MOUNTAIN VIEW, Calif. — February 13, 2026 — The average Fortune 1000 company boasts over 30,000 employees, with engineering, sales, and marketing teams numbering in the hundreds. Similar-sized groups exist within government, scientific, and defense organizations. Yet, research indicates the ideal size for a truly productive, real-time conversation is surprisingly small: just 4 to 7 people.
The Limits of Large-Group Dialogue
Why bigger isn’t always better when it comes to brainstorming and problem-solving.
- As groups grow, individual speaking opportunities diminish, leading to frustration.
- Traditional methods like polls and surveys capture data but lack the dynamic exchange of conversation.
- Hyperchat AI divides large groups into smaller, interconnected discussions facilitated by AI agents.
- Studies show Hyperchat AI can amplify collective IQ and foster a greater sense of collaboration.
The reason is straightforward: as groups expand, each person has less chance to speak and must wait longer for their turn, increasing frustration that their views aren’t adequately considered. This holds true whether teams collaborate in person, via video conference, or even through text chat, where messages quickly become buried and stifle meaningful deliberation. Simply put, productive team conversations don’t scale.
So, what’s the solution when you need to tap into the collective knowledge of a large team? Many organizations resort to polls, surveys, or interviews. While these methods gather individual perspectives, they often leave people feeling unheard and rarely uncover optimal solutions.
These tools aren’t designed for deliberation. They lack the back-and-forth exchange where team members debate issues, offer reasoning, present arguments, and ultimately converge on solutions through their collective merits. Surveys treat people as oversimplified data points, while interactive conversations recognize them as thoughtful data processors—a profound difference.
After more than a decade of studying this challenge, I’m convinced the best way to unlock the true collective intelligence of large teams is through authentic, real-time conversations at scale. I’m talking about thoughtful discussions where scores of people can brainstorm, prioritize, and forecast together, ultimately arriving at solutions that genuinely leverage their combined knowledge, wisdom, and insight.
But aren’t conversations impossible to scale?
Not anymore. Over the last few years, a new communication technology, Hyperchat AI, has emerged. It enables large, distributed teams to hold productive discussions where they can debate issues, brainstorm ideas, prioritize alternatives, provide arguments and counterarguments, and efficiently come up with solutions.
Inspired by large natural systems, Hyperchat AI combines the biological principles of Swarm Intelligence with the power of AI agents. It works by dividing any large, networked group into a set of small, interconnected subgroups, each sized for thoughtful real-time conversation via text, voice, or video. The key is a swarm of AI agents, called “conversational surrogates,” that participate in each local discussion and connect all the subgroups into a single, coherent deliberation.
With Hyperchat AI, groups of any size can debate issues, brainstorm ideas, prioritize options, forecast outcomes, and solve problems in real-time. Research demonstrates that teams using this approach converge on smarter, faster, and more accurate solutions. In one study I participated in, groups connected by Hyperchat AI amplified their collective IQ to the 97th percentile.
Another study, conducted with Carnegie Mellon University, found that groups of 75 people using Hyperchat AI felt more collaborative, productive, and heard compared to traditional communication tools like Microsoft Teams, Google Meet, or Slack. They also reported greater buy-in to the solutions they reached.
Super Bowl Ads: A Real-World Test
To demonstrate Hyperchat AI’s capabilities, the research team at Unanimous AI (the developers of Thinkscape, a platform utilizing Hyperchat AI) brought together 110 members of the public who watched this year’s Super Bowl to debate which ad was the most effective, and why. With 30-second spots costing between $8 to $10 million (excluding production), the stakes are high for brands aiming to stand out.
The 110 participants were divided into 24 subgroups, each with 4 or 5 people and a single AI agent. Each agent observed its subgroup, identified key insights in real-time, and then shared those insights with AI agents in other subgroups. When agents received external insights, they integrated them into their local conversation, expressing the insight as a member of their group. This process weaves all deliberations into a single, real-time conversation that converges in unison.
The 110 participants initially suggested 54 different ads for consideration, reaching a decisive answer in just 10 minutes. Because the AI agents tracked the dynamics within all 24 debates, the system instantly generated a ranked list of all 54 ads based on conversational support across the entire population.
The Pepsi ad featuring the Coke polar bears was identified as the most effective by a significant margin (p<0.01). The system’s analysis of the ad’s effectiveness included this reasoning: “We found it effective due to its humor, clever use of polar bears, jab at Coca-Cola, memorability, nostalgic elements, wide appeal, product focus and ability to spark conversations. While some of us criticized it for focusing on a feud, a large majority felt it successfully captured the essence of a classic Super Bowl ad.”
The group also identified the least effective ad: the Coinbase spot. Their reasoning: “We found it lacking in clarity, with confusing messaging and a failure to explain the product effectively. Additionally, the ad was found by many to be annoying, cringey and low-effort, with little promotion of the product and a disconnect from Coinbase’s services. Overall, it failed to build trust and was off-putting to many viewers.” This selection was also statistically significant (p<0.01).
While this was a lighthearted experiment, I’ve observed similar results with large groups—from financial analysts to scientists—using this technology to tackle complex issues. In all cases, the groups converged on solutions with increased speed, accuracy, and buy-in.
For more information on the academic studies surrounding Hyperchat AI, see this recent paper.
Louis Rosenberg earned his PhD from Stanford University, was a professor at California State University (Cal Poly) and has been awarded over 300 patents for his work in human-computer interaction, AI and collective intelligence.
