Muse AI Calls: Meta’s Paid Changes (2026) ⚡
Discover Meta Muse AI calls for paid subscribers—costs, real limits of AI phone agents, AI customer service tips, and AI ethics to consider in 2026.

Meta's Muse Can Call Businesses. The Hard Part Is Getting Them to Answer
Calling a cable company, a government office or a restaurant can take longer than the actual problem. You work your way through the phone tree, wait on hold, and hope whoever picks up can fix what you need. Meta's Muse can now make outbound calls to U.S. businesses on a user's behalf - so you don't have to do it yourself.
This isn't just for a small group of early testers anymore. Meta says the feature is available to anyone who pays for Muse. In other words, you're not only getting a consumer AI that talks - you're getting one that tries to act for you.
There are still big limits. It's a paid feature, it's currently for calls to U.S. businesses, and some companies may hang up as soon as they realize the caller is an AI. Even with better voice tech, that doesn't automatically make the calls welcome, trustworthy, or effective. This article breaks down what Muse offers on the phone, why Meta is charging for it, where these calls tend to fall apart, and how to decide if delegating a call is worth it.
What can Meta's Muse do on a phone call?
Muse can place outbound calls to U.S. businesses on a user's behalf. Meta describes the feature as a way to handle calls people would rather avoid - things like contacting a cable company, the DMV, or a restaurant that doesn't accept reservations online. The basic idea is simple: you ask the assistant to take over the phone call that you'd normally have to make.
The more important distinction is between an AI that gives you information and an AI that takes action. A chatbot can explain how to contact a business. A phone agent can try to make that contact. That shift is exactly what makes this useful - and what raises the bar for what "success" looks like.
A good answer can be checked quickly. A good call has to do more than sound smooth: it needs to reach the right person, explain the request clearly, handle what the other side says, and leave you with an outcome you can actually use.
Meta first tested the calling feature with selected early adopters in September. Since then, access has expanded to paid Muse subscribers. Meta says it had more than 10 million users the previous month, but that number doesn't tell you how many people use the calling feature or how often they try it.
The task is more than speaking
A phone call isn't just a voice interface. The caller has to make decisions in real time. They may need to explain why they're calling, answer questions, clarify details, handle transfers, or deal with follow-ups. A restaurant reservation is comparatively self-contained. A disputed cable bill or a complicated government issue often requires personal information, context the caller can't reliably guess, and additional steps after the first conversation.
Meta's announcement doesn't lay out the exact types of call outcomes Muse can complete, what details users must provide, or how it handles every kind of follow-up. Those specifics matter. A call that connects but doesn't actually resolve the task still isn't a win just because the agent spoke clearly.
For now, the cleanest way to think about Muse's calling feature is as a delegation option, not a promise that it can finish every phone task end to end. Whether it helps will depend on the request, the business's willingness to engage, and how clearly the user defines what they want.
Why is AI calling a paid feature?
Meta's Muse subscription plans start at $20 per month for 500 million Muse tokens per week. A higher-use plan costs $100 per month and includes 3 billion tokens per week. Meta is selling business calling as part of the paid Muse experience, not as a free add-on.
This pricing also doesn't answer some practical questions. Meta doesn't say how many tokens a call uses in real life, whether calls have a separate meter, or what a "typical" subscriber can reasonably expect to do each month. Those are details people should check inside the product, not infer from the headline plan limits.
For Meta, charging for calling also tests whether consumers will pay for an AI that performs tasks, not just one that generates text or answers questions. Calling is an easy use case to understand - most people have felt the pain of waiting on hold. But understanding the problem doesn't automatically translate into real willingness to pay. Subscribers will decide based on whether it saves them enough time and effort, especially if they only need phone calls occasionally.
A subscription changes the product question
It's not enough to ask, "Can Muse make a call?" The better question is whether the entire service is worth it - measured in money, time, and attention - compared with making the call yourself. A feature can be technically impressive and still be a bad deal if it struggles with common requests or requires lots of oversight.
Before paying extra specifically for calling, it helps to ask:
- Frequency: Do you make these calls often, or just once in a while?
- Task value: Is the call repetitive and low-stakes, or does it affect money, health or legal obligations?
- Outcome quality: Can you verify what changed (or what was agreed) after the call?
- Plan fit: Does Muse make sense for your other uses, beyond calling?
There's also a product signal here. Meta says creating a Muse email address is expected to roll out to free users, while calls remain paid for now. That implies Meta isn't treating every capability as automatically free. The company is still sorting out which actions belong in the paid tier, even if it doesn't fully explain where that line will land later.
What can go wrong when an AI calls a business?
The biggest obstacle isn't whether the AI can understand speech. The other side has to accept the call and actually be willing to help. Some businesses and receptionists hang up once they realize the caller is an AI. When that happens, the agent's fluency doesn't fix the problem - because the call fails before it reaches the right person.
That means this isn't a one-way system. You delegate to Muse, but the business still controls its processes, staff availability, and expectations. A phone agent can only operate within the channel the business allows. The company may require identity verification, refuse to discuss an account with a third party, or ask questions the agent can't answer because the user didn't provide the needed information.
These aren't rare edge cases for customer service. They're part of how businesses manage access and responsibility.
Meta isn't the only company exploring this space. The source report also points to Google's Gemini and startup Instinct as offering similar calling capabilities. That suggests more than one team is working on AI phone agents, but it doesn't prove all products behave the same way - or succeed at the same tasks. The real operational question is whether an agent can reliably complete requests across different businesses, different scripts, and different call flows.
Better voice creates a trust problem
As AI voices get more convincing, people on the other end might find it harder to tell whether they're speaking with a real customer or an AI agent acting on someone's behalf. That could make conversations feel smoother - but "smooth" isn't the same as "transparent."
If a business believes it's talking to the customer directly, while it isn't, the interaction can create confusion about who made commitments or provided information. A responsible product should be clear about what the agent can and can't do, and it should avoid nudging users toward misrepresentation.
Businesses also need a straightforward way to decide whether they'll engage with automated callers. Meta's announcement doesn't spell out how Muse handles disclosure, consent, recordings, or sensitive information - so users shouldn't assume details that weren't stated.
The stakes matter, too. Asking whether a restaurant has a table available is different from changing a service contract, discussing a medical issue, or disputing a charge. On higher-impact calls, a polished voice can make a weak or misunderstood result sound more authoritative than it should. A good safeguard is treating what the agent reports as something to verify, not something to accept blindly.
When should you delegate a call to AI?
A reasonable rule of thumb: delegate calls that are repetitive, low-stakes, and easy to verify. If the request has a clear outcome and an error would be easy to correct, an agent may save you time. If the call involves sensitive information, major financial decisions, or a complicated dispute, you're usually better off staying in control.
When you think about it, picture the call in three parts: what the agent needs to know, what it's allowed to do, and how you'll confirm the result. That bit of preparation reduces the odds that the assistant will run with an ambiguous instruction - or hand back a vague summary that leaves you guessing what actually happened.
A practical checklist before handing over
- Define one outcome. Ask for a specific result (like checking whether a reservation is available) instead of a broad request to "handle the issue."
- Limit the scope. Be explicit about what the agent may ask for and what it may agree to. Don't assume it can infer your authority or preferences.
- Share only necessary information. Give the minimum details needed for the task, especially when account, identity or payment info is involved.
- Set a stopping point. If the business asks for a decision or information you didn't authorize, the agent shouldn't guess. You may need to take over directly.
- Verify the result. Confirm any booking, account change, or commitment through a channel you trust. A fluent recap isn't a substitute for confirmation when the consequences matter.
Those steps help whether the assistant is making the call or just preparing you for it. They also highlight where Muse needs clarity: what information it receives, what authority users give it, and what counts as a completed task.
The economics matter, too. If a call takes you five minutes and you only make one every few months, a subscription probably won't pay off based on calling alone. If you routinely spend time on repetitive customer-service issues - and you already use Muse for other tasks - the calling feature may feel more worthwhile. The right comparison is your real routine, not the broad promise that an AI can do everything.
Conclusion
Meta's business-calling feature is a step toward consumer AI that tries to take action, not just return answers. For routine calls, the upside is clear: less time spent on phone trees and waiting for help. But actual usefulness depends on whether businesses engage, whether the agent handles the request accurately, and whether you can verify what happened.
Treat Muse calling as a tool for bounded, low-risk tasks - not as a replacement for your judgment in sensitive conversations. Before paying for it, think about how often you'll use it and whether Muse is valuable to you beyond calling. The technology is moving forward, but building trust on both sides of the line is the harder part.