Synthetic user: the persona who replies

What is a synthetic user? Simply put: a persona or archetype is a description you read; a synthetic user is a persona with whom you have a conversation.

Based on target audience research, interviews or other user data, an AI model is fed that subsequently behaves as a specific type of user. The difference compared to a traditional persona is that the latter is static. A synthetic user "comes to life": you can ask follow-up questions, explore scenarios, and test responses.

We call that a synthetic user. You may also hear other terms, such as AI persona, digital twin, or simulated participant. These are not necessarily wrong, as long as you understand what is actually being simulated and where the limitations lie.

Synthetics are fake. Period. They simulate credible answers, based on their training data. Their knowledge of users extends no further than ours. 

Think of it as an actor preparing for a biopic. Rami Malek could study every interview with Freddy Mercury, watch every recording, and master his mannerisms and voice perfectly. But: Rami does not know what Freddie would order in a restaurant he has never been to before, or how he would react if his pension provider suddenly changed their login flow. 

Synthetics are brilliant at reproducing the known. They struggle with the truly unknown. They do not predict, but anticipate. That's why synthetic output is always a hypothesis, never a finding. Saying that out loud actually makes your research stronger. 

So the question is: how do we get as close to the truth as possible? When well-built, they stand on real user data, not on assumptions. UX teams want to capture their user in a workable form. First with the persona or other artefact, based on real research.

Synthetics are essentially the interface: a window to the user knowledge that lies beneath. Companies that take their users seriously work towards a growing knowledge base about their users. The synthetic is the way to unlock that.

That is the wrong question. They are not a miracle cure. The key word is 'when': when does a synthetic really add value, and when do you lose yourself in it?

For the Faraday team, this is not a theoretical question. We already deploy synthetics for clients. For instance, we scan pages early in the process from the context of a specific user group. This allows us to quickly identify where the friction lies before we sit down with real people.