Newsletters
In the not-so-distant future, having your own personal journalist—one who understands your tone, preferences, and worldview—may no longer be a fantasy. Thanks to advances in artificial intelligence, media is rapidly evolving into a more personalized, character-driven experience.
In Türkiye, one digital platform is pioneering this shift by combining real-time data, AI personas, and lifelike voice synthesis to offer fully automated radio broadcasts without any human hosts.
F.A.S.O.N. Radio, with the efficiency of a platform that seems to have journalists working from various cities, compiles the most significant news on relevant topics and presents it to its listeners. Remarkably, it does all of this using only AI technology.
At the heart of F.A.S.O.N. Radyo is a system built to mimic how humans think, react, and present content. The AI behind the platform doesn’t just read out news headlines. Instead, it gathers articles and social media updates in real time, summarizes them intelligently, and then delivers them through the voice of a character that adds its own tone and angle.
These personas are carefully crafted to sound like real people, with their own ways of speaking, preferences, and ideological leanings. A listener might hear the same news story explained with traditional values by a “Conservative Analyst,” then later revisited with a progressive spin by a “Liberal Commentator.”
What makes this approach more than just an AI gimmick is its underlying critique: the synthetic diversity of these characters draws attention to the predictable, formulaic nature of real-world media voices.
F.A.S.O.N. Radyo broadcasts a wide range of content, including daily news updates, AI-narrated summaries of opinion pieces, and commentary on entertainment and the arts. Each piece is matched with a fitting AI persona, giving listeners the impression of a curated editorial voice. Political updates might come from a formal, no-nonsense character, while cultural stories are presented by a more laid-back, ironic figure.
The head of the initiative, Burak Koy, sat down with Türkiye Today to discuss the future of the initiative and the broader landscape of digital media.
To begin, we’d love to get to know you. How did the idea for F.A.S.O.N. Radio come about? What kind of background does the team behind it bring?
About nine years ago, I moved from Türkiye to the U.K. following a position change at the company I work for. I’m still with the same organization, focusing on digital user experience. I used to follow current affairs back in Türkiye too, but after moving abroad, my interest in Türkiye’s agenda intensified significantly—something I think many expats experience. You inevitably start comparing the media narratives of your home country with those of where you live.
One day, I heard a news story on British radio about doctors leaving the country and the strain it’s putting on the healthcare system. I couldn’t help but wonder: where are they going? That curiosity turned into a small research project.
I figured if this topic intrigued me so much, it might also interest people back in Türkiye—they should hear about it too. But with Türkiye’s news cycle being so relentless, it's not always easy to take a step back and look outward. I felt this was a gap worth addressing.
I knew I wouldn’t have the time for daily content creation, so I designed an automated publishing system powered by AI modules and APIs. My professional background in digital product development and communications strategy—combined with earlier personal content projects, came together in the form of F.A.S.O.N.
I wish I could say it was developed by a team, but this is a solo project built entirely by me. I recently learned that there’s even a term for this kind of one-person tech venture: “solopreneur.” F.A.S.O.N. is essentially the product of technical curiosity meeting personal observation.
The fictional commentators on F.A.S.O.N. sound incredibly realistic — and opinionated. Did you base them on real-life media figures? Or did you use a different method to “teach” AI how to take sides?
There’s definitely inspiration from real life. Most listeners end up picturing certain commentators, and more often than not, they guess correctly.
Technically, training AI to express a particular bias isn’t difficult. You can optimize the modules using specific language patterns, source preferences, and attitudinal cues. But my goal with these personas wasn’t to gain followers or leverage polarization—that space is already crowded.
The opinionated modules on F.A.S.O.N.—Ozlem, Vahit, and to a degree, Stelyo—are a small part of the system. Their purpose is to approach today’s media climate with a touch of distance, even satire. They’re designed to show how media language works, to trigger a “seriously?” reaction, and encourage critical thinking. Sometimes humor reveals reality better than seriousness ever could.
In your view, how might AI-powered media production become the new norm? What trends might drive this experimental approach to scale, both in Türkiye and globally?
Media—and content creation more broadly—is a natural space for AI, especially since what we call AI today is, at its core, a language model. It learns from human-produced texts and uses that data to create human-like, often creative content.
We’re already seeing it transform many content-related tasks, especially operational ones like editing, summarizing, and formatting. That trend will likely accelerate.
But when it comes to truly original content—framing topics from unique angles, building context, and engaging in creative thinking — AI still has its limits. Some cutting-edge models are pushing these boundaries, but they’re not yet at a level to fully replace human creativity.
Journalism, in particular, is a distinct craft. Being on the ground, observing, asking the right questions—these are all skills today’s AI simply doesn’t possess. So for now, AI media systems function more like curators and editors. Still, that alone represents a major shift for many countries and newsrooms.
Technically speaking, how do you build commentator modules that reflect different political leanings — say, “conservative” or “oppositional”? What metrics, data sources, or filtering methods are involved?
These modules are built around three key elements: tone of voice, source prioritization, and conceptual framing. An “oppositional” persona, for instance, might pull news from diverse sources, use a more skeptical tone, ask different kinds of questions, and frame issues with different metaphors. A “conservative” module may refer to the same facts but interpret or prioritize them differently. Think of any recent event in Türkiye and how it was covered by different media outlets—these AI models do something similar.
Technically, this is achieved through prompt engineering, content direction systems, and fine-tuning if needed. The systems are trained using filters based on content reliability, agenda prominence, and sentiment metrics.
So, we’re not making the AI biased per se — we’re teaching it to emulate different interpretive lenses.
We often say “a human journalist senses the news.” So how does F.A.S.O.N.’s AI system determine what’s important? How does your agenda-setting algorithm work?
F.A.S.O.N. doesn’t rely solely on click rates or trending tags. It looks at patterns like topic recurrence, presence across diverse outlets, and conceptual density.
For instance, if a topic appears consistently across multiple news sources in one day, the system prioritizes it. If it's covered by media outlets from various ideological backgrounds, it’s flagged as part of the “general agenda.”
From this, the system creates an artificial “agenda heat map.”
Much like editorial meetings at newsrooms, the AI assesses which stories should take center stage. The commentator modules then respond to this map from their respective viewpoints, offering varied interpretations of the same story. In that sense, the system mimics how a journalist evaluates what’s newsworthy—but does so structurally.
Are there similar AI-driven media projects abroad with the scope or vision of F.A.S.O.N.? Where would you place your project in this landscape?
We’ve seen examples like Bloomberg’s automated news systems or BuzzFeed’s AI content experiments. But most of those focus on formatting or content recommendations. F.A.S.O.N. is more experimental in that it goes into editorial tone and character creation. I’d say it sits at the creative automation frontier of media production—and I hope it inspires new directions in the field.
Do you think we’ll eventually have personalized AI journalists — commentators who know our interests and speak directly to us? How close are we to that technically?
That’s one of the core ideas we’re exploring — personalized content production. The goal isn’t just to show users what they’re interested in, but to curate and narrate the news for them in a way that feels individually relevant. Technically, we’re already close.
Previously, we followed accounts to curate our social media feeds. Now, algorithms mostly decide what we see. While this shift can be disconcerting, it’s also an accepted part of the user experience.
A personalized news narrator could restore some agency to the user. Imagine being able to define not just what you want to see, but how you want to see it. That means AI has to play an active role in not just handling data—but in shaping meaning.
F.A.S.O.N.’s visual identity is striking. Why the pixel-art mask and retro radio aesthetic? What cultural or visual inspirations influenced that decision?
We wanted to bridge nostalgia with the digital age. The radio aesthetic recalls a time when people engaged with news through voice alone — a simpler, more personal era where commentary was often tied to lived experience. So, radio was both a medium and a symbol.
The pixel-art style emphasizes F.A.S.O.N.’s digital nature. It signals that these personas are simulations, not real faces, but distinct viewpoints. It suggests they carry perspective, not identity.
We also considered long-term flexibility. These personas were designed to be adaptable to different formats—audio, animation, maybe even in-game characters. The visual style makes those transitions feel seamless and consistent.
Why only Turkish? Is there a reason you’ve limited the project to Turkish for now? Are there plans for multilingual expansion?
Turkish was the most natural starting point, both personally and culturally. Without a deep understanding of Türkiye’s agenda dynamics, media language, and social reflexes, it wouldn’t have been possible to create such a realistic media simulation.
In fact, being able to simulate different political and cultural identities in Turkish using AI was a test in itself—technically and sociologically.
From an infrastructure standpoint, we’re already multilingual-ready. But creating the same sense of authenticity in other languages requires more than just translation.
Every language comes with its own cultural context, media codes, and news priorities. That means separate modules, different datasets, and unique character development.
In the near term, we’re planning to pilot some English-language content. Long-term, we imagine a multilingual F.A.S.O.N. universe with AI personas rooted in local contexts across the world.
2 min read
