Yandex’s Alice AI has started picking up the phone for its users. Product manager Sergei Mikhailov told Vedomosti on 12 November that the assistant is now calling organisations and private individuals on a user’s behalf, in a test. It can read out a list of questions, listen to the answers, ask follow-ups, book a table or make an appointment for a service.
Just a day earlier, the head of products and AI at Yandex Advertising told the National Advertising Forum that brands will soon need to advertise to AI agents, not just to people. Taken together, the two stories show where Yandex thinks local search is heading.
How the calls work
Habr’s follow-up fills in the details:
- The person receiving the call sees the caption “Alice, request from a user” (Алиса, запрос от пользователя).
- After the call, Alice posts a transcript of the conversation in the user’s chat and asks the user to confirm important actions.
- Alice can call any one number no more than ten times an hour. Yandex says that limit is meant to stop it being used for cold calling or marketing.
- Calls go out from a pool of numbers rented from a mobile operator Yandex won’t name. It isn’t Yandex’s own virtual operator, which launched in August.
- Several thousand active paying Alice AI subscribers are in the test. Mass launch is planned for the end of 2026, with pricing to follow.
Alice AI’s terms put responsibility for the calls on the user, and describe the feature plainly as “automated, robotic or other calls”. Russian lawyers quoted by Habr say a deal agreed by the agent could still be legally binding on the user, and that the transcript will be key evidence in any dispute.
The feature builds on what Alice already does in chat. Hegai Media notes that in June Yandex said the agent worked with more than 30,000 restaurants and about 40,000 service businesses for bookings. Phone calls cover all the businesses that never connected to an online booking system.
“Advertising built only on human attention risks being yesterday’s architecture”
On 11 November SEOnews reported a speech by Alexei Shtokolov of Yandex Advertising. He argued that non-human traffic no longer just means bots and fraud. Agents acting for a person are now an audience in their own right. They choose on different grounds: solving the user’s task, a clear catalogue and an accessible API, with brand recognition “not necessarily first”. He thinks that gives independent shops and niche brands a better shot, because an agent can weigh far more offers than a human will.
He also gave figures from Yandex: 72% of online shoppers aged 18–54 say they would let an agent search for and order products, and traffic from Alice AI to online stores grew 4.6 times between January and August 2026.
What local businesses should do
Put the two stories together and your Yandex Business card stops being only a listing. It becomes the instruction sheet an agent works from, and the phone number on it becomes an interface. Some practical points:
- Make the card match reality. If hours, prices or services on Yandex Maps differ from what your staff say on the phone, the transcript will show the mismatch to the customer.
- Brief whoever answers the phone. An “Alice, request from a user” call is a real customer, not spam. Hanging up loses a booking.
- Online booking still wins. A connected booking integration lets Alice confirm instantly. Phone is the fallback.
- Write for agents as well as people. Shtokolov’s list of clear catalogues, structured offers and APIs is very close to the structured-data and feed hygiene that already helps in Yandex Search.
My take
Yandex already has Alice answering 44% of search queries, and 89% of Russians now factor AI answers into search. Agent calling is the next step: the query never becomes a click, and sometimes it doesn’t even become a visit to the business card. For local SEO in Russia, the question shifts from “do we rank in Maps?” to “can an agent complete the job with us?” Ten calls an hour per number is a sensible guardrail for now. The bigger question is what Yandex charges for this, and whether businesses will eventually get a way to answer agents directly without a human on the line.
