Polling In 2026: What To Expect from AI Research
Polling In 2026: What To Expect from AI Research
https://www.forbes.com/sites/willjohnson/2026/05/18/polling-in-2026-what-to-expect-from-ai-research/
Publish Date: 2026-05-18 09:11:00
Source Domain: www.forbes.com
Colourful overlapping silhouettes of people voting. EPS10 file, best in RGB, CS5 versions in zip
getty
Leading up to the November midterms, polling can say one thing today and another tomorrow. Will there be a “blue wave,” or is that prediction overblown? Are Democrats or Republicans more enthusiastic? How much of a factor is the Iran war?
Then there is artificial intelligence, which is changing public opinion research as we know it. Many Americans are worried about this change, expressing concerns that AI will make polling less accurate and more misleading. Even Gen Zers, who are naturally more digitally inclined, express increasingly negative sentiment about AI.
Cynicism and skepticism reign supreme for a reason. Because the highest-quality forms of electoral and market research take time to execute, not to mention double- and triple-checking (by human beings), the impulse for some researchers is to speed up the process. Not unlike the news business, there is a strong incentive to “be first” at all costs.
Cue the predictive human intelligence startup Brox creating a “parallel universe” populated by 60,000 digital twins of real, living humans, using AI to simulate demographic profiles and consumer preferences. Synthetic polling companies are now a dime a dozen, with AI-generated synthetic voters replacing real human beings to replicate their behavior. This can be exponentially quicker and significantly cheaper, with higher profit margins.
But it isn’t necessarily more accurate. Recent research suggests that the biases skewing old-school results can skew silicon sampling numbers even more strongly. AI is known to hallucinate, inventing incorrect answers. Simulated answers may also smooth out unconventional, but insightful responses. For example, a “common sense” response from an AI avatar of a certain demographic captures what people in that group should think, but we need to know what people actually think.
The takeaway for researchers is to put AI…
