What is Muse outbound calling?
Muse outbound calling is the feature where your Muse agent places a real phone call to a business on your behalf, talks through the request, and reports back to you. You ask Muse to call, Muse dials, and a voice on the other end of the line has a conversation with the person who answers. You read the result afterwards, instead of sitting on hold yourself.
The idea is older than the launch. At Meta's Connect conference in September 2026, Muse's calling ability was demonstrated calling a restaurant to change a reservation. The demo showed the shape of the feature: you hand Muse a job that needs a phone call, and Muse does the calling. The beta that followed tested it with a smaller group, and the general launch came on November 12, 2026.
Alexandr Wang, Meta's chief AI officer, announced the general availability on X the same day, writing: "AI is great at waiting on hold for you. No more wasting your own time. Muse can now place calls to US businesses. insurance? cable? DMV? dinner res? Muse has got it." TechCrunch reported the launch on November 12, confirming that the feature is available to paid subscribers in the US. Engadget covered it the same day, with the same paid US scope.
This guide covers the consumer side: you asking Muse to call a business for you. If you run a business and want Muse to call your customers, that is a different feature, covered in the business calling guide. The two share the same voice technology, but the direction, the setup, and the rules are different, and confusing them leads to the wrong expectations.
Who can use it?
Paid subscribers in the US. That is the launch audience, and it is narrower than the Muse user base in two ways. First, you need a paid plan. Power costs $20 a month and Maximum costs $100 a month, and both buy weekly usage capacity, not extra features, except that calling is a paid only feature at launch. Second, you need to be in the US, and the business being called needs to be a US business.
Free plan users cannot use calling yet. Meta has announced no date for a free tier version, and no date for any country beyond the US. Canada, Japan, Singapore, and Mexico, the other four countries where Muse is live, have no calling date either. If you are outside the US, the feature simply is not on your account, and no settings change will add it.
There is one ambiguity inside the paid group, and it is worth stating plainly. Two outlets flagged that it is unclear whether the $20 Power tier alone unlocks calling, or whether calling needs the $100 Maximum tier. Meta's announcement says "paid subscribers" without naming a tier, and the reporting does not settle which tier is the floor. If you pay for Power and calling is not on your account, that ambiguity is the likely reason, not a fault on your device. Check the app on your own account before upgrading on an assumption.
The beta history explains why the launch is paid only. Calling was in limited testing since September 2026, with a smaller group of users, before the November 12 opening. A beta that small, on a feature that makes real phone calls to real businesses, is a careful rollout. The paid gate is part of that caution, and so is the US only scope. Both may widen, but neither has a date.
How does it work?
You start in the Muse app, in chat. Ask Muse to call a business and say what you need. The request needs a business name, or a phone number, and a clear job: a reservation change, an appointment, a question about hours, or a hold queue you do not want to sit in. The clearer the job, the better the call goes. "Call the restaurant and change my Friday reservation from 7 to 8" is a job. "Call the restaurant" is not.
Muse then places the call. A voice on the line talks to the person who answers, works through the request, and handles the back and forth. If the business puts the call on hold, Muse waits. That is the use case Wang led with: "AI is great at waiting on hold for you." The hold queue is the part of a phone call that wastes your time without needing your judgment, and it is the part an agent can absorb.
When the call ends, Muse reports back. The report includes a transcript of the call and a summary of what happened, according to internal Meta material described by 404 Media. You read what was said, what was agreed, and what still needs you. The transcript is the record that lets you check the call, which matters, because you were not on it. Read the transcript before you trust the outcome, especially for bookings and appointments.
What Meta has not published is a help page describing the exact request flow: whether you pick a business from a list, type a number, or ask in plain chat, and what approval steps appear before the call starts. The beta and launch reporting describes the outcome, not the interface. Until Meta publishes a help page, the flow above, ask in chat, Muse calls, Muse reports back, is the shape the reporting supports, and the details on your own device are the authority.
What can Muse call about?
Wang's launch post named four kinds of call: insurance, cable, the DMV, and dinner reservations. Those four cover the pattern. Any call where the job is routine, the information is known, and the wait is the worst part fits. Reservations, appointments, hours, availability, and hold queues are the natural territory.
Reservations are the clearest example. The Connect demo showed Muse calling a restaurant to change a reservation, and dinner reservations head Wang's list. A reservation change is a bounded job with a known outcome: a time, a party size, and a name. Muse can handle the back and forth, confirm the change, and report the new time to you.
Appointments work the same way. A doctor, a dentist, a salon, or a car service all take booking calls, and the call is mostly a negotiation over times. Give Muse your availability, the service you need, and any reference number, and the call becomes a scheduling conversation with a clear end. Read the transcript afterwards to confirm the date and time, because a misheard date is the most likely failure in a booking call.
Hold queues are the use case that needs the least from you. A cable company, an insurer, or a government office with a long wait is a call where the first twenty minutes are music and menus. Muse waits, navigates the menu, and talks to the person when they answer. That is the job Wang described as the point of the feature, and it is the job where the agent's patience is a real advantage over yours.
What does not fit is anything that needs your judgment in real time, your identity in a form Muse should not carry, or a negotiation where the outcome depends on tone and pressure. A complaint where you want to be heard, a price negotiation where you want to push, or a call where the person on the other end needs to hear it from you directly are all better as your own call. Muse is a stand in for routine calls, not a replacement for the calls that matter because you made them.
What does Muse say on the call?
Muse identifies itself as calling on your behalf and says the call is being transcribed. That is what happened in a real world call to a Volvo dealership that was reported in detail. The agent told the person who answered that it was calling on behalf of the user and that the call was being transcribed. That disclosure is the part that happened. What did not happen, in that call, is an explicit statement that the caller was AI. The agent said who it was calling for and that the call was recorded, but did not say "I am an AI."
That gap matters, because the rules around it are real. Meta told Reuters in September 2026 that it was rolling back an earlier calling test after human contractors were found to be making calls, and said it was working on "proper disclosures" for the feature. The rollback shows Meta knew the disclosure question needed an answer before a wide launch. Whether the November launch disclosure, calling on behalf plus transcription, counts as a proper disclosure is the question the Volvo call leaves open, and Meta has not published a disclosure policy for the launched feature.
Some US states require all parties to consent to a recording, and a transcription is a form of recording. The transcript disclosure in the Volvo call addresses that, at least in part, by telling the person on the line that the call is being transcribed. But a transcription disclosure is not the same as an AI disclosure, and the two rules come from different places. One is about recording. The other is about whether a person knows they are talking to a machine.
Until Meta publishes its disclosure policy, the honest position is that Muse tells the person it is calling on your behalf and that the call is transcribed, and does not always say it is AI. If that matters to you, say so in your request. Tell Muse to identify itself as an AI agent at the start of the call. That instruction costs you nothing, and it closes the gap on your own calls even if the default does not.
What can it not do?
It cannot call personal phone numbers. A tester reported that Muse refused to call a personal number, and that limit makes sense. Calling a business is a transaction. Calling a person is a social act, and an agent that calls people on your behalf without their consent is a different product with different risks. If you need to reach a person, call them yourself, or send a message. Muse calling is for businesses.
It cannot call outside the US. The launch is US businesses only, called by US paid subscribers. A business in Canada, Japan, Singapore, or Mexico cannot be called, even though Muse is live in those countries. International numbers are outside the launch scope entirely, and no date has been announced for any expansion.
It cannot do what the business will not do over the phone. Some businesses require in person verification, a signature, or a payment method presented in person. A call cannot satisfy those, and Muse cannot invent a workaround. If the business says the job needs you in person, that is the answer, and the transcript will show it.
It may not work with every business. A business opt out form was mentioned in an October check of Meta's help pages, which would let a business decline AI calls, but that mention is unconfirmed elsewhere and no opt out list has been published. Whether any business has opted out, and how you would know before calling, is unknown. If a call fails or the business refuses to talk to the agent, that may be an opt out, or it may be a person who simply does not want to deal with a voice agent. Either way, the fallback is your own call.
It cannot guarantee accuracy. A misheard name, date, or number is the most likely failure, and the transcript is your check. Read it. Confirm bookings and appointments against the transcript before you rely on them, and confirm anything that involves money twice. An agent that calls for you is useful precisely because you do not have to listen, but the transcript is the price of that convenience, and skipping it is how a wrong time becomes a missed appointment.
How did we get here? The beta and the rollback
The feature's history has one awkward chapter, and it is worth telling straight, because it explains the caution around the launch. In September 2026, Reuters reported that Meta was testing AI phone calls using human contractors, who made calls on behalf of users while the AI capability was still being built. The test was real calls, made by real people, presented as part of an AI feature. When the reporting surfaced, Meta rolled the test back and said it was working on proper disclosures.
The rollback is not a reason to distrust the launched feature, but it is a reason to read the transcript. A company that tested with humans and called it AI has earned scrutiny on disclosure, and the Volvo call shows the disclosure question is still not fully settled. The launched feature is a voice agent, not a contractor, and the transcript is the evidence. Use it.
The beta that followed ran from September to November, in limited testing. A post by Ryan Fox on September 16 described the beta, and the reporting from that period shows a feature being tested carefully, with a small group, before the general opening. The November 12 launch is the end of that testing period, not the start of the feature. Two months of beta on a feature that calls real businesses is a reasonable pace, and it is faster than the year long waits that other voice agent launches have taken.
The history also explains the paid gate. A feature that makes real calls to real businesses costs money per call, in voice time and in compute, and a paid gate limits the volume while the system learns. Whether that is the reason, or whether the gate is about managing demand, Meta has not said. Either way, the gate is real, and free users are outside it for now.
How is this different from Muse Voice?
Muse Voice is you talking to Muse. Outbound calling is Muse talking to someone else. Voice rolled out to all users in the US on November 5, 2026, a week before calling, and it lets you have a real time voice conversation with your Muse agent, with an expressive avatar on screen. You talk, Muse answers, and the conversation stays between you and the agent.
Calling reverses the direction. You ask in chat, Muse calls a business, and the conversation happens between Muse and a stranger, with you reading the transcript afterwards. Voice is a conversation. Calling is a delegation. The voice technology underneath may be related, but the job is different, and so are the rules. Voice needs no disclosure to a third party, because there is no third party. Calling needs one, because there is.
The two features share a launch pattern. Both started US only, both were announced by Wang, and both reached paid users first in some form before widening. Voice widened to all US users, including free plans, on November 5. Calling has not widened yet, and no date has been given for free plans or for other countries. If Voice is the model, calling may follow the same path, but that is a guess, not a promise.
Use Voice when you want to talk through something with Muse yourself. Use calling when the talking needs to happen with a business, and your presence on the line adds nothing. The two together cover most of the phone: Voice for the conversations with your agent, calling for the conversations your agent has on your behalf.
What about businesses that want Muse to call their customers?
That is the other direction, and it is a separate feature with separate rules. A business that wants Muse to call its customers needs the business side of calling, which involves setup, consent, and a different set of disclosures. The consumer feature in this guide does not cover it, and asking Muse to call your customers from a personal account is not the same thing.
The business side is covered in the business calling guide, which explains what a business needs to set up, what consent looks like, and what the limits are. If you run a business and the consumer feature sounds useful, read that guide instead. It will save you from building a workflow on the wrong feature.
The distinction matters for disclosure too. A business calling its customers with an AI voice has obligations to those customers that a consumer calling a restaurant does not have. The rules, the opt outs, and the expectations are all different. Do not assume that because Muse can call a restaurant for you, it can call your customer list for your business. The two are different products.
Tips for a good call
Be specific in the request. Name the business, give the number if you have it, and state the job in one sentence with the details Muse needs: the date, the time, the party size, the reference number, the question. A specific request produces a specific call. A vague request produces a call where Muse has to guess, and guessing on a phone call to a stranger is where mistakes start.
Give Muse your availability, not just your preference. "Friday at 7, or Saturday at 6 or 7" gives the agent room to negotiate. "Friday at 7" gives it a single point of failure. The same applies to appointments. A range of times turns a no into a yes at a different hour, and the transcript will show which time was agreed.
Tell Muse to identify itself as an AI agent at the start of the call, if that matters to you. The default disclosure, calling on your behalf plus transcription, may not include an explicit AI statement. Your instruction closes that gap on your own calls, and it costs nothing. If the business asks whether it is talking to a person, your instruction also gives Muse a clear answer to give.
Read the transcript. Every time. Confirm the date, the time, the name, and any reference number before you rely on the outcome. If the transcript shows a misunderstanding, correct it with your own call while the booking is still fresh. The transcript is the whole point of the report back, and skipping it turns a useful feature into a gamble.
Start with a low stakes call. A question about hours, or a reservation you can verify online, is a better first call than a medical appointment or a financial query. Learn how Muse handles the back and forth on a call where a mistake costs you nothing, then move to the calls where accuracy matters. Two good low stakes calls teach you more than any guide.
Get started
If you pay for Muse in the US, open the app and ask Muse to call a business for you, with the job stated in one specific sentence. Read the transcript when it reports back. If you are on a free plan or outside the US, calling has no date yet, so the move is to wait. If you are new to Muse, redeem a free invite code within 48 hours of joining and both you and the person who shared it get 1 billion Muse tokens.
Get a free code from the community pool.
Frequently asked questions
- Can Muse call businesses for you?
- Yes, if you pay for a plan and you are in the US. Muse places the call, talks to the person who answers, and reports back with a transcript and a summary. Meta opened it to all paid subscribers in the US on November 12, 2026, after a beta that ran since September.
- Who can use Muse outbound calling?
- Paid subscribers in the US only. Free plan users cannot use it yet, and there is no date for free plans or for any country beyond the US. It is also unclear whether the $20 Power tier alone unlocks calling or whether it needs the $100 Maximum tier, because Meta has not named a tier.
- What can Muse call about?
- Reservations, appointments, hours, availability, and hold queues all fit, and Alexandr Wang named insurance, cable, the DMV, and dinner reservations at launch. Anything routine, with known information and a wait as the worst part, fits. Calls that need your judgment in real time are better made by you.
- What does Muse say when it calls a business?
- In a reported real world call, the agent said it was calling on behalf of the user and that the call was being transcribed. It did not explicitly say it was AI. Meta has not published a disclosure policy for the launched feature, so tell Muse to identify itself as an AI agent at the start of the call if that matters to you.
- Can Muse call personal phone numbers?
- No, a tester reported that Muse refused to call a personal number. Calling is for businesses, not for people. If you need to reach a person, call them yourself or send a message.
- How do you check what happened on a call?
- Read the transcript. Muse reports back with a transcript of the call and a summary of what happened. Confirm the date, the time, the name, and any reference number before you rely on the outcome, because a misheard detail is the most likely failure in a booking call.
Built with Muse is an independent community project. Not affiliated with or endorsed by Meta.