synthetic users

A synthetic user is not a human being but a computer program that interacts with an app, a website, or a service the way a human would. The program is given a description: for example, “43 years old, enjoys shopping online, avoids complicated menus.” It then acts the way a human with this background would presumably act. Companies use this to find out early where real users would fail or drop off – without first having to survey actual people. The profiles are usually driven by a large language model, that is, an AI system that understands and produces text.

Speed and cost in product testing

Classic user studies are laborious. You search for participants, schedule appointments, and evaluate video recordings. That takes weeks and costs a lot of money. Synthetic users can run through thousands of scenarios within minutes.

This speed advantage is especially valuable when a product is close to launch and weaknesses still need to be found quickly. A team can implement changes overnight and already evaluate test results the next morning – a rhythm that would not be possible with human testers.

On top of that comes scalability. While a real user study rarely has more than a few dozen participants, synthetic profiles can be created in any number. You can specifically simulate rare user groups, such as people with very poor internet connections or people who are barely familiar with digital forms.

How a profile is created and how it tests

The starting point is a so-called persona – a brief description of a fictional person with age, habits, goals, and possible frustrations. This description is handed over to a language model. The model, so to speak, slips into the role of this person.

The model then receives a task: “Book a flight to Vienna for under 200 euros.” It navigates through the interface, makes decisions, aborts if something is unclear – and logs every action along the way. The log shows developers at which point the simulation got stuck.

An important distinction: a synthetic user does not follow a fixed script written by hand beforehand. It reacts flexibly to whatever the interface currently shows – similar to a real human. This distinguishes it from classic automated tests, which only check whether buttons are clickable, not whether the flow feels intuitive.

Synthetic users in practice and their limits

Start-ups in the UX field – UX stands for User Experience, i.e., the experience someone has when using a product – have begun building entire platforms around synthetic users. Companies like Synthetic Users or Personas AI let customers configure their own profiles and have studies evaluated automatically.

Large technology companies are also experimenting with this internally. They deploy synthetic users before real beta phases start – that is, test phases in which real people use a still-unfinished version. Synthetic users filter out gross errors so that human testers are later confronted with something that already works on a basic level.

However, there is a hard limit: synthetic users learn from texts about human behavior, not from lived experience. Emotions, cultural nuances, or physical limitations – such as how someone with trembling hands operates a touchscreen form – are hardly reliably captured by them. That is why experts consider them a useful complement to real studies, not a complete replacement.