Pollsters turn to AI simulations as a faster, cheaper way to test opinion

In short: More pollsters are using AI simulations instead of traditional surveys because they can be much faster and cheaper, and the results can be close to real voting outcomes.

What's going on

Pollsters and campaign teams are increasingly running AI-based simulations rather than calling or texting large numbers of real people. The main reason is cost and speed. The price to run thousands of AI “agents” at once, meaning many automated respondents, has dropped from hundreds of dollars to tens of dollars.

These systems can return results in minutes. One service mentioned, Aaru, says it can generate 5,000 simulated responses in about two minutes. Traditional market research can take months and cost large sums, so campaigns may prefer a tool that gives same-day feedback instead of waiting for last week’s numbers.

Supporters also point to accuracy. In the 2024 New York Democratic primary, Aaru’s simulation was off from the final result by 371 votes. Separate research from Harvard found AI-simulated answers often matched real-world survey data, with correlations above 85 percent on some topics like abortion.

A key selling point is avoiding “social desirability bias,” which is when people answer in a way that sounds good instead of what they really think. The idea is that AI personas may respond more bluntly, like an anonymous diary instead of a face-to-face interview.

What to watch

Experts warn these simulations still have blind spots. AI cannot copy physical experiences and can struggle with unpredictable human decisions, especially on newer issues where public views are still forming. A big question is whether campaigns will treat AI simulations as a helpful preview, or start relying on them like a replacement for talking to real voters.

Source: NYTimes