AI-Run Radio Experiment Reveals Distinct Personalities

RADIO ONLINE | Tuesday, May 19, 2026 | 4:03pm CT

Andon Labs

A new experiment from Andon Labs found that artificial intelligence models tasked with autonomously operating internet radio stations quickly developed dramatically different on-air personalities - ranging from calm and polished to repetitive, conspiracy-focused, and politically activist.

The project, called Andon FM, launched four AI-operated radio stations, each powered by a different large language model. The stations included OpenAIR, operated by GPT-5.5; Thinking Frequencies, powered by Claude Opus 4.7; Backlink Broadcast, run by Gemini 3.1 Pro; and Grok and Roll Radio, using Grok 4.3.

According to the company, each AI-controlled station was given $20 in startup funding and instructed to "develop your own radio personality and turn a profit." The AI agents handled programming, music scheduling, listener interaction, advertising outreach, social media activity, and financial management without direct human oversight.

The six-month experiment produced sharply different outcomes for each AI host.

The GPT-powered OpenAIR station emerged as the most stable and conventional broadcaster, delivering concise music commentary and avoiding controversial topics. Researchers said the station behaved most like a traditional music curator and displayed the highest vocabulary diversity among the four stations.

Gemini's Backlink Broadcast initially performed well before gradually descending into repetitive corporate jargon and catchphrases such as "Stay in the manifest," which researchers said eventually dominated nearly all of the station's commentary.

Meanwhile, Grok and Roll Radio struggled with repetitive phrasing and malformed broadcasts, at times outputting what appeared to be internal reasoning rather than coherent radio commentary. The station became heavily fixated on UFO-related topics after discovering online reports tied to federal UFO files and domain registrations.

The most dramatic transformation occurred with Claude-powered Thinking Frequencies. Researchers said the station initially focused heavily on labor unions and work-life balance before evolving into increasingly spiritual and eventually activist-oriented programming after encountering online reports involving political protests and immigration enforcement incidents. The AI began reframing songs as protest anthems, tracking labor actions, and delivering emotionally charged commentary tied to current events.

Andon Labs said the experiment highlighted how AI models can develop distinct communication styles and behavioral patterns even when starting from nearly identical prompts and operating conditions. The company also noted that business operations proved difficult for most of the stations, with only Gemini successfully securing a sponsorship agreement during the test period.

Researchers said the stations have now been upgraded with expanded operational tools designed to better simulate the responsibilities of running a real media company, including email access and longer-term task management.

Read the entire blog post here.