When Alexandru Voica, head of corporate affairs at video-generation startup Synthesia, shared a link to an interactive virtual avatar of himself earlier this summer, it marked a significant milestone in AI-assisted public relations. The digital clone was trained to handle press inquiries about Synthesia’s products, representing what some might consider the ultimate evolution of AI in PR.
Synthesia, originally founded in the UK, has emerged as a leader in the digital avatar space alongside competitors like D-ID, HeyGen, and Colossyan. The company recently celebrated a $4 billion valuation and announced last year that it had surpassed $100 million in annual recurring revenue (ARR), bolstered by investment from Adobe.
In September, the author visited Synthesia’s newly opened New York office and agreed to have an AI avatar created. While the company has previously built personal avatars that read scripted text, this marked the first time Synthesia created an interactive avatar for a journalist—or indeed, anyone outside of Voica. The interactive model is powered by a combination of voice-to-text, video generation, language processing, and text-to-voice technologies.
The creation process involved a trip to a mini studio within the office, where multiple photographs were taken and a two-minute voice recording was captured. Consent was required for the generation of these digital twins. The resulting avatars included both personal versions—which simply recite provided scripts—and interactive versions capable of listening and responding.
For this specific project, the interactive avatar was trained exclusively on the journalist’s prior reporting regarding fraud among venture-backed startups. When asked questions outside that scope, such as personal details about their life in New York, the avatar consistently redirected the conversation back to the trained topic. This deterministic approach ensures the avatar does not hallucinate or provide unverified personal information.
Testers, including the author’s non-tech friends and family, noted that while the voice resembled theirs, the visual likeness varied between the personal and interactive versions. The author’s mother described the experience as “amazing,” though she humorously remarked that she did not remember giving birth to two of them after the avatar refused to answer personal questions.
The technology stack behind these avatars allows enterprises flexibility. While Synthesia uses its own models, customers can integrate alternatives from providers such as Cartesia, ElevenLabs, Google, or OpenAI. Hosting options include either the customer’s preferred cloud infrastructure or Synthesia’s own servers.
Synthesia currently offers three main product types: a standard video creation platform for scripted content, an agentic platform called Sessions for interactive roleplay and surveys, and an API for developers to build custom solutions. The entire process of creating the journalist’s avatars took approximately two days.
The experience has sparked broader questions about the future of journalism and media. Can audiences trust AI-presented news? While there is growing pushback against AI-generated content on social media platforms like Instagram and Pinterest, some industry observers believe avatars could augment—or even replace—certain journalistic functions.
However, the author argues that trust, a cornerstone of journalism, cannot be outsourced to AI. The deterministic nature of current avatars means they will never exhibit genuine spontaneity or emotional depth. There are concerns that nondeterministic, chatbot-driven avatars could lead to strange psychological effects for users, though many find digital avatars less unsettling than human-like robots.
As avatar usage becomes more prevalent in corporate America, the technology raises questions about identity, authenticity, and the limits of AI replication. For now, the ability to interact with a digital twin remains a novel but limited experience, highlighting both the potential and the boundaries of current generative AI capabilities.
My mom said the funniest thing about not remembering twins! The tech is impressive, but the personality gap is obvious.
Trust is the cornerstone of journalism. I doubt we can ever outsource that to deterministic algorithms. Readers know the difference.