How to Use Muse to Automate a Report

Mostly yes — think 'semi-automate.' Muse, Meta's personal AI agent, can gather information across the services you connect and the web, draft the narrative and summary sections, format the result, and prepare the send to your client or boss. It works asynchronously, so the draft can be finished before you are back at your desk, and it remembers your format, metrics, and tone between runs. What it will not do, as of launch, is fire on a schedule by itself: Meta has not announced recurring scheduled tasks, so you re-prompt each cycle — which its memory turns into a one-line job.

• No scheduled or recurring automation has been announced. If your report must run every Friday at 9 a.m. with zero human involvement, Muse alone cannot do that yet.
• Muse can only see data in services you have connected or sources it can reach with its browser. Numbers it cannot pull should be clearly marked in the draft, not estimated.

1. 1

   Show it one finished report first

   Paste or describe last cycle's report. Muse's memory holds the structure, so every future run starts from your format instead of a generic one.

2. 2

   Name the data and where it lives

   List the metrics, their sources, and anything it must browse for. Be explicit about which numbers it should pull and which you will paste in yourself.

3. 3

   Ask for a draft, not a send

   'Prepare the report and the covering email; send nothing.' Muse checks before sensitive actions anyway — make it contractual in the prompt.

4. 4

   Build the checklist into the ask

   Tell Muse to flag missing data, odd numbers, and anything it could not verify, so your review takes minutes instead of a forensic audit.

A freelance SEO consultant produces a Friday report for one client covering keyword rankings, traffic, and the week's completed work.

Goal: Build this Friday's SEO report for Harbor & Lane and prep the send.
Context: Same structure as last week: summary, rankings, traffic, work completed, next week. The client only reads the summary, so lead strong.
Requirements:
- Pull current rankings for our 20 tracked keywords and this week's traffic from the dashboard I can reach in the browser.
- Write variance notes versus last week for anything that moved more than 10%.
- Mark anything you couldn't pull as UNVERIFIED instead of guessing.
Actions:
- Draft the covering email to Dana: two sentences, plain tone, report pasted below the signature.
Approval: Nothing sends. Both drafts wait for my review.
Output: Summary, full report, email draft, and the list of metrics you couldn't reach.

• A past report as the model

  One real example teaches Muse your structure and tone; after that, its memory reproduces them on every run.

• Metric definitions and sources

  'Traffic' means five different numbers. Muse needs to know which one, from where, for which period.

• The comparison baseline

  Variance notes need last cycle's numbers — say whether Muse should recall them, re-gather them, or receive them from you.

• Recipient and tone

  The covering email to your CFO and to a friendly client should not sound the same. Say which voice this run needs.

• What is out of its reach

  Data behind logins or services you have not connected should be flagged for manual paste, never improvised.

For this task, it is smart to make Muse check with you before it takes certain actions. Decide your rules up front and include them in the prompt.

• Sending the report or covering email — This is the consequential step. Muse checks before sensitive actions like sending an email — keep it that way and read every draft.

• Connecting a new service to reach the data — Credentials live in Muse's secure VM, but you decide which services it connects to. Grant new access deliberately, one at a time.

• Conclusions drawn from partial data — A report with one invented number is worse than a late report. Make Muse flag gaps, then fill them yourself.

• Changing the report format unasked — If Muse 'improves' your structure mid-run, your client notices. Lock the format in the prompt and reject drift.

• ✓Keep a standing 'report mode' message — structure, metrics, sources, and the send-nothing rule. By run three, 'same as last week, new numbers' is the whole prompt.
• ✓Asynchronous is the point: launch the gather-and-draft run before dinner and review it when it is done. Muse keeps working after you close the app.
• ✓Make the UNVERIFIED label mandatory in your prompt. It converts silent error risk into a visible to-do list you can clear in minutes.
• ✓For data Muse cannot reach, ask it to leave clearly marked slots instead of estimates. Filling slots beats editing invented numbers.
• ✓Re-prompting each cycle is a feature while the product is days old: you sanity-check every run until you trust it — which Reuters' launch reporting suggests is the right instinct, given internal tests hit failure modes.