The average Fortune 1000 company has more than 30,000 employees and engineering, sales and marketing teams with hundreds of members. Equally large teams exist in government, science and defense organizations. And yet, research shows that the ideal size for a productive real-time conversation is only about 4 to 7 people.
The reason is simple: As groups grow larger, each person has less opportunity to speak and must wait longer to respond, increasing their frustration that their views are not sufficiently considered. This is true whether groups collaborate in person, by video or teleconference, or even by text chat (which buries users in a backlog of messages that reduce participation and undermine deliberation).
Simply put, productive team conversations do not scale.
So, what do you do if you have a large team and you want to leverage their knowledge, wisdom, insight and expertise? For many organizations, their only choice is to resort to polls, surveys or interviews. This will capture data about individual perspectives, but nobody will “feel heard” when the process is over, and it will rarely find optimal solutions.
This is because polls, surveys and interviews are not deliberative instruments. There is no give and take as team members debate issues, provide reasons and rationales, present arguments and counterarguments and ultimately converge on solutions by virtue of their deliberative merits. Surveys treat people as over-simplified data points, while interactive conversations treat people as thoughtful data processors. This difference is profound.
I have been studying this issue for more than a decade, and I’m convinced that the best way to unlock the true collective intelligence of large teams is through authentic real-time conversations at scale. I am talking about thoughtful discussions where scores of people can brainstorm, prioritize and forecast together, ultimately converging on solutions that genuinely leverage their combined knowledge, wisdom and insight.
But conversations are impossible to scale, right?
Wrong — 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 emerging 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 by text, voice or video. The magical ingredient is a swarm of AI agents called “conversational surrogates” that participate in each local discussion and work to connect all the subgroups together into a single coherent deliberation.
Using Hyperchat AI, groups of potentially any size can debate issues, brainstorm ideas, prioritize options, forecast outcomes and solve problems in real-time. And it works — research shows that when large teams hold conversations this way, they converge on smarter, faster and more accurate solutions. In one study I was personally involved in, groups connected by Hyperchat AI amplified their collective IQ to the 97th percentile.
In another study, conducted in collaboration with Carnegie Mellon University, groups of 75 people holding conversations using Hyperchat AI technology said they felt more collaborative, productive and heard compared to traditional communication structures like Microsoft Teams, Google Meet or Slack. They also felt greater buy-in to the solutions that emerged.
To test the virtues of Hyperchat AI in a fun and timely format, I asked the research team at Unanimous AI (developer of Thinkscape, a platform that uses Hyperchat AI) to bring together 100 members of the public who watched the Super Bowl this Sunday and debate which Super Bowl ad was the most effective, and why?
I know this is not a question of grand social importance, but the Super Bowl is among the most watched events in the world, both for the athletic spectacle and the ads. This year, a 30-second spot cost between $8 to 10 million, not including production costs. With that level of investment, every brand is looking to stand out, yet only a few can achieve that.
So, we brought together 110 random members of the public — their only qualification being that they watched the Super Bowl — and asked them to discuss and debate the ads. Sixty-six unique ads ran during the game. Did any of them stand out strongly above the rest, and if so, why was it so effective?
The 110 participants were divided into 24 subgroups, each with 4 or 5 humans and a single AI agent. Each agent was tasked with observing their subgroup, identifying key insights in real time, then share those insights with AI agents in other subgroups. When agents received those outside insights, they then participated in their local conversation, expressing the insight as a member of their group. This process weaves all the deliberations together into a single real-time conversation that flows seamlessly and converges in unison.
All told, the 110 human participants suggested 54 different ads for consideration, and they reached a decisive answer in only 10 minutes of hyper-connected discussion. And, because the AI agents were tracking the dynamics within all 24 local debates, the instant the conversation finished the system generated an ordered list of all 54 ads based on the conversational support across the full population.
Here are the top ten as identified by the deliberating participants:
As you can see, the Pepsi ad that used Coke’s polar bear was found to be the most effective of the night by a wide margin. In fact, the Thinkscape system reported that this was a statistically significant result for a population of randomly selected consumers (p<0.01).
In addition, the system automatically tracks the reasons that emerge in every subgroup, and the reactions to those reasons (whether it swayed opinions of others, inspired counterarguments, or both). This enables the system to instantly produce a deliberative overview for every ad produced, assessing why the group viewed each ad the way it did.
Here is the reasoning instantly generated for the Polar Bear ad:
“Our collective perspective is that the most effective Super Bowl ad of 2026 was the Pepsi Polar Bears spot. 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.”
For the record, the team at Unanimous AI also asked this real-time collective to consider a follow-up question, Which Super Bowl ad was the least effective and why? This is what the system reported after 10 minutes of deliberation:
“Our collective perspective is that the worst 2026 Super Bowl ad was the Coinbase spot. 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.” Note: The selection of this ad was a statistically significant result (p<0.01) across the population.
Again, this was just a fun example for engaging the public, not a large deliberation of grand importance. That said, I have observed large groups, from analysts in large financial institutions to scientists at the Department of Energy, discussing important issues using this technology — and in all cases the groups seem to converge with increased speed, accuracy and buy-in.
For an overview of academic studies on Hyperchat AI, check out 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.
