A useful first Meta Muse prompt describes one job, its constraints and what a finished result should contain. “Help me shop” leaves too much undefined; a shortlist with a budget and retailer links gives you something you can check.
Meta describes Muse as an agent that can browse, handle tasks across services and continue longer jobs in the background. These examples are suggested prompts, not tasks we completed in a hands-on test. Actual results depend on access, connected services and the task.
AI-generated editorial illustration of the five suggested task types; not a Muse screenshot or evidence of completed tasks.
If you have not installed it, start with our Muse download and availability checks. For an overview of the product, read what Meta Muse is.
1. Compare a purchase with a real budget
Find three wireless headphones I can buy in the US for no more than $100 including delivery. Prioritize comfort for two-hour calls and multipoint Bluetooth. Use retailer or manufacturer links, distinguish new from refurbished units, and flag any missing delivery price. Show a comparison table. Do not buy anything.
Why this prompt is useful: it defines the total budget, the use case and how uncertainty should be handled. Without those details, a technically good recommendation can still be the wrong purchase.
Check the result: open each retailer page. Compare the model number, condition, stock, delivery cost and return policy. If shipping is unknown, the total has not yet been established.
Follow-up: “Remove any option whose total cost cannot be confirmed, and explain the trade-off between the remaining two.”
2. Research one day of a trip
Plan a rainy-day afternoon in Amsterdam on November 23, 2026, for two adults. Start near Amsterdam Centraal at 13:00 and finish by 18:00. Keep admission under €40 per person. Include travel time, official opening-hours links and a backup if tickets are unavailable. Do not book.
Why this prompt is useful: a date, starting point and time window make the recommendations easier to assess than a general list of attractions.
Check the result: confirm opening hours for that specific day and inspect the official booking pages. Look for a route that fits the time window after allowing for queues and transport.
Follow-up: “Reduce walking and keep only two main stops. Show which details remain unconfirmed.”
3. Build a meal plan around what you have
I have rice, canned chickpeas, eggs, carrots and frozen peas. Suggest three dinners for two adults using these ingredients. Ask about dietary restrictions first. Then give quantities, steps and a shopping list containing only missing items. Keep each meal under 30 minutes.
Why this prompt is useful: it provides starting inventory and defines what belongs on the shopping list. It also gives the assistant a reason to ask one important question before writing recipes.
Check the result: compare ingredient quantities with what you actually own. Check labels for relevant allergens and assess whether the cooking time is realistic for your kitchen.
Follow-up: “Combine duplicate shopping items and suggest one substitution for anything expensive.”
4. Turn a project into a short checklist
Help me organize my home office this weekend. I have 90 minutes on Saturday and 45 on Sunday. Ask which areas are causing problems, then make a prioritized checklist. Include a stopping point for each session and a place to put items I cannot decide about yet. Do not recommend buying storage before we know what needs storing.
Why this prompt is useful: the time limit prevents an ambitious plan from becoming a second unfinished project. A stopping point also makes the plan easier to follow when you run out of time.
Check the result: add up the estimated minutes. Move optional work below the tasks that actually solve your problem.
Follow-up: “I finished the desk but not the drawers. Adjust Sunday’s plan to fit 45 minutes.”
5. Review calendar conflicts before changing anything
This example is for when you deliberately choose to connect a calendar. First inspect the permissions offered to your account.
Read my calendar for the next seven days and identify overlapping appointments or days with less than 30 minutes between meetings. Summarize the conflicts with dates and times. Do not create, move or cancel events. Ask me about travel requirements before suggesting changes.
Why this prompt is useful: it gives a concrete purpose for the connection and separates identifying problems from modifying events.
Check the result: compare every flagged conflict with the calendar itself. Confirm time zones, recurring events and whether a short gap is actually a problem for you.
Follow-up: “Suggest changes for Tuesday only, but leave the calendar unchanged.”
Meta’s security documentation describes permission controls and an approval system. Where supported, access to read information can be separated from permission to change it. A written instruction in a prompt is not a substitute for reviewing those controls.
A prompt template you can reuse
Help me [specific task]. My constraints are [budget, dates, location or time]. Use [preferred sources or information]. Return [table, shortlist, draft or checklist]. Flag anything you cannot confirm. Before [sending, booking, buying or changing data], ask for approval.
Replace the brackets with actual requirements. If you do not know a key detail, ask the agent to clarify it before proceeding. Otherwise, give enough information to let it start.
Keep the requested output small enough to inspect. A three-option shortlist with sources is often more useful than a long list you cannot verify.
How to judge whether Muse helped
Consider the full task: writing the request, waiting, checking and correcting. Ask whether the result saved time overall and whether you can identify what the agent actually did.
For connected tasks, check the underlying app rather than relying entirely on a completion message. Meta’s security documentation acknowledges that the agent can make mistakes. Treat a confident answer as something to assess, not confirmation that every step succeeded.
Sources and scope
- Meta’s launch announcement: advertised task and background-work capabilities.
- How We Built Safety Into Muse: permissions, approvals and acknowledged limitations.
Checked September 30, 2026. Prompts and evaluation advice are editorial examples. This article does not claim tested task completion, specific integrations available to every account or a measured success rate.
