Top 5 in AI

Guides

Can Muse Make Phone Calls for You — and Is a Human Actually on the Line?

By the Top5Apps editorial team · Published September 23, 2026 · Updated September 23, 2026 · 7 min read

Share

Short answer: yes — in a US-only beta that's off until you ask for it. Tell Muse to call a business and, if you're in, it dials, holds the conversation, and reports back. It calls US businesses: the salon, the store that might have the thing in stock, the restaurant with no OpenTable page. It won't call your mom — testers report it refuses personal numbers — and there's no sign of international calling. And the honest asterisk: on September 22 Reuters reported that Meta had quietly tested having trained human contractors place some of those calls when the AI kept failing, without telling the people whose calls they were. Meta says it 'rolled back' that test. Everything below separates what Meta has confirmed from what people assume.

One thing to know before anything else: there is no official announcement, help-center article, or @Muse post about calling. We checked Meta's launch post and all fourteen Muse help articles — zero mentions of phone calls. The public record is two X posts from Meta staff, TechCrunch, Reuters, and 404 Media. Where a fact comes from a user or an internal Meta post rather than Meta itself, we say so.

How to get it: ask, and wait

Ryan Fox@wailord

we just expanded the @Muse beta for outbound calls to US businesses, prioritizing folks who'd asked their Muse to let us know they wanted it first :) if you're in, give it a shot and tell us what you think. phone calling was one of our top requests and your feedback helped make it happen! 📞

View original post ↗

That September 16 post from Muse team member Ryan Fox is the closest thing to an announcement. The mechanism is exactly what it sounds like: ask your Muse to make a call. If you're in the beta, it dials. If you're not, it tells you it can't yet — and that refusal is your request; Meta is 'prioritizing folks who'd asked their Muse to let us know they wanted it first.' TechCrunch called the design 'essentially… a feature request mechanism.' There's no toggle, no waitlist page, and no promise that asking gets you in. Per Reuters' reporting on internal posts, employees have had the feature since August and Meta 'gradually started rolling it out to Muse users in the days after the app's debut.'

What it will and won't call

  • Will: US businesses. That's Meta's own phrase. The use cases people report are the ones you'd expect — a haircut, a stock check, a contractor quote, a restaurant that doesn't take online bookings. One Meta employee had his Muse call five Tesla locations at once to find a Model Y on the floor: 'It found one.'
  • Won't: personal numbers. User report: a tester who asked Muse to call his own phone was refused, with Muse saying the test could dial verified US business numbers rather than personal ones. Meta hasn't published that rule; treat it as how the beta behaves, not policy.
  • Not found: international calls, calls that accept payment, or any way for Muse to receive calls.
  • What you get back: per an internal Meta post quoted by 404 Media, Muse 'handles the conversation, completes your request, and reports back with a transcript and a summary.' Meta hasn't said this publicly. One employee posted that 'if your muse needs you they will patch you in.'

The legitimate question: who's actually talking?

Here's what Reuters' Katie Paul reported on September 22 from internal Meta posts. Businesses were hanging up — an employee wrote that they 'keep hanging up on Muse when they hear it is AI.' So last week Meta enabled a 'human concierge,' also called 'human agent calls,' for half its employees: 'Muse is now able to hand requests to a trained agent, who places the call and works it through.' Contractors could push success rates 'up to the 95% to 98% range' versus the 'lower percentage of AI calling.' Then it went wrong in the ways you'd predict: per 404 Media, one tester 'was not made aware of the caller being human and was only told after,' and per Reuters an employee who had Muse negotiate a cable bill found the transcript showed 'the human contractor had made a racist reference during the call.'

What Meta confirmed, on the record: a Meta Superintelligence Labs VP wrote internally that starting contractor calls 'without proper disclosures' was 'a miss,' and that the company had 'rolled back this feature' while it keeps 'hillclimbing AI calling.' Spokesperson Daniel Roberts told Reuters: 'We're working with merchants to continue improving this potential calling feature, and will only roll it out when it's ready and with the proper disclosures.' Note the word potential — Meta is describing the calling feature you may already have as pre-launch.

What people assume, and shouldn't: that every Muse call is the model talking, and that it introduces itself as an AI. The employee's 'when they hear it is AI' line implies disclosure. But the only published Muse call transcript — screenshotted by a Meta employee on September 16 — opens: 'Hi, this is Hailey calling on Matt's behalf. I'll be transcribing this call for notes.' A human name, no mention of AI, and a recording notice rather than a consent request. Whether 'Hailey' was a model with a persona or a person, we can't tell you, and neither can Meta's public documentation. That's the gap this feature has to close before it's ready.

What goes wrong

  • Businesses hang up. Meta's own testers say so; a 404 Media source described a rep who 'hung up on the agent.' Expect transcripts of calls that went nowhere.
  • Nobody's documented the consent rules. Several US states require all parties to consent to a recording. Muse 'transcribing this call for notes' is a notice, not consent, and Meta has said nothing about recording law or AI-disclosure law. Until it does, assume the business on the other end wasn't asked.
  • You own what gets promised. If your agent — or a contractor speaking for it — agrees to a price, a time, or a cancellation fee, that's your agreement. The same rule that makes Link purchases yours applies to a verbal one.
  • No off switch is documented. There's no help-center article on stopping a call in progress, what's stored, or how to delete a recording. Ask Muse to forget it, or use Reset Muse.

Muse vs Instinct Concierge vs Grok Bot: same job, three products

Noah Shinn@noahrshinn

Your Instinct can now handle phone calls. Introducing Instinct Concierge – a white glove service meant to handle high-touch cases, such as making phone calls, high-end service booking, and more… We're slowly rolling this out to our early access group today and will be expanding access soon.

View original post ↗

Meta MuseInstinct ConciergeGrok BotGemini
Makes calls to businesses?Yes — US beta since Sept 16, unlocked by askingYes — early-access group since Sept 16No — nothing in xAI's docsNo — it dials your phone's app for you to talk
Who's talking?AI; a human-contractor layer was tested internally and 'rolled back'Not stated. 'White glove service' — Instinct hasn't said whether humans are involvedYou
Discloses it's an AI?Undocumented; the one public transcript uses a human nameUndocumented
Official documentationNoneFounder's post onlyHelp article
Sources: Meta staff posts, Reuters (Sept 22), Instinct founder post (Sept 16), docs.x.ai/grok-bot, Google's Gemini help center. Both agent products launched calling the same day; both are early.

Resist the framing that one of these is 'real' and the other isn't. Instinct branded Concierge as white-glove; Muse branded calling as a beta you unlock by asking. Neither has published who's on the line or how the call is disclosed — Instinct's post doesn't say, and TechCrunch lists humans-in-the-loop for Concierge as 'not specified.' The difference that matters today is access: Muse's beta is reachable by anyone with the app who asks; Instinct requires an invite and then early-access status on top. Google's comparable product isn't Gemini Spark at all — it's Search's 'Ask for Me,' which has been calling businesses for quotes since 2025 and does disclose itself.

Should you use it?

  • Good for: the reservation with no web form; 'does anyone nearby have this in stock'; a first-pass contractor quote; getting on a cancellation list. Low stakes, public information, easy to verify afterward.
  • Bad for: insurance, anything involving an account number or a date of birth, a medical office, a bill negotiation, or any conversation you wouldn't want a contractor in a call center to hear. Until Meta documents who places the call and how it's disclosed, assume a third party might.
  • Before every call: read what Muse is going to say — it will confirm the goal — and set an explicit limit ('don't agree to anything over $150'). Afterward, read the transcript, because the transcript is the only record you'll get.

Our read

Calling is the feature that makes a personal agent feel like the thing science fiction promised, and it's also where the 'agent' abstraction is thinnest — because the other party is a person, and they get a say. Meta's testers discovered that people hang up on robots, reached for the oldest fix in tech (hide a human behind the software), and got caught by its own employees within a week. The rollback was the right call, and Meta's statement — 'only when it's ready and with the proper disclosures' — is the right standard. Hold them to it: the day Muse tells the business it's an AI at the start of every call, and tells you whether a person ever joins one, this becomes the best reason to install it. Until then, use it for the salon, not the insurer. Our Muse review and Muse vs Instinct vs Grok Bot guide both reflect this.