Muse Phone Calls: US Eligibility, Paid Plans and How to Give a Clear Request

Meta says Muse can now make phone calls for paid subscribers in the US. Alexandr Wang's November 12 announcement names errands such as restaurant reservations and calls to cable providers. The useful question is what authority to give the agent, and what evidence to expect when it comes back.

The announcement from Meta's chief AI officer confirms the rollout's audience, but does not publish a call-minute allowance, maximum duration or success rate. Do not read “paid subscribers” as a promise of unlimited calling, or assume that access to Muse in another country includes this feature.

Which Muse plan do you need?

Meta's subscription help page lists Power at $20 per month and Maximum at $100 per month. The weekly allowances are 500 million and 3 billion Muse tokens respectively. Subscriptions renew monthly unless cancelled, and Meta says available benefits can vary by region and account.

Those are general subscription allowances, not a published conversion into telephone minutes. The launch post says paid US subscribers; it does not identify a separate calling package or explain how a call consumes the allowance. Check the benefits shown for your account before paying specifically for calling.

If Muse says it cannot make a call, first check your subscription and US eligibility. Then ask whether calling is available on your account. Repeatedly rephrasing the same request is not a reliable way to distinguish a rollout restriction from an unsupported task. A text answer about a business is also not evidence that a call occurred.

Give Muse a bounded job, not a vague errand

Start with a low-stakes business inquiry where the possible outcomes are clear. A restaurant is a better first test than changing an insurance policy or resolving an urgent account problem. This is a suggested workflow, not a claim that I have tested Muse's calling feature.

For a reservation inquiry, separate finding availability from making a commitment. Here is a prompt template you can adapt:

Call [restaurant name and city] using the number on its official website.
Ask whether it has a table for four on [date] between 7:00 and 8:00 p.m.
local time, and whether any deposit, cancellation or no-show fee applies.

For this call, only collect information. Do not book, join a waitlist,
provide payment details or accept a charge. If the restaurant needs a
name, use [the name I authorize]. Ask me before sharing anything else.

Report the available times and terms, whether you reached a person,
and whether anything remains unconfirmed. If nobody answers, tell me;
do not keep calling repeatedly.

This template deliberately asks for less than the feature may be capable of doing. It makes the first result easy to inspect and prevents an availability inquiry from quietly becoming a paid booking. Once you have the details, give a separate booking instruction for the specific time and terms you accept.

The date, city and local-time wording matter. “Friday evening” is convenient in conversation but leaves room for a wrong day or timezone. Likewise, “get me a table” leaves unanswered whether a deposit, an outdoor table or a waitlist would count as success.

Verify the outcome, not just the activity

After an inquiry, expect a concrete answer: who was reached, which times were offered, what conditions apply and what is still unknown. After an authorized booking, ask for the restaurant's confirmation details and any confirmation message it provided. “I called” and “your reservation is confirmed” describe different outcomes.

If a response is ambiguous, avoid immediately asking for another booking attempt. First establish whether the restaurant already has a reservation under your name. Repeating an uncertain action can create duplicate bookings rather than resolve the uncertainty.

Treat a voicemail, a request to call back and a waitlist as their own statuses. None is a confirmed table. For time-sensitive plans, keep enough time to call the business yourself if the agent cannot establish the result.

A prompt helps define permission; it does not guarantee execution

Meta's published safety approach describes a separate Sentinel system that evaluates actions and can ask for approval. Meta also acknowledges that Muse can make mistakes. That is broader product architecture, not a call-specific guarantee that every instruction above is mechanically enforced.

Keep consequential decisions with you. Do not treat a general instruction to “handle it” as a sensible substitute for deciding what information can be shared, what money can be committed or which account changes are acceptable. If the conversation moves into a medical, financial or legal decision, take over rather than expanding the task casually.

Phone automation is useful when it removes waiting without making the result harder to trust. A narrow request and a clear confirmation standard are what let you delegate the hold music without delegating every decision that might follow.

I'm Paras. I work at the intersection of AI, engineering and distribution.

I help AI teams explain product capabilities and real workflow limits in stories people can understand and use.

Book a 30-minute call