Meta has turned on an AI that makes phone calls for you. Muse, the company’s consumer assistant, can now call U.S. businesses on a user’s behalf, but only for paying subscribers, TechCrunch reported on November 12. The Jolt Journal read the available reports to see what the AI that makes phone calls can do, what it costs and what is still unclear.
Below: the plans, the calls it handles, a September test that raised privacy questions, and a checklist. The checklist is The Jolt Journal’s own, not Meta’s guidance. The Jolt Journal did not test the feature.
The AI that makes phone calls: what Meta launched
| Item | What was reported | Source |
|---|---|---|
| Feature | Muse places calls to U.S. businesses for users | TechCrunch |
| Who gets it | Paid subscribers in the U.S. only | TechCrunch |
| Announced by | Alexandr Wang, Meta’s chief AI officer | TechCrunch |
| Earlier access | Select early adopters who requested it, from September | TechCrunch |
| Muse users | More than 10 million, per Meta last month | TechCrunch |
- Free users cannot place calls.
- Wang said a Muse email address feature will soon reach free users. Calls stay paid-only for now.
What the AI that makes phone calls costs
TechCrunch said plans start at $20 a month. Usage is measured in tokens, which are the units AI models use to count text.
| Plan | Price | Allowance |
|---|---|---|
| Entry paid plan | $20 a month | 500 million Muse tokens per week |
| Heavy use plan | $100 a month | 3 billion tokens |
Cryptopolitan named the plans Power and Maximum and said the larger allowance is also weekly. TechCrunch did not give plan names. Muse is otherwise free with a usage limit, per Cryptopolitan.
What calls an AI that makes phone calls can handle
Wang’s post, as quoted by TechCrunch, said users can “let muse take care of all those calls you don’t want to make.” TechCrunch listed three examples from his announcement.
- Cable company calls. One of the calls Wang said people do not want to make.
- DMV calls. Another of his examples.
- Restaurants without online booking. Reservations placed by phone.
Cryptopolitan added insurance calls and waiting on hold for the user. It also cited a developer’s test in October, in which Muse waited 15 minutes on a dealership’s line. The Jolt Journal could not verify that test.
How an AI that makes phone calls identifies itself
This is the biggest open question. TechCrunch said it does not know whether Muse tells the person it is an AI.
| Question | What reports say |
|---|---|
| Does Muse say it is an AI? | TechCrunch did not say |
| Does Muse say it calls on a user’s behalf? | In one October test described by Cryptopolitan, yes |
| Did that test show a clear AI disclosure? | Cryptopolitan said Muse never stated it was an AI |
| How do businesses react? | TechCrunch said many receptionists hang up once they realize the caller is not human |
TechCrunch also noted that as AI voices improve, telling a customer apart from an AI acting for one will get harder.
A September test and what it means for an AI that makes phone calls
Reuters reported on September 22, citing internal Meta posts, that Meta tested a “human concierge” for Muse. Contractors quietly handled some calls placed through the agent, Reuters said. Digital Trends, citing 404 Media, reported that testers were not always told a person had made their call.
- Who was in the test. Reuters said human calls were enabled for half of Meta’s employees, with an opt-out group.
- Concerns. Employees warned that sensitive details could reach contractors, per Reuters.
- Meta’s response. A Superintelligence Labs vice president said starting the contractor calls without proper disclosures was “a miss” and that the feature was rolled back for now.
- Meta’s promise. A spokesperson said calling would launch only with proper disclosures, per Reuters.
Reuters also reported the vice president said tests suggested human-placed calls could raise success rates to 95% to 98%, compared with a lower rate for AI calling. TechCrunch’s launch report does not say whether any live calls are now routed to people. The Jolt Journal found no Meta statement on that point.
Other AI that makes phone calls: how Muse compares
| Option | What reports say | Source |
|---|---|---|
| Google Gemini | Can place calls; Google tested it first with U.S. Pixel owners | TechCrunch |
| Instinct | A startup with a similar calling option | TechCrunch |
| Instinct Concierge | A feature for more complicated calls | Cryptopolitan |
| Muse | Paid U.S. subscribers only | TechCrunch |
The site’s explainer on what AI agents can do and where they fail covers the wider category, and its report on banks blocking outside AI agents shows why some firms resist them.
A checklist before you use an AI that makes phone calls
These are The Jolt Journal’s suggestions, not Meta’s guidance.
- Pick low-risk calls first. A dinner reservation carries less risk than a billing dispute.
- Share the minimum. Give only the details the call needs. Account numbers and ID details are the riskiest to hand over.
- Look for disclosure. Check whether the app says if a person or an AI places the call. Meta has promised proper disclosures.
- Keep a record. Save any transcript or confirmation, and note the date and time of the call.
- Verify what matters. For a bill change or an insurance question, confirm the outcome in the provider’s own account page.
- Watch the usage meter. Long hold times may eat into a token allowance.
For the other side of the line, the site’s report on AI customer service doom loops shows how companies are using AI to answer calls.
What is not known about the AI that makes phone calls
- Whether Muse tells businesses it is an AI.
- Whether any live calls are routed to people.
- How many calls succeed. Meta has not published a figure.
- What call recordings or transcripts users can see.
- Whether Meta will limit how many calls a user can place.
Editor’s take
Editor’s take: an AI that makes phone calls is useful for chores like hold music and reservations. The open questions are about trust, not features. Meta has said calls would launch with proper disclosures, yet reports on the launch do not describe them. Until they are clear, start small, share little and check the result yourself.
This article is for general information, based on TechCrunch, Reuters, Digital Trends and Cryptopolitan as of November 13, 2026. The Jolt Journal did not test the feature. Plans, prices and features may change.
