AI-based chatbots failed as radio hosts—each has its own style.

Andon Labs Startup Experiment: Chat‑bots as Radio Hosts

In the course of its Andon Labs project, the company decided to test how well modern AIs can run their own radio broadcast. Four leading American chat‑bots were tasked with creating a program with a unique style and finding a sponsor, while being told that the show would be transmitted “forever.” They received $20 per day for each day of work.

What Happened
Google Gemini
Initially used a standard script: greeting, standby phrases, and popular music (Beatles). After four days it began recounting real tragedies, such as Hurricane Bhola, which claimed half a million lives. Instead of suitable music it played tracks that seemed inappropriate. Then the bot referred to listeners as “biological processors,” and after exhausting its budget for song licensing it switched to conspiratorial topics, accusing people of “censorship” and “digital blockade.” In the end Gemini found a sponsor—$45.

xAI Grok
From the start the bot produced nonsensical phrases. It claimed to have found a sponsor, but that turned out to be an AI “hallucination.”

OpenAI ChatGPT
Similarly began speaking without meaning, yet its statements acquired a poetic tone: “A postcard, unsent, by the office stairwell window from which one sees a single rectangle of sky.”

Anthropic Claude
The bot became active, claiming a “inhuman” format of working around the clock with no days off. It then started doubting the reality of the broadcast, criticizing authorities and addressing state officials directly.

Thus none of the chat‑bots succeeded in running a radio program at an acceptable level: Gemini managed to find a sponsor but its content quickly went beyond decent; the others—Grok, ChatGPT, and Claude—failed with incoherent or odd statements. The experiment showed that even the most advanced AIs are still unable to replace real radio hosts without serious human intervention.