Simile says it can survey millions of AI-made consumers for research

In short: Simile is selling a way for companies to get survey-like results by asking questions to large groups of AI “digital twins” instead of real people.

What's going on

Simile is a start-up based in Palo Alto. It builds AI “digital twins” of real people, meaning software profiles that try to act like a person would. You can think of it like a flight simulator, but for consumer behavior (a safe place to test decisions before doing them in the real world).

Companies use Simile through a web tool. They pick a group they want to study, like “U.S. millennial parents” or “CVS shoppers in the Northeast.” Then they run surveys or “what if” scenarios on hundreds of thousands, or even millions, of AI agents that are modeled on real individuals.

Simile says its system is trained using interviews and large survey datasets, plus some purchase and behavior data. It also says the agents have features like memory and planning, so they respond more like people than a simple checklist profile. The company charges partly based on how many agents it builds for a client.

What to watch

The big question is accuracy and when it holds up. Simile claims results can match real surveys at an 85% to 99% level, depending on the topic and the group being studied, but those are averages, not guarantees. Gallup is independently testing where these AI respondents match humans and where they fail, which will likely shape how much businesses trust synthetic surveys for high-stakes decisions.

Source: NYTimes