Tavus Griffin AI Avatar: 48% Mistook It for a Real Person

What happened

Tavus has launched Griffin, which the company describes as the first "Human Interaction Model." The system holds real-time video calls and processes facial expressions, tone of voice and gestures as the conversation unfolds. In a Tavus study, 48 percent of participants believed they were speaking with a real person after a one-minute call, according to reporting by The Decoder.

The company's comparison point is significant. The Decoder reports that earlier systems reached only about two percent in the same kind of test, which suggests a sharp jump in perceived realism rather than a incremental improvement. The announcement centres on Griffin's ability to read non-verbal cues, not just generate speech or lip-sync to a face.

Why it matters

The headline figure matters less as a novelty and more as a signal about where conversational AI is heading. When a system can respond to a person's expression, vocal tone and body language in real time, the interaction starts to resemble the rhythm of a human exchange. For organisations already piloting AI in customer support, sales and onboarding, that changes the design question from "can the AI answer correctly?" to "how should the interaction feel, and what does the customer understand about who or what they are talking to?"

For leaders in experience and digital transformation, the implications run in two directions. Realistic AI agents could expand access to service at lower cost and with greater availability. At the same time, perceived realism raises the stakes for trust, consent and disclosure, particularly in regulated sectors where customers need to know who is accountable for a decision.

By the numbers

• 48% of Tavus study participants believed Griffin was a real person after a one-minute video call.
• About 2% is the figure The Decoder cites for earlier systems in the same type of test.
• One minute was the length of the call after which participants formed that belief.

The Renascence take

Most coverage will focus on the shock value of the 48 percent figure. The more useful reading is that realism is now a design variable, and it is one that organisations can set deliberately or drift into by accident.

The real risk is not that customers will be fooled for a minute; it is that they will be fooled by an interaction they cannot later explain or challenge. Good service design makes the nature of the interaction legible, with clear disclosure, an obvious route to a human, and a record of what was decided and why. Operators testing avatars like Griffin should treat human-likeness as something to be earned through transparency, not assumed as a conversion tactic. The customers who feel most respected will be the ones who never had to wonder who they were talking to.