Playbooks · Everyday tasks · last verified 2026-10-23

Sort and summarise a messy inbox into next actions

For a freelancer × turning a backlog of unread messages into a next-actions list × with ChatGPT, pasting messages in small batches — the chatbot never opens your mailbox.

Before you start

You need a mailbox you can copy text from, and a decision about scope: pick one backlog — for example, unread messages from the last two weeks, or messages from one client. Do not try to paste an entire inbox in one sitting. OpenAI's help page describes working with text you bring into a conversation (documents, spreadsheets, presentations, images, and text files you attach), and pasting message text into a chat follows the same pattern: you supply the content, the chatbot works on what you supplied.

Time and cost: NOT_MEASURED. No inbox-zero claim is made here; the result is a list, not an empty inbox.

Before pasting anything, remove passwords, payment details, account numbers, and other people's private information from the pasted text. Paste only the messages you need sorted, and only the parts that carry the request or the decision. If a message contains sensitive material unrelated to the action, cut that part before pasting.

What "done" looks like

At the end of one sitting you have a single next-actions list — each item with the sender, what they asked for, and the date they asked — plus a short list of messages that need no action and a list of messages you could not sort from the text pasted. Replies are not sent in this sitting; writing them is the next sitting's work.

Steps

1. Pick the batch and say what counts as an action. Paste:

   I am going to paste messages from my inbox in small batches.
   For each message, decide: ACTION (someone is waiting on me),
   WAITING (I am waiting on them), or NO ACTION (information
   only). Do not invent deadlines, senders, or requests that
   are not in the pasted text. If a message is unclear, put it
   in UNCLEAR and say what is missing.

2. Paste one small batch at a time. Paste 5 to 10 messages, with sender and date at the top of each, then paste:

   Sort this batch into ACTION / WAITING / NO ACTION / UNCLEAR,
   as a table with columns: Sender / Date / What they asked /
   Category. Quote the sentence that carries the request.

   Quoting the request sentence is the check that makes the table auditable: every action traces to a sentence you can find in the original message.

3. Merge the batches into one list. After the last batch, paste:

   Combine all the batches above into one next-actions list,
   grouped by sender, with no duplicates. If the same request
   appears in two messages, keep the latest one and note the
   earlier one as a duplicate. Keep the UNCLEAR items as a
   separate section at the end.

4. Resolve the UNCLEAR items yourself. For each UNCLEAR item, open the original message and decide: reply asking for clarification, file it as information, or delete it. Do not let the chatbot resolve an unclear message by guessing what the sender meant.
5. Copy the list out of the chat. Paste it into your task list or notes. Work through it in your mail app, where replies, filing, and deleting actually happen — the chatbot does none of those in this playbook.

Check the agent's work

• Every ACTION item quotes a sentence from a message you pasted, and the sender and date match that message.
• No deadline appears that was not in the pasted text. A deadline the chatbot inferred from tone ("urgent") is not a deadline — check the message.
• Duplicates were actually merged: the same request from the same sender does not appear twice as two actions.
• WAITING items are genuinely things you are waiting on, not things you owe — mixing those two up is how a follow-up goes to the wrong person.
• UNCLEAR items were not quietly sorted into ACTION or NO ACTION. If the UNCLEAR section is empty after a messy backlog, re-check the batches.

What can go wrong

• It invents a request. A polite closing ("let me know your thoughts") becomes an action item. The quoted-sentence rule catches this: no quote, no action.
• It loses context across batches. A reply in batch 3 can change the meaning of a request in batch 1. The merge step exists for this, but if a thread spans batches, paste the whole thread in one batch instead of splitting it.
• It sorts by tone instead of content. A long, friendly message with no request is NO ACTION, and a short, blunt message with a request is ACTION. Category follows the request sentence, not the length or warmth of the message.
• Sensitive text gets pasted by habit. Copy-pasting whole messages is fast and leaky. Trim each message to the request and its context before pasting, every time.

Limits

• The chatbot does not open, read, or access your mailbox in this playbook. It only works on text you paste or files you attach. It cannot see messages you did not supply, and it cannot know about messages that arrived after you pasted a batch.
• It does not send, file, delete, or label mail, and it does not update a task system. Those are your steps, in your mail app and task list.
• It cannot judge urgency the way you can: a client's real deadline and a newsletter's manufactured urgency look similar in text. The final order of your list is your judgement, not the chatbot's table order.

Verification

Tier: AUTHOR_DOCS_VERIFIED · 2026-10-23. This page was not run end to end; its capability claims rest on provider documentation fetched this run. Sources: OpenAI Help Center — Uploading files and audio to ChatGPT (ChatGPT can work with documents, spreadsheets, presentations, images, and text files you attach to a conversation; "describe what you want ChatGPT to do with it"; review the response and ask follow-up questions). The inbox playbook relies on that same bring-your-own-content pattern, with pasted text instead of an attachment. Omitted: no claim is made that the chatbot can connect to a mailbox, read unread mail, or send, file, or delete messages — those steps were cut because they are not grounded in the fetched page, and the playbook says so in Limits. Limits of this verification: provider pages describe working with supplied content, not whether any given sort matches a given inbox — that check is the "Check the agent's work" section above.

Back to: Everyday playbooks hub (../../guides/everyday/) · Skills index (../../skills/)

Also: Small-team hub