As artificial intelligence becomes more deeply embedded in online life, researchers at Pew Research Center say the polling industry is facing growing threats from AI-generated survey responses, fake participants and low-quality data — but they argue rigorous polling methods still remain reliable.
In a new Q&A published this week, Pew Research Center Vice President of Methods and Innovation Courtney Kennedy said the organization does not use AI to simulate public opinion, a practice sometimes referred to as “silicon sampling,” where companies ask AI systems to predict how people would respond to surveys instead of interviewing real people directly.
“Polling is fundamentally about humans — what they’re thinking and experiencing,” Kennedy said. “If we stop polling people and just assume AI knows the answer, we risk misunderstanding what’s actually happening in the public.”
Kennedy said research into AI-generated polling has shown systems can stereotype demographic groups, struggle to accurately reflect Republican viewpoints and understate disagreement in public opinion.
The report also highlighted concerns about “opt-in” online surveys, where participants self-enroll through ads or reward systems. Pew researchers said those systems are increasingly vulnerable to AI-driven fraud and bogus respondents completing surveys rapidly for financial gain.
According to Pew, the organization avoids many of those risks by using probability-based sampling methods. Rather than recruiting survey participants online, Pew randomly selects households from national address databases and contacts people directly through mail invitations.
Kennedy said that structure makes large-scale fraud significantly harder because respondents cannot create multiple accounts or repeatedly take surveys for money.
The report also warned about “bogus respondents” — people who rush through surveys without answering honestly. Researchers said those participants often skew results by disproportionately selecting positive responses, contributing to misleading polling data and, in some cases, media corrections tied to flawed surveys.
Still, Pew cautioned that even probability-based polls are not automatically trustworthy. Kennedy said poorly weighted or badly designed surveys can still produce inaccurate results, even when using stronger sampling methods.
The organization acknowledged that rigorous polling remains expensive because of the labor-intensive process required to recruit representative samples and provide both online and phone participation options.
The discussion comes as artificial intelligence increasingly reshapes research, political analysis and media industries, while public trust in polling continues facing scrutiny after several high-profile election misses in recent years.
FingerLakes1.com is the region’s leading all-digital news publication. The company was founded in 1998 and has been keeping residents informed for more than two decades. Have a lead? Send it to [email protected].