For years the answer to "can my AI assistant make phone calls" was a flat no. That changed quietly, and then it changed loudly: on September 4, 2026, Google began switching off Google Assistant on Android phones, tablets, Wear OS watches, headphones and Android Auto, handing assistant duties to Gemini. Gemini can dial. So the honest 2026 answer to the AI assistant phone call question is no longer "no". It is "some of them, some of the time, for some of the call".
That distinction is the whole story, and it is where most people get burned. Dialing a number is the easy five seconds. The painful part of a phone call is the forty minutes after the number connects: the menu tree, the hold music, the transfer, the second hold, the human who asks for an account number, the callback that never comes. Almost no mainstream AI assistant does that part.
Quick Answer: Can AI Make Phone Calls for You in 2026?
- Gemini on Android: yes, it dials. It can call a contact, a business or a number by voice, including from a locked screen. It places the call and hands you the phone.
- Google's agentic calling: yes, narrowly. Google's "Ask for me" experiment in Search Labs calls local businesses for pricing and availability, US English users only, and it started with categories like nail salons and auto repair shops.
- ChatGPT: no. It has no telephony. It will tell you so directly, and it cannot contact people on your behalf.
- Siri: it dials, nothing more. Voice commands place a call. It does not sit in a queue for you and it does not answer your incoming calls.
- Waiting on hold and beating IVR menus: almost nobody. Pixel's Hold for Me handles toll free numbers on Pixel hardware and then gives the call back to you.
- Full end to end calls: a dedicated AI phone agent such as Assindo makes the call, presses through the menu, waits in the queue, talks to the human, and reports back with a summary and a transcript.
The short version: mainstream AI assistants got the dialer. They did not get the patience.
What Changed on September 4, 2026
Google Assistant, launched in 2016, started its shutdown on September 4, 2026. The rollout takes weeks rather than hours, and Google has been clear that once your device flips, there is no switching back. Assistant survives for now inside cars with Google built in, while Gemini expands the other way into Google TV, Google Home speakers and smart displays.
For phone calls specifically, the swap is a real upgrade and a real regression at the same time.
The upgrade: Gemini understands messy call requests. "Call the vet where we took the dog in June" is a query, not a command, and a language model handles that far better than the old intent matching did. Google's help documentation confirms Gemini can call contacts, businesses and raw numbers, can pick up a business from the context of the conversation you were just having, can call by relationship if you have told it who your sister is, and can dial without unlocking the phone when you allow it.
The regression: Assistant was deterministic. It did a small number of things the same way every time. Gemini is probabilistic, which is wonderful for interpretation and less wonderful for the muscle memory of "call Mum" while you are driving. Several reviewers have flagged exactly this risk with basic phone and smart home commands. Expect a rocky few weeks.
Neither of those changes the ceiling. Gemini starts calls. It does not conduct them for you.
What Gemini Can and Cannot Do on the Phone
It is worth being precise, because "Google's AI makes phone calls" has been reported three different ways in the last eighteen months and they are not the same feature.
Gemini's Phone app integration. This is a dialer. You speak, it places the call, you talk. Requirements are mundane: an Android device with an active mobile plan, contacts synced to your Google account, phone permissions granted for the Google app, a default phone app set. It explicitly cannot accept or reject incoming calls, it is not available in Gemini Live chats, in Gemini inside Google Messages, or on the web app, and emergency calls must go through your device's own dialer.
Agentic calling, or "Ask for me". This is the genuinely autonomous one. Opt in through Search Labs, search something like "oil change near me", fill in a card describing what you want, and Google's AI phones multiple businesses, asks about price and availability, and texts or emails you a summary, typically inside half an hour. It is built on the Duplex line of work that Google demoed back in 2018.
The limits matter as much as the capability. It is US only and English only. It launched around a small set of service categories where the conversation is a clean question and answer, such as nail salons and auto repair shops. It is shopping shaped: it asks businesses for information. It is not built to argue a billing error with your insurer, sit in a thirty five minute queue for a utility, or reschedule a specialist appointment that needs your date of birth and a referral number.
So: Gemini calls a shop to ask a price. Gemini does not call the airline about your rebooked connection.
Why ChatGPT Cannot Make Phone Calls on Your Behalf
This is one of the most searched questions about AI assistants and phone calls, and the answer has not moved: ChatGPT has no phone line. You can talk to it with voice, which is a different thing entirely. Voice mode is a microphone on a chatbot. It does not place a call over the telephone network, it cannot contact a third party, and OpenAI's own guidance says so plainly, which is why the model itself declines when asked.
The reason is not that OpenAI forgot. It is that a phone call is a completely different product to a chat completion:
Telephony is infrastructure, not a feature. Real calls need SIP trunking, PSTN interconnects, carrier relationships, number provisioning and per country telecom compliance. That is a regulated business, not an API endpoint.
Latency is unforgiving. A two second pause in a chat window is invisible. On a call it is an awkward silence that makes the person on the other end say "hello? hello?". A natural sounding voice pipeline needs speech recognition, reasoning and speech synthesis to complete in well under a second, continuously, for the length of the call.
IVR menus are adversarial by design. "Press 1 for billing, press 2 for claims" was built for a human thumb. An AI agent has to hear the options, choose, send the right DTMF tones, sometimes key in a sixteen digit account number, and recognise when it has been dumped back at the root menu five levels deep.
The blast radius is bigger. A wrong restaurant recommendation costs you a mediocre dinner. A wrong sentence to your bank costs you an account problem. Real world actions carry real world consequences, so the acceptable error rate is far lower than for text.
Add disclosure law to that list. Regulators in the EU and several US states now expect an automated caller to identify itself as automated, which is a product requirement, not an afterthought.
Siri, Alexa and the Rest Hit the Same Wall
Siri places calls on command and has done for over a decade. Apple's rebuilt, Apple Intelligence powered Siri has been the most delayed product in consumer AI, and the capabilities that have shipped centre on text, images and on screen context rather than autonomous calling. Siri does not wait in a queue for you. It does not screen your incoming calls, interrogate an unknown caller, or hand you a structured summary afterwards.
Alexa is the same shape: excellent at "call Dad", absent for everything that happens after the line connects.
Pixel owners get the closest thing to relief. Hold for Me keeps the line for you when you call a toll free number, processes the audio on device, and buzzes you when a human picks up. It is genuinely good, and it is genuinely narrow: toll free numbers only, Pixel hardware, a handful of countries including the US, UK, Canada, Australia and Japan. It also gives the call back to you. You still have to be free at an unpredictable moment, and you still have to conduct the conversation.
Call Screen, its sibling, is the incoming half and works the same way: useful, on device, and it stops well short of handling the call.
The Part Nobody Automated: Queues, Menus and Callbacks
Here is the gap that survived all of 2026's assistant launches. Americans spend an estimated 900 million hours a year on hold. That time is not lost to dialing. It is lost to waiting, to menu trees, to being transferred, to being disconnected and starting over, and to businesses that still take appointments by phone and nothing else.
A dialer does not touch any of that. What actually removes those hours is an AI agent that owns the entire call:
- Places the outgoing call and stays on the line however long it takes
- Navigates the phone menu, including entering account and reference numbers
- Waits in the queue without getting bored, distracted or hung up on
- Talks to the human who eventually answers, with the context of why it is calling
- Reports back with a written summary and a transcript, so you know what was actually agreed
- Answers your incoming calls when you cannot, screens the caller, and takes a message
This is where a dedicated AI phone assistant is a different category from a chatbot with a microphone. Assindo is built for exactly this shape of work: you give it a phone number and a goal in the app, it makes a real call, works the menu, holds the line, speaks to the person, and sends you the outcome. It also answers your incoming calls, screens them, and turns what the caller said into a task. It works on iOS, Android and the web, with no server to run and nothing to configure, which is the other reason self hosted agent frameworks never made it into normal people's lives. Our guide to what AI can and cannot do on the phone covers the hold time side in detail.
There is an honest limit worth stating: an AI agent cannot pass a voice identity check, cannot answer security questions only you know, and should not pretend to be you. Calls that hinge on proving who you are still need you on the line. Everything before that point, which is most of the elapsed time, does not.
Four Calls People Still Make Themselves in 2026
The specialist appointment. The clinic's online booking covers routine visits only. Anything else is a phone number, a five option menu, and a receptionist who is on another line. Gemini will dial it. It will not sit through the eleven minute queue, explain the referral, and get you a Tuesday morning slot.
The billing dispute. A charge appeared that should not have. This is a forty minute call with two transfers, a hold, and a supervisor. No mainstream assistant will make it, and the reason is not technical modesty: it is a conversation with an unpredictable script and a real outcome.
The subscription cancellation. Plenty of services still make you phone to cancel, precisely because the friction works. An agent that will hold, decline the three retention offers, and get a cancellation reference is worth more than the subscription. We wrote about the cancel by phone problem separately.
The missed inbound call. You are in a meeting and a client rings. It goes to voicemail, and most people no longer leave voicemails. Nothing Google shipped this year answers your phone for you. That is the incoming half of the same problem, and it is the half small businesses feel hardest, as our small business calling guide goes into.
How to Tell a Chatbot From an AI Agent That Can Call
Five questions that separate the two, and they take a minute to check:
1. Does it dial, or does it conduct? Ask whether it stays on the line after connecting. If the product describes handing the call to you, it is a dialer.
2. Does it handle incoming calls too? Outgoing only means you are still tethered to your phone all day. The full version screens and answers.
3. Does it work on hold and in menus? Ask specifically about IVR navigation, DTMF tones, and queue waits. This is the differentiator, so any product that does it will say so.
4. Does it give you a transcript? A call you cannot review is a call you have to make again. Summary plus transcript is the minimum standard for trusting an agent with anything that matters.
5. Does it require hosting? If setup involves servers, API keys and configuration files, it is a developer tool. Self hosted agent frameworks are powerful and they are not a substitute for something that works from a phone. Our cost comparison breaks down what the managed options actually charge.
AI Assistants and Phone Calls: Quick Answers
Can AI make phone calls for me? Yes, but only some AI can. Gemini on Android places calls, Google's "Ask for me" experiment calls local businesses for prices in the US in English, and dedicated AI phone agents such as Assindo handle the full call including menus and hold time. General chatbots cannot.
Why does my AI assistant say it cannot make phone calls? Because it has no connection to the telephone network. Assistants like ChatGPT run entirely inside their own interface. Placing a call requires telephony infrastructure, carrier agreements and per country compliance that a text model simply does not have, which is why the assistant refuses rather than fails.
Can ChatGPT make phone calls on your behalf? No. Voice mode lets you speak to ChatGPT, not through it. It cannot ring a business, a clinic or a person for you, and OpenAI's documentation states this directly.
Can Gemini call a business for me and ask a question? Only inside Google's agentic calling experiment, which is US English, opt in through Search Labs, and scoped to straightforward question and answer categories such as auto repair and salons. Outside that, Gemini dials and you talk.
Can you call your AI assistant on the phone? With most assistants, no. You talk to them through their app. An AI phone assistant works the other way around: it has its own phone number, so it can both make calls and receive them for you.
How do I place a phone call with an AI assistant? On Android, say "call" plus a name or business to Gemini. To have the AI conduct the call rather than start it, you need an agent with its own number: you describe the goal and the number in the app, and it calls, holds, talks and reports back.
Where This Goes Next
The direction is not in doubt. Assistants are moving from answering questions to completing tasks, and phone calls are the most stubbornly manual task left in an average week. Google shipped the first mainstream consumer agentic calling. Apple is rebuilding Siri around actions. The gap closes eventually.
What is in doubt is the timeline, and the shape. Big platform assistants tend to ship the safe, shopping shaped version of a capability first, because the reputational cost of an AI agent saying the wrong thing to a bank is enormous. The calls that hurt most, the queue, the dispute, the reschedule, are exactly the ones a cautious platform ships last.
So the useful question in 2026 is not whether AI will make phone calls for everyone. It is whether you want to keep making them yourself while you wait.
Stop Waiting for Your Assistant to Learn How to Call
Gemini dials. Assindo makes the whole call: the menu, the hold, the human, and a written summary when it is done.