Koral announced the plan on Thursday in Istanbul, saying an AI version of his voice and face would handle common questions for 11 municipal call centres. The pilot is expected to reach as many as 2.4 million residents, and the start-up behind it says it has raised 480 million lira (about €12 million) to build and run the service.
The former national football captain said callers could ask about waste collection, transport cards and local appointments, then receive spoken answers in Turkish or Arabic. The municipalities have not yet published the service’s test results, data-retention rules or a list of the questions the avatar will be allowed to answer.
A familiar voice, new risks
“People hear a famous, reassuring voice and lower their guard. That is exactly when a mistaken answer can travel furthest,” said Peetu Yrjölä-Aarnio, a researcher at Istanbul’s Centre for Digital Public Life. She warned that residents may mistake polished AI responses for official decisions, especially when the system speaks with confidence.
Koral said the avatar would not make decisions or collect payment details. “It will point people to the right desk; it will not replace the person who has authority to say yes or no,” he said. The start-up says calls will be labelled as AI-generated, though it has not explained how that warning will work for people who join after a call has begun.
A famous voice can make a weak answer sound like an order from the state.
Peetu Yrjölä-Aarnio, digital policy researcher
That distinction matters in Turkey, where municipal hotlines handle practical requests that can affect a household’s daily routine. A system that gives the wrong date for a service, or sends someone to the wrong office, can waste hours even without exposing sensitive information. Alexander Peña, a telecom-security specialist, said callers need a simple way to reach a human operator.
Who controls the voice
Koral’s announcement follows a broader rush to package public figures’ voices as software. The start-up says it trained the avatar on recordings supplied by Koral and will delete raw audio after processing, but it has not said whether outside auditors can verify that promise or how quickly the synthetic voice could be switched off.
Peña said the key question was not whether the software could imitate Koral, but who could authorise new scripts and correct errors. “A voice model needs a brake that works faster than a public-relations team,” he said. He wants every answer to carry a traceable source and every call to offer an immediate transfer to a live operator.
The pilot’s first phase is due to begin in November, with a review after 90 days. The municipalities have not named the participating districts, and none has said whether residents can opt out of automated calls. If the rollout expands, the test will be whether people can tell when a celebrity voice is offering guidance—and when only a public office can make a binding decision.
