A journalist's experiment with her own AI avatar reveals the uncanny valley of digital PR

Alexandru Voica, corporate affairs lead at video-generation platform Synthesia, surprised a TechCrunch reporter this summer by introducing her to an unexpected new member of the company's PR operation: a virtual avatar of himself, programmed to field routine inquiries about Synthesia's capabilities and business model. The encounter felt like a watershed moment in how artificial intelligence is reshaping media relations. Synthesia, which achieved a $4 billion valuation earlier this year and reported crossing $100 million in annual recurring revenue last year, operates in a crowded field alongside competitors like D-ID, HeyGen, and Colossyan.

The platform enables enterprises to construct interactive training videos featuring AI avatars. Its recent Roleplay Sessions product allows employees to rehearse scenarios—such as sales conversations—with an interactive AI avatar that provides real-time responses and performance feedback. When Synthesia opened a new office in New York and offered to create a digital avatar of the reporter, the opportunity seemed too compelling to decline.

The creation process took place in a small film studio within Synthesia's office, where the company captured multiple photographs and a two-minute voice recording. Synthesia generated four distinct avatars: a personal version that recites any provided script, another personal version trained on the reporter's venture fraud investigation, and two interactive variants—all available with and without glasses. The interactive avatars operate through a technology stack combining voice-to-text, language modeling, and text-to-voice components, alongside Synthesia's proprietary video animation models. Customers can select alternative components from providers including Cartesia, ElevenLabs, Google, and OpenAI, and choose their own cloud hosting or rely on Synthesia's infrastructure.

Synthesia offers three distinct product categories: a video creation platform where users input scripts for avatars to deliver, an agentic platform called Sessions enabling interactive avatar engagement in surveys and roleplay scenarios, and an API layer allowing developers to combine Synthesia's models with external services. The company completed the reporter's avatars within several days. When tested with a generic script about autumn in New York, the personal avatar's voice reproduction proved accurate, avoiding the hoarseness present in the original recording. Non-technical acquaintances found the result simultaneously fascinating and unsettling.

The interactive avatar operates deterministically, responding only to questions within its training scope—in this case, the venture fraud story. When asked unrelated questions about the reporter's background or residence, it consistently redirected conversations back to its designated topic. Friends noted the voice lacked authenticity and the visual likeness fell short of the personal avatar, yet remained sufficiently convincing to provoke discomfort. The reporter's mother called it "amazing," though she and the reporter's father attempted to pose questions only they would know, each time receiving redirects to the venture fraud narrative.

I don't remember giving birth to two of you

the reporter's mother, joking after testing the model

The experience prompted reflection on journalism's future trajectory. Would audiences accept news delivered by avatars? One investor dismissed the prospect outright. Growing resistance to AI-generated content flooding social media and news platforms suggests skepticism. Yet other industry figures expressed uncertainty. Could avatars supplement or supplant human journalists? Might executives prefer engaging with AI journalist avatars rather than humans?

The reporter values the interpersonal dimensions of her work—connecting with sources, crafting narratives, and investigating emerging topics. Yet journalism's foundation rests on trust, something that cannot reasonably be delegated to artificial intelligence. Beyond journalism, the appeal of creating a digital self-clone seems evident: no need to manage work accumulation after time away, since an avatar version could perpetually address inquiries.

Avatar adoption across corporate environments remains uncertain. Following the initial excitement of possessing a digital twin, the reporter developed conflicted feelings. Upon observing her avatar after it finished speaking, she found herself waiting—uncertain for what. Perhaps anticipating a blink, an unprompted remark, a smile, or some indication of recognition. Since her digital twins operate deterministically, such moments will never occur. The reporter acknowledges how readily someone could develop what she terms "AI psychosis" with a nondeterministic avatar powered by an unrestricted chatbot.

The reporter shared a prediction with an investor: her generation, Gen Z, will likely resist normalizing digital twins. They evoke science fiction—the kind of technology previously confined to movies now made tangible. Still, she finds avatars less jarring than humanoid robots. If circumstances become uncomfortable, logging off remains an option. For now, her personal avatar stands ready to summarize the week's top stories.

https://share.synthesia.io/embeds/videos/ca910fc3-5aae-4fcb-b799-e594ddbbf66b