How Inflection AI mapped the state of consumer AI with Prolific
Inflection AI builds Pi, a personal intelligence partner designed to help people navigate everyday life. To help shape the future of personal intelligence, the company's research arm set out to answer a deceptively simple question about the consumer AI market: who actually uses these tools, what do they want from them, and who stays away?
The research was led by Jofish Kaye, Ph.D., Director of New Product Research at Inflection AI. Jofish has spent 25 years in user research and has had his studies cited thousands of times. Using Prolific, his team fielded two census-matched surveys covering 1,100 US adults and published the findings as the State of Consumer AI Research Report, providing deep insight into a landscape of public opinion that’s evolving every day. The challenge: Finding out what consumers want from AI.
"We set out to do this survey to really understand what people want from AI," Jofish explains. "I have a Post-it on my desk right now that says, 'Who should love us?' That's the question I'm trying to answer."
The study design came with three demands:
Unbiased recruitment. Advertising for chatbot users would bias the sample toward enthusiasts. The survey had to be framed around general technology use, with the chatbot question buried among 20 others covering everything from smartwatches to wired headphones.
Quality at scale. At 1,000 respondents, some people will check every box just to qualify. The team needed a participant pool where honeypot screening would leave a clean sample intact.
Honest opinions, including negative ones. Understanding why people choose, abandon, or avoid AI tools requires unvarnished views on real products, real companies, and the industry itself.
The team also built in room to get it right. Before the main study, they ran a 100-person pilot. "I always think of it as a pancake study," Jofish says. "When you're making pancakes, you always want to try out a first pancake before starting to crank them out.”
The solution: A verified pool
Screening that preserves honesty
A good practice in recruiting is to include honeypots: attractive options that will catch unqualified people just trying to get into the survey. For example, the main survey invited people to talk about their technology use. The first question listed some 20 technologies, including a fake one, “a barcode-based household multimedia system," which appeared right at the top of the alphabetized list. No such product exists. The 64 people who claimed to have used it were removed, along with 3 who later selected a fictional chatbot called "Afitoto," and roughly 15% of entrants who hadn't used a chatbot in the past week.
What remained was a clean sample of 1,000 chatbot users, matched to the US Census on sex, age, and region.
Quality respondents, responsive support
Over the past 25 years, Jofish has worked with many research vendors, and some experiences have been less than positive. But Prolific impressed him with both the quality of respondents and customer support.
"I've been impressed by the quality of the responses, the quality of the work, and the responsiveness of the team,” he says.
The surveys are one part of a wider research practice. Inflection AI also uses Prolific for hands-on user testing, putting real people in front of the product to see how they behave with it over time.
Fast, precise recruitment
Asked whether it took long to find respondents, Jofish's answer was short: "It's very, very fast." The average study fill time was 59 minutes, and the median completion time was 13 minutes.
The quality of responses was high, with an approval rate of 99.9% - 1,000 responses were approved, and only one was rejected.
The results: A published report and a sharper strategy
The State of Consumer AI Research Report was published by Inflection AI in July 2026, with the full methodology, honeypots and all, laid out for readers. Among the headline findings:
Chatbot use is a plural, healthy ecosystem. Users report using an average of 4.2 chatbots and reach for roughly 1.8 of them daily. The market looks less like the mobile OS duopoly and more like podcasting, with big players alongside a vibrant long tail.
People want relational AI. 80% of users picked at least one supportive role, such as mentor, coach, listener, or companion, among the five future chatbot roles they'd find most valuable. 55% picked more than one.
There are five kinds of chatbot users, from outcome-focused Capability Seekers to Emotionally-Led Power Users who describe their favorite tool as a "bestie" or "the one that knows me best."
Non-use runs deeper than missing features. Asked in their own words, most non-users simply had no need for a chatbot. Handed a checklist, 62% selected environmental concerns, a reason only around 7% had volunteered unprompted. The report interprets that gap as evidence that 'environmental concerns' function as a more acceptable way of voicing a broader distrust of the AI industry.
What's next: More niches, more surprises
Inflection AI Labs continues to study how different niches use chatbots because the market remains wide open. "Everything is still to play for," Jofish says. "There's still growth potential."
His advice to other research teams comes down to one focusing question: who are we trying to convince of what? Every study should change a decision. And the reason to keep asking real people is the thing no model can replicate.
"I'm always and consistently surprised by users," he says. "That's why you need to keep doing user research: to be surprised."
Get honest human data for your AI research
Prolific gives AI teams direct access to a global pool of 300k+ verified participants, with the screening rigor to keep your data clean and the honest, detailed responses that synthetic users can't provide. Get started for free today.
