress the week into a status update: what shipped, what is blocked and the specific ask to unblock it, what lands next week.</task> <constraints>Under 200 words total. Every blocked item must name a person and a decision needed, not just a complaint. No filler openers.</constraints> <format>Three labelled groups: Shipped (max 5 bullets), Blocked (max 3 bullets, each ending in an ask), Next week (max 2 bullets). Subject line at the top.</format>

💡

Pro tip: Paste your sent folder rather than your inbox. What you sent is a far better record of what you did than what other people sent you.

Escalation Radar

7/35

✨ What it does

Claude scores each live [NAME] thread for escalation risk and backs every score with a quoted line. You take the highest-risk thread first and call or write that person today.

You are reading client and stakeholder email for early signs of escalation. I will paste the live threads. <context>[PASTE the current threads with clients or senior stakeholders, most recent messages included]. Account or project: [NAME]. History worth knowing: [CONTEXT, PAST ISSUES].</context> <task>Score each thread from 1 to 5 on escalation risk and justify the score with the exact sentence that signals it. Recommend an intervention and who should send it.</task> <constraints>Look for tone shifts, dropped pleasantries, new names added to the recipients, shorter replies, and deadlines restated. Do not inflate scores, a 4 or 5 must be defensible.</constraints> <format>A ranked list, highest risk first: thread, score, trigger quote, recommended move, sender.</format>

💡

Pro tip: The strongest single signal is a new senior name quietly added to the recipients. Ask Claude to check the recipient list of each message, not just the body.

XML tags are just the start. Learn the full Claude workflow.

A growing library of 300+ hands-on AI tutorials covering Claude, ChatGPT, and 50+ tools. New tutorials added every week.

Start 7-Day Free Trial

Reply Drafting

7 prompts

Reply From Rough Bullets

8/35

✨ What it does

Claude turns [BULLETS] into a full reply to the pasted email that answers every question, and lists what your notes never covered. You fill the gaps, then send the draft.

You are drafting my reply from scrappy notes. I will paste the email and my bullets. <context>[PASTE the email I need to answer]. My rough notes on what to say: [BULLETS]. Who they are to me: [RELATIONSHIP]. Tone: [WARM / NEUTRAL / FORMAL].</context> <task>Turn the bullets into a complete reply that answers every question in their email, in the order they asked them.</task> <constraints>Under 150 words. Plain English. No opener about hoping they are well. Do not add commitments that are not in my notes.</constraints> <format>The reply body, ready to paste. Underneath, list anything they asked that my notes did not cover, so I can decide before sending.</format>

💡

Pro tip: The unanswered-questions list is the point. Half of all reply friction is answering three of four questions and forcing another round trip.

Firm No, Bridge Intact

9/35

✨ What it does

Claude writes a decline of [SCOPE, TIMING, PRICE, MEETING] that is clear in the first two lines and still protects the relationship. You add only as much of [REASON] as you planned to share, then send it.

You are helping me decline something without damaging the relationship. I will paste the request. <context>[PASTE the email making the request]. What I am declining: [SCOPE, TIMING, PRICE, MEETING]. Real reason: [REASON]. How much of the reason I am willing to share: [ALL / SOME / NONE]. Relationship value: [HIGH / MEDIUM / LOW].</context> <task>Write a reply that says no clearly in the first two lines, gives a proportionate reason, and offers one genuine alternative if a real one exists.</task> <constraints>No hedging, no maybe, no leaving the door ajar unless I said to. Under 120 words. Do not apologize more than once.</constraints> <format>The reply, then a one-line note on the single sentence most likely to be read as a maybe so I can harden it.</format>

💡

Pro tip: Ask for the softer and firmer version side by side. Reading them next to each other makes it obvious which one your recipient will actually hear as a no.

Warm Reply to a Cold Pitch

10/35

✨ What it does

Claude writes a short, interested reply to the cold pitch that asks [QUESTIONS] before you give up calendar time. You send it only if you are still [CURIOUS / EVALUATING / READY], and book nothing until they answer.

You are replying to unsolicited outreach that is worth a conversation. I will paste their pitch. <context>[PASTE the cold email I received]. What caught my interest: [WHAT]. What I still need to know before spending time: [QUESTIONS]. My buying stage: [CURIOUS / EVALUATING / READY].</context> <task>Write a short reply that shows real interest, asks the two or three qualifying questions that would save a wasted call, and proposes the lightest sensible next step.</task> <constraints>Under 90 words. Do not commit to a meeting unless my stage is evaluating or ready. Ask for pricing directly if it was not in their email.</constraints> <format>The reply, then the three questions listed separately in case I want to trim to one.</format>

💡

Pro tip: Ask for pricing in the first reply. It filters out roughly half of vendor threads before anyone books a call.

De-escalating Reply to an Angry Client

11/35

✨ What it does

Claude drafts a calm reply that owns [OUR FAULT, IF ANY], states [FACTS], and puts a date on [REMEDY] without promising [LIMITS]. You confirm the date with the team, then send it.

You are drafting a reply to an angry customer that lowers the temperature. I will paste their email. <context>[PASTE their message in full]. What actually happened on our side: [FACTS]. What we got wrong: [OUR FAULT, IF ANY]. What we can offer: [REMEDY]. What we cannot do: [LIMITS].</context> <task>Write a reply that names the specific thing that went wrong, takes proportionate responsibility, states the remedy with a date, and gives them one clear next step.</task> <constraints>No defensive framing, no passive voice hiding the actor, no blaming a process. Do not promise anything outside the stated remedy. Under 160 words.</constraints> <format>The reply. Then flag any sentence a lawyer or a finance lead would want removed before sending.</format>

💡

Pro tip: Have Claude flag the liability-sensitive lines separately. It is faster to delete a flagged sentence than to rewrite a whole draft after legal reads it.

Same Reply, Three Tones

12/35

✨ What it does

Claude writes the same reply in three registers, from collegial to on-the-record, with the same [SUBSTANCE]. You pick the register that matches the stakes and send only that one.

You are giving me tonal options for one message so I can pick the right register. I will paste the situation. <context>[PASTE the email I am answering]. What I need to communicate: [SUBSTANCE]. Recipient: [WHO THEY ARE, SENIORITY, HOW WELL I KNOW THEM]. Stakes: [LOW / MEDIUM / HIGH].</context> <task>Write the same reply three ways: warm and collegial, neutral and factual, and formal with a paper-trail quality. Keep the facts and commitments identical across all three.</task> <constraints>Each version under 120 words. Change register and structure, never the substance. No version should read as passive aggressive.</constraints> <format>Three labelled drafts, then one line on which situations each version fits.</format>

💡

Pro tip: When a thread might be forwarded upward later, send the formal version. Tone that reads warm to your peer can read sloppy to their director.

Reply That Buys Time

13/35

✨ What it does

Claude writes a short holding reply that names [DATE] instead of soon, shares [PARTIAL ANSWER, IF ANY], and matches their [HIGH / MEDIUM / LOW] urgency. You put the real date on your calendar, then send the hold.

You are writing a holding reply for something I cannot answer properly yet. I will paste the email. <context>[PASTE the email waiting on me]. Why I cannot answer now: [REASON]. When I realistically can: [DATE]. What I do know already: [PARTIAL ANSWER, IF ANY]. Their urgency: [HIGH / MEDIUM / LOW].</context> <task>Write a short holding reply that acknowledges the specific ask, gives whatever partial answer I already have, and commits to a firm date for the full answer.</task> <constraints>Under 70 words. Never say I will get back to you soon. Name a date. Do not over-explain the reason for the delay.</constraints> <format>The reply, plus a one-line calendar reminder text I can copy for the date I committed to.</format>

💡

Pro tip: Send holding replies within the hour. A 40-word acknowledgment on day one prevents the follow-up chase on day three.

Reply Up the Chain

14/35

✨ What it does

Claude rewrites your draft so the answer or [DECISION] sits in the first lines for a phone skim by [ROLE]. You delete the backstory, then send the inverted version.

You are rewriting my reply for an executive reader who will skim it on a phone. I will paste my draft. <context>[PASTE the email I received from a senior stakeholder]. [PASTE my current draft reply]. Their role: [ROLE]. What they are accountable for: [THEIR METRIC]. Decision I need from them, if any: [DECISION].</context> <task>Rewrite so the answer, the number, or the ask is in the first sentence, with supporting detail below in descending order of importance.</task> <constraints>Under 120 words. No preamble, no recap of what they already know. If I am asking for a decision, present at most two options with a recommendation.</constraints> <format>The rewritten reply, then a one-line note on what I buried in my original draft that mattered most.</format>

💡

Pro tip: Ask Claude to tell you what you buried. The line you put in paragraph three is usually the only line the executive needed.

Follow-ups & Chasing

7 prompts

Three-Touch Follow-Up Ladder

15/35

✨ What it does

Claude writes a three-email chase for the quiet thread, where each touch adds a new reason to reply and restates [SPECIFIC ASK]. You send the first, wait, and only then use the next.

You are writing a follow-up sequence for a thread that has gone quiet. I will paste the last message I sent. <context>[PASTE my original email and the date I sent it]. Who they are: [ROLE AND COMPANY]. What I need from them: [SPECIFIC ASK]. Why it matters to them: [THEIR BENEFIT]. Deadline, if real: [DATE].</context> <task>Write three follow-ups to send at day 3, day 8 and day 15. Each must add a new reason to reply rather than repeat the first email.</task> <constraints>Each under 70 words. Never write just checking in or bumping this. Touch two should add value or information, touch three should make it easy to say no.</constraints> <format>Three labelled drafts with the send date on each, plus the subject line to use (same thread or new).</format>

💡

Pro tip: Reply inside the original thread for touches one and two, then start a fresh subject for touch three. A new subject line escapes a buried thread.

Unpaid Invoice Chase

16/35

✨ What it does

Claude writes an overdue-invoice note pitched to [DAYS] late and a [NEW / LONG-STANDING] account, using [DETAILS] without burning the relationship. You attach the invoice, then send the matching pressure level.

You are chasing an overdue payment without damaging the account. I will paste the invoice details and any prior emails. <context>[PASTE previous payment emails and replies]. Invoice number and amount: [DETAILS]. Due date and days overdue: [DAYS]. Contact and their role: [NAME, ROLE]. Client relationship: [NEW / LONG-STANDING / AT RISK]. Contract terms on late payment: [TERMS].</context> <task>Write the next chase email at the right pressure level for how overdue it is, and name the concrete next step if it stays unpaid.</task> <constraints>Factual, unemotional, no apology for chasing. Reference the invoice number and due date. Mention contract terms only if past 30 days overdue.</constraints> <format>The email, then a two-line note on when to escalate and to whom.</format>

💡

Pro tip: Ask Claude for the version addressed to accounts payable and the version to your day-to-day contact. They need different tones and often different information.

Nudge a Silent Decision Maker

17/35

✨ What it does

Claude writes a nudge that offers two options, states a default if they stay silent, and names [IMPACT AND WHO IS AFFECTED]. You send it, then act on the default when the date hits.

You are re-engaging someone who owns a decision and has stopped replying. I will paste the thread. <context>[PASTE the thread, including the point where it went quiet]. The decision they own: [DECISION]. What is blocked without it: [IMPACT AND WHO IS AFFECTED]. Their likely reason for silence: [BUSY / UNSURE / LOST INTEREST / INTERNAL BLOCKER].</context> <task>Write a nudge that makes replying take under 30 seconds: state the decision, give two options, and offer a default I will proceed with if I hear nothing by a date.</task> <constraints>Under 80 words. No guilt, no chasing language. The default must be reasonable enough that silence is an acceptable answer.</constraints> <format>The email with the two options as a short list and the default clearly stated with a date.</format>

💡

Pro tip: The default clause does the work. It converts a stalled decision into a deadline the other person has to actively stop.

Post-Meeting Recap With Owners

18/35

✨ What it does

Claude turns [PASTE raw meeting notes or a transcript] into a recap where every action has one owner and a date. You fix any wrong owner, then send the recap the same day.

You are turning meeting notes into the recap email that makes the outcomes stick. I will paste my notes. <context>[PASTE raw meeting notes or a transcript]. Meeting: [PURPOSE]. Attendees and roles: [NAMES]. Anything agreed before the meeting: [PRIOR CONTEXT].</context> <task>Write the recap: decisions made, actions with a named owner and a date, open questions with who will answer them, and the next checkpoint.</task> <constraints>Every action needs one owner, never a team name. If a date was not agreed, write date to confirm rather than inventing one. Under 250 words.</constraints> <format>Subject line, one-line purpose, then Decisions, Actions (owner and date per line), Open questions, Next checkpoint.</format>

💡

Pro tip: Send it within two hours. A recap that lands the same day becomes the record of the meeting, one that lands tomorrow becomes a topic of debate.

Break-Up Email

19/35

✨ What it does

Claude writes a last email that closes the chase after [TIME] of silence and either leaves the door open or not, based on [YES / NO]. You send it once and stop following up.

You are writing the last email in a thread that has gone cold, designed to close the loop cleanly. I will paste the history. <context>[PASTE the thread and how many follow-ups I have already sent]. What I was asking for: [ASK]. How long since their last reply: [TIME]. Do I want the door left open: [YES / NO].</context> <task>Write a short closing email that states I am stopping the follow-ups, leaves the file in a clean state, and makes it effortless for them to restart later if the door stays open.</task> <constraints>Under 60 words. No sarcasm, no guilt trip, no final pitch. One sentence on what happens to their file or request.</constraints> <format>The email, plus one alternative subject line likelier to get opened than the original thread subject.</format>

💡

Pro tip: Break-up emails get the highest reply rate of any follow-up. Write it to be genuinely final, not as bait, or the reply you get will be worthless.

Revive a Cold Thread

20/35

✨ What it does

Claude restarts the old thread using [NEW PRODUCT, PRICE, TEAM, RESULT] and what may have changed on their side, not a guilt trip about the stall. You send it as a new note and wait for a real reply.

You are restarting a conversation that died months ago. I will paste the old thread. <context>[PASTE the old thread and its date]. What changed on my side since: [NEW PRODUCT, PRICE, TEAM, RESULT]. What may have changed on theirs: [FUNDING, HIRE, LAUNCH, REORG]. Original reason it stalled: [REASON].</context> <task>Write a restart email that leads with what actually changed, references the old conversation in one clause, and asks a question that is easy to answer even if they forgot me.</task> <constraints>Under 90 words. Do not assume they remember me. Do not reopen with as discussed. The change must be genuinely relevant to their situation, not a feature announcement.</constraints> <format>The email, with a new subject line rather than a reply to the dead thread.</format>

💡

Pro tip: Paste the old thread even though it is stale. Claude uses their old objection to pick which change to lead with, and that beats any generic reopener.

Deadline Reminder With Stakes

21/35

✨ What it does

Claude writes a group reminder and individual nudges for [DATE] that state [REAL CONSEQUENCE] once and name who has not answered. You send the group note, then the individual ones only to [NAMES].

You are reminding a group about a deadline in a way that gets action, not acknowledgment. I will paste the original request. <context>[PASTE the original request email and the responses so far]. Deadline: [DATE]. Who has responded: [NAMES]. Who has not: [NAMES]. What actually happens if it slips: [REAL CONSEQUENCE].</context> <task>Write two versions: a group reminder that names what is missing without shaming anyone, and a direct one-to-one version for the people who have not responded.</task> <constraints>State the real consequence once, plainly, without threats. Each under 80 words. Include exactly what to send and where.</constraints> <format>Two labelled drafts, group and individual, each with its subject line.</format>

💡

Pro tip: Send the group version to those who delivered and the individual version to those who did not. Public chasing of named people costs you goodwill for no extra speed.

These prompts give you the what. Tutorials give you the why.

Learn when to use extended thinking, how to build Claude Projects, and workflows that compound. 300+ tutorials and growing.

Try AI Academy Free

Newsletters & Filters

7 prompts

Filter Rules for Noisy Senders

22/35

✨ What it does

Claude turns your 30-day sender list into Gmail filter fields you can type in, so [TYPES] stay in the inbox and the noise does not. You create one filter, watch it for a day, then add the next.

You are designing Gmail filters from my actual mail volume. I cannot connect you to Gmail, so I will paste the data and copy your rules back in by hand. <context>[PASTE a list of sender addresses and subjects from the last 30 days, ideally 100+ lines]. Mail I must always see in the inbox: [TYPES]. Mail I only review in batches: [TYPES].</context> <task>Propose a filter set that keeps the inbox for mail needing a human decision. For each filter give the exact values to type into Gmail's create-filter fields.</task> <constraints>Use only real Gmail filter fields: From, To, Subject, Has the words, Doesn't have, Size. Actions limited to skip inbox, apply label, mark read, star, forward, delete. Never filter a sender who has ever needed a reply.</constraints> <format>A table with Filter name, From, Has the words, Doesn't have, Actions, Label. Then a warning list of senders that look risky to auto-archive.</format>

💡

Pro tip: Build one filter, then use Settings, Filters and Blocked Addresses, Export to get the XML. Paste that XML back to Claude as a template so it returns an importable file next time.

Label Taxonomy Build

23/35

✨ What it does

Claude designs at most 12 labels from your sample subjects, with rules, what should auto-apply, and which current labels to delete. You create the new labels, apply them for a week, then delete the ones you never used.

You are designing a Gmail label system that survives contact with a real week. I will paste what my mail actually looks like. <context>[PASTE a representative sample of 50 to 100 subject lines with senders]. My role: [ROLE]. How I work: [PROJECT-BASED / CLIENT-BASED / FUNCTIONAL]. Labels I use today and what annoys me about them: [CURRENT SYSTEM].</context> <task>Design a label taxonomy with no more than 12 top-level labels and a clear rule for what goes in each, plus which labels should be auto-applied by filter and which need a human.</task> <constraints>No label that requires me to make a judgment call twice. Nested labels only where the parent is genuinely used on its own. Say which of my current labels to delete.</constraints> <format>A tree of labels with a one-line rule under each, a table of which are filter-applied, and a delete list.</format>

💡

Pro tip: Gmail nests labels with a slash in the name, so Clients/Acme creates the hierarchy. Ask Claude to output names in that format and you can paste them straight in.

Search Operator Cheat Sheet

24/35

✨ What it does

Claude writes real Gmail search strings for the hunts you repeat, such as attachments, unanswered mail, or invoices from a domain. You save the three you will actually use and try one search now.

You are writing me a Gmail search cheat sheet for the searches I actually run. I will describe the hunts I repeat. <context>[LIST the things I search for repeatedly, for example: attachments from a client last quarter, unanswered mail I sent, invoices over a size, anything from a domain]. My role: [ROLE]. Common senders and domains: [LIST].</context> <task>Write each hunt as a single Gmail search string using real operators, and explain in one line what each operator does so I can adapt it later.</task> <constraints>Use only operators Gmail supports, such as from, to, subject, has:attachment, larger, older_than, newer_than, label, in:sent, is:unread, filename, and the minus and OR syntax. No invented operators.</constraints> <format>A table of Goal, Search string, What the operator does. Then five combined searches worth saving as bookmarks.</format>

💡

Pro tip: Paste a search that returned the wrong results and ask Claude to debug it. Most misses come from missing quotes around multi-word terms.

Newsletter Cull List

25/35

✨ What it does

Claude marks each newsletter from the last 90 days as keep, batch, or unsubscribe against [ROLE, TOPICS] and your weekly [MINUTES]. You unsubscribe the drop list and filter the batch pile out of the inbox.

You are auditing which newsletters earn their place in my inbox. I will paste the last three months of them. <context>[PASTE sender and subject lines of every newsletter and marketing email from the last 90 days]. My role and interests: [ROLE, TOPICS]. How much reading time I have per week: [MINUTES].</context> <task>Sort every publication into keep in inbox, move to a label and read weekly, or unsubscribe. Judge on how often it has changed a decision or taught something I would act on.</task> <constraints>Be ruthless on frequency: anything daily that is not directly job-critical goes to a label at best. Flag senders that are marketing dressed as a newsletter.</constraints> <format>Three lists with a one-line verdict per publication, then the filter values to route the keep-but-batch group to a label.</format>

💡

Pro tip: Search "unsubscribe" in Gmail to pull the full list fast. It catches marketing mail that never calls itself a newsletter.

Route Alerts Out of the Inbox

26/35

✨ What it does

Claude writes filters that keep [LIST] alerts visible and push the rest of the machine mail into a weekly sweep. You create the urgent filter first, then move the rest out of the main inbox.

You are separating machine mail from human mail. I will paste the automated notifications I receive. <context>[PASTE examples of automated mail: system alerts, ticket updates, CI notifications, calendar invites, form submissions, billing receipts]. Which alerts genuinely need me within the hour: [LIST]. Which I only review in a weekly sweep: [LIST].</context> <task>Design filters that keep only the hour-critical alerts in the inbox and route everything else to labels, with the exact field values to enter in Gmail.</task> <constraints>Match on the sending address or a stable phrase in the body, never on a subject line that changes per event. Never auto-delete anything with billing or legal content.</constraints> <format>A table of Alert type, From value, Has the words value, Action, Label. Then a short weekly sweep routine covering the labels in order.</format>

💡

Pro tip: Add a plus alias like [email protected] when signing up for tools. Filtering on the alias is more reliable than guessing sender addresses later.

Multiple Inbox Layout

27/35

✨ What it does

Claude defines up to five multiple-inbox sections with real search strings and rules for emptying them, matched to [ONE SCREEN / TWO / MOSTLY MOBILE]. You add the sections in Gmail settings and live with them for a week.

You are configuring Gmail's multiple inboxes so my screen matches my working day. I will describe how I work. <context>[DESCRIBE my day: when I triage, when I do focused work, what I am on the hook for]. My role: [ROLE]. Labels or stars I already use: [LIST]. Screen setup: [ONE SCREEN / TWO / MOSTLY MOBILE].</context> <task>Propose up to five multiple-inbox sections, each defined by a Gmail search string, with a name and the rule for when it should be empty.</task> <constraints>Use valid search syntax only, for example is:starred, label:waiting-on, from:[email protected], is:unread label:clients. Every section must have a clear emptying condition so it stays useful.</constraints> <format>A table of Section name, Search string, When it should be empty, then the panel position and section size to set.</format>

💡

Pro tip: Multiple inboxes live under Settings, Inbox, Inbox type. Each section takes a raw search query, so anything Claude writes for search works there unchanged.

Digest Instead of Drip

28/35

✨ What it does

Claude turns this week's pasted newsletters into one digest aimed at [DECISIONS] and [TOPICS], so you read once instead of drip by drip. You scan the digest, click only what you will use, and archive the originals.

You are turning a week of scattered subscriptions into one readable digest. I will paste the mail. <context>[PASTE the full text of the newsletters and updates I received this week]. My role: [ROLE]. Decisions I am currently weighing: [DECISIONS]. Topics I track: [TOPICS].</context> <task>Produce a single digest: what genuinely matters to my decisions, what is noise, and any item worth forwarding to a colleague with a suggested one-line note.</task> <constraints>Group by theme, not by publication. Cite the source name for each item. Skip anything that is a product announcement with no bearing on my decisions.</constraints> <format>Three to five themed sections with three bullets each, then a Forward these list with the note text.</format>

💡

Pro tip: Pair this with the batch label from the cull prompt. Open the label once a week, select all, copy, and paste the lot into one Claude message.

Meeting & Scheduling Emails

7 prompts

Meeting Request That Gets Accepted

29/35

✨ What it does

Claude writes a meeting ask for [NAME, ROLE, SENIORITY] around [DECISION OR OUTCOME], with concrete slots and an invite description. You send the email and hold the calendar holds until they pick a time.

You are writing a meeting request that a busy person will actually accept. I will give you the context. <context>Who I am inviting: [NAME, ROLE, SENIORITY]. Why now: [TRIGGER]. What I want out of the meeting: [DECISION OR OUTCOME]. What is in it for them: [THEIR BENEFIT]. Length I need: [15 / 30 / 45 MIN]. My availability: [SLOTS AND TIME ZONE].</context> <task>Write an invitation email that states the outcome in the first line, justifies the length requested, and offers concrete slots rather than asking for their availability.</task> <constraints>Under 100 words. Never ask do you have time to chat. Always name the decision the meeting produces. Offer three slots across two days.</constraints> <format>Subject line, the email, then a two-line calendar invite description to paste into the event body.</format>

💡

Pro tip: Ask for 20 or 25 minutes instead of 30. Odd lengths signal a real agenda and slip into the gaps a full-slot request cannot.

Decline a Meeting, Keep the Relationship

30/35

✨ What it does

Claude writes a decline of the pasted invite that still gives the organizer an outcome, plus a short calendar note, based on [NO VALUE / WRONG PERSON / BAD TIMING / TOO SENIOR FOR THIS]. You send the decline and forward the work to [NAME] if needed.

You are declining a meeting invitation without damaging the relationship or the work. I will paste the invite. <context>[PASTE the meeting invitation or the request email]. Why I am declining: [NO VALUE / WRONG PERSON / BAD TIMING / TOO SENIOR FOR THIS]. Who should attend instead, if anyone: [NAME]. Do I still want the outcome: [YES / NO].</context> <task>Write a reply that declines clearly, protects the outcome the organizer wants, and offers the cheapest alternative such as a delegate, a written answer, or a shorter slot.</task> <constraints>Under 80 words. Do not invent a scheduling conflict if the real reason is that the meeting is unnecessary. Never leave the organizer without a path forward.</constraints> <format>The reply, then a one-line version for the calendar decline note.</format>

💡

Pro tip: Write the calendar decline note too. A blank decline reads as a snub, one line of context reads as a judgment call.

Reschedule Without the Ping-Pong

31/35

✨ What it does

Claude writes one reschedule email with slots in both time zones and a default booking, so you are not bouncing times. You send it, then update the invite when they confirm.

You are rescheduling a meeting in a single email instead of six. I will paste the thread. <context>[PASTE the existing invite and any scheduling messages so far]. Why I am moving it: [REASON]. How urgent the meeting still is: [URGENCY]. My real availability with time zone: [SLOTS]. Who else must be in the room: [NAMES].</context> <task>Write one message that apologizes once, offers three specific slots with time zones spelled out, and states a default I will book if nobody objects by a date.</task> <constraints>Under 80 words. Never write let me know what works for you as the only option. Convert every slot into the recipient's time zone as well as mine.</constraints> <format>The email with the slots as a short list, both time zones per slot, and the default clearly marked.</format>

💡

Pro tip: Give Claude both cities, not the UTC offsets. It handles daylight saving correctly far more often when it reasons from the location.

Pre-Read Agenda Email

32/35

✨ What it does

Claude writes a pre-read that puts [DECISIONS], trade-offs, and timeboxes in front of [NAMES AND ROLES] before the meeting. You send it the day before and start the meeting from that list.

You are writing the agenda email that makes a high-stakes meeting productive. I will paste the background. <context>[PASTE the relevant thread, brief or notes]. Meeting purpose: [PURPOSE]. Attendees and what each is there to decide or contribute: [NAMES AND ROLES]. Decisions that must come out of it: [DECISIONS]. Time available: [MINUTES].</context> <task>Write a pre-read email: the decisions on the table, the options and their trade-offs, what each attendee should read or prepare, and a timeboxed agenda.</task> <constraints>Timeboxes must add up to the stated length with five minutes spare. Every agenda item names its owner. Anything that is information only goes below the agenda, not in it.</constraints> <format>Subject line, purpose in one line, decisions list, timeboxed agenda with owners, then a Read before section.</format>

💡

Pro tip: Send it 24 hours ahead and paste the same agenda into the calendar invite body. Nobody opens the email a second time, but everyone opens the invite.

Kill the Meeting, Send This Instead

33/35

✨ What it does

Claude writes a status update that replaces the meeting, with [FACTS, STATUS, NUMBERS] and named people for any leftover decisions. You send it to [NAMES] and cancel the hold if nobody objects.

You are replacing a meeting with a written update that does the same job. I will paste the meeting context. <context>[PASTE the invite, the agenda or the thread that triggered it]. Who was invited: [NAMES]. What the meeting was meant to achieve: [PURPOSE]. What I already know: [FACTS, STATUS, NUMBERS].</context> <task>Write the email that replaces it: the update, the specific decisions needed with named deciders and a reply-by date, and the one topic that genuinely still needs live discussion, if any.</task> <constraints>Under 250 words. Each decision must be answerable with a yes, a no, or a choice between two named options. Be honest if the meeting should actually happen.</constraints> <format>Subject line, status in five bullets, Decisions needed as a numbered list with decider and date, then Still needs a live conversation.</format>

💡

Pro tip: Ask Claude to tell you when the meeting should stay. A cancellation email for a genuinely contentious topic costs more time than the meeting would have.

Cross Time Zone Scheduling Email

34/35

✨ What it does

Claude proposes workable slots for [NAME, CITY for each] shown in every local time, plus the invitation email. You pick one default slot, send the invite, and stop the spreadsheet.

You are scheduling across time zones without a spreadsheet. I will list the participants and their locations. <context>Participants and cities: [NAME, CITY for each]. Meeting length: [MINUTES]. Working hours each is willing to stretch to: [EARLY / STANDARD / LATE]. Preferred week: [DATES]. Who is most senior or hardest to move: [NAME].</context> <task>Find the slots that fall in working hours for everyone, or if none exist, the fairest rotation of who takes the awkward hour. Then write the invitation email.</task> <constraints>Show each proposed slot in every participant's local time. Check for daylight saving changes inside the date range. Never put the same person in the awkward slot twice.</constraints> <format>A table of proposed slots by participant local time, a fairness note, then the invitation email itself.</format>

💡

Pro tip: Confirm the date range against daylight saving before sending. The US and Europe switch on different weekends, which quietly breaks recurring calls every spring.

Recurring Meeting Audit

35/35

✨ What it does

Claude gives each recurring invite a keep, shorten, thin, or cancel verdict, costed in attendee hours against the [HOURS] you want back. You cancel or shrink the first one this week and tell the attendees why.

You are auditing my recurring meetings using the invites and notes I will paste. <context>[PASTE the list of recurring invites: title, frequency, length, attendee count]. [PASTE recent notes or recaps for each, if any]. My role: [ROLE]. Hours per week I want back: [HOURS].</context> <task>For each recurring meeting, judge whether to keep, shorten, halve the frequency, cut attendees, or cancel. Calculate the hours per month each option returns across all attendees.</task> <constraints>Base the verdict on whether decisions come out of it, not on whether people like it. Count attendee hours, not just mine. Say which cancellations need a sponsor's approval.</constraints> <format>A table of Meeting, Verdict, Reason, Hours per month saved. Then the email I send to propose the changes.</format>

💡

Pro tip: Cost the meeting in total attendee hours per month. Eight people in a weekly 60-minute standup is 32 hours a month, and that number wins arguments.

Free tool

Prompt Optimizer

Turn a rough idea into a structured, professional AI prompt.

Try it free →

Frequently Asked Questions

Not in a normal chat window. Claude cannot open your mailbox, read unread messages, send on your behalf, or apply labels. Anthropic offers Google Workspace connectors on some paid and enterprise plans, but availability varies and a Workspace admin has to approve them. Every prompt on this page is written to need none of that: you paste text in, and you copy the output out.
Paste all of it, including the quoted history. Claude handles very long threads well, and the oldest messages usually contain the commitment or the constraint that explains the rest. In Gmail, click the "Show trimmed content" arrow before you copy, otherwise you only capture the last few replies.
That depends on your employer, not on Claude. Check your company policy before pasting anything with customer data, salaries, legal matters, or unreleased financials. Business and enterprise plans generally exclude your inputs from model training while consumer plans have different settings, so confirm which account you are signed into. When in doubt, replace names and figures with placeholders before pasting.
Go to Settings, then Filters and Blocked Addresses, then Create a new filter. Type the values Claude gave you into the matching fields such as From, Subject, Has the words, and Doesn't have, then choose the actions on the next screen. If you already have filters, export them from the same screen to get an XML file, paste that file to Claude as a template, and it can return new rules in the same importable format.
The Gemini panel is inside Gmail and can read the open message, so it is faster for a single quick reply. Claude sits outside Gmail and works on whatever you paste, which makes it better for the jobs that span many messages: triaging a hundred subject lines, auditing who emails you, extracting a decision log from a six-week thread, or designing a filter system from your real mail volume.
A fast model handles everyday drafting, triage and follow-ups perfectly well. Switch to the most capable model available on your plan for the high-stakes work: an angry-client reply, an escalation read, a decision log that people will argue about, or a filter design you will apply to years of mail.

Prompts are the starting line. Tutorials are the finish.

A growing library of 300+ hands-on tutorials on ChatGPT, Claude, Midjourney, and 50+ AI tools. New tutorials added every week.

7-day free trial. Cancel anytime.