AI Prompts to Prep for a Meeting From Your Email

You have a call in twenty minutes — maybe five. Your email history with the person on the other end runs back weeks, across multiple threads, split between a pricing discussion and a follow-up on something they asked about last month. You could scroll for context. Under time pressure you will miss half of it and skim the rest. The ai prompt to prepare for a meeting from emails solves this directly: paste the thread, run a targeted question, and get the context you need before the call starts.

The problem with a generic summarize request is that it gives you a recap, not a briefing. A recap tells you what was discussed. A briefing tells you where you stand — what was agreed, what is still open, what you owe them, and what they are likely to push on. These are four distinct outputs with four distinct prompts, and collapsing them into one request produces a vague paragraph that answers none of them cleanly.

The four prompts below are built for the five-minute window before a call. They work in any capable AI chat interface — ChatGPT, Claude, Gemini, Microsoft Copilot — and each produces a specific, actionable output that a general summary would bury.

What do you need before running these prompts?#

You need the email thread. In your email client, find the relevant correspondence with the person you are meeting, expand it fully so quoted history is included rather than collapsed, and copy the text. Oldest-first is the most reliable order: a model reading the conversation as a story from the beginning produces more accurate output than one working backwards from the most recent reply. If your client can only give you newest-first, add one line to your prompt noting the order.

How far back to go depends on the relationship. For a first call with a new contact, one or two threads usually cover what matters. For a long-running client relationship, the last three to six months captures current commitments without burying the model in context from closed projects. When in doubt, start with the five or six most recent messages and add more history if the output misses something important.

Strip the noise before you paste. Email signatures, legal disclaimers, unsubscribe footers, and repeated quoted blocks waste context and can mislead the model — a confidentiality disclaimer has been read as a substantive decision more than once. Delete obvious boilerplate and the output will be shorter and more reliable. One more step that matters: pasting a business thread into a consumer AI chatbot sends client names, deal terms, and your internal positions to a third-party server. On free and lower consumer tiers your inputs may train the model unless you turn that off. Use a temporary chat or disable history and training before pasting anything commercially sensitive.

• Expand the full thread before copying — collapsed quoted history is invisible to the model and will be skipped
• Paste oldest-first where possible; if you cannot, tell the model the order in one sentence
• Keep sender names and dates — they are the raw material for who agreed to what and when
• Strip signatures, disclaimers, and repeated quoted blocks before pasting to save context and avoid misleading the model
• On free and consumer-tier chatbots, disable history and training or use a temporary chat before pasting anything commercially sensitive

The four prompts, step by step#

Run each prompt as a separate message in the same chat session. The thread only needs to be pasted once — after that, the model still has it in context, so each follow-up prompt can reference it without a second paste. Paste the thread first, confirm the model has acknowledged it, then work through Prompts 1 to 4 in sequence.

1. 1

   Prompt 1: where you last left things

   This establishes your baseline. Paste the thread, then send: "You are preparing me for a call with [name]. From the email thread above, summarize where we currently stand in two to three sentences. Focus on the last agreed position, any commitments made by either side, and the current status of what we are working on together. Use only facts in the thread; write unclear for anything not stated." The output tells you what was agreed and whether it is still current — the single most useful thing to know before a call opens.

2. 2

   Prompt 2: open asks in both directions

   Before any call, you want to know who is waiting on what from whom. In the same chat, send: "From the thread above, list every open request or ask still waiting on a response. Give me two labeled bullet lists: What they asked me or my side, and What I asked them. Use only the current state — if a request was already answered, leave it out. Write None if there is nothing outstanding on either side." Seeing both lists side by side shows you what you owe and what you are entitled to follow up on.

3. 3

   Prompt 3: questions you owe them an answer to

   This is a narrower version of Prompt 2, focused only on your outstanding obligations. Send: "From the thread above, list every question they asked me or my organization that I have not yet answered. For each, include the question, the date it was asked, and which message it appeared in. Include questions that were asked but never formally addressed, replied to only partially, or may have been overlooked. Write None if there are no outstanding questions I owe them." Running this before a call means you are not caught by a question from three weeks ago that went unanswered.

4. 4

   Prompt 4: what they are likely to raise

   This prompt makes an inference rather than an extraction — read the output against the thread before relying on it. Send: "Based on the thread above, what are the three topics [name] is most likely to raise in our next conversation? Base your answer only on the thread: their stated concerns, items they have followed up on more than once, pending matters on their side, and anything unresolved. For each, give a one-sentence description and one sentence on the best response position based on what we have agreed so far." Use the output to think through your positions before the call, not as a script.

How these prompts perform across the main AI platforms#

The four prompts above work in any capable AI chat interface, but the experience varies. Verify current features against each vendor's own documentation — AI products change quickly, and packaging as of August 2026 may differ by the time you read this.

What to do when the prompts do not produce useful output#

The most common failure is incomplete input. If Prompt 2 returns two empty lists and you know there are open asks in the thread, the model did not see the full conversation. Expand every collapsed reply in your email client before copying — most webmail interfaces have a show trimmed content control or an ellipsis that reveals the quoted history. If you copied only the visible top message, the model summarized one reply and presented it as the whole relationship.

The second issue is missing attribution. If your email client strips sender names and dates from the copied text, the model cannot identify who asked what or build the timeline Prompt 1 depends on. Add one line before the thread: "The thread is between me, [your name], and [their name] at [their company]." That single header restores the attribution the model needs to fill in both sides of the open-asks list.

The third is over-relying on Prompt 4. That prompt asks for inference, not extraction. A model will produce a confident-sounding list of likely topics even when the evidence in the thread is thin. On a short thread or a new relationship, treat the output as a prompt for your own thinking before the call, not as a reliable prediction. Cross-check each suggested topic against the actual messages before you carry it in.

A faster way: meeting briefs that build themselves#

The four prompts above work. The cost is the loop: find the right threads, expand them, paste the text, run the prompts in sequence, copy the outputs into a note before the call starts. We build AI Emaily — an AI-native email client that removes that loop entirely. Its Living Brief feature generates the last agreed position, open asks, and outstanding questions directly from your real email thread, inside the inbox where the messages already live. There is no paste and no tab switch. The brief draws on your user-set Context brain, so it reflects how you actually communicate with this contact, not a generic model output. Your email is never copied to a third-party chatbot and is never used to train models. It works across Gmail, Outlook, iCloud, Fastmail, Proton, and IMAP. A 7-day free trial gives you full access — see aiemaily.com/pricing for plan details, or connect your inbox at aiemaily.com.