People rarely post their real opinion online. They post the version that will not cost them anything: the safe agreement, the vague joke, the silence that reads as neutrality. Underneath that public feed sits a much larger, mostly unrecorded layer of private belief, things people think but do not say out loud, especially when a topic touches religion, politics, relationships, or anything else that might make them look out of step with the people around them.

Kiko Zang is the CEO and Co-Founder of Chomp, a consumer app built to pull that private layer into view. Chomp asks players a short question every day, asks them to guess how the rest of the room answered, then reveals both sets of results side by side. The company describes the difference between those two numbers, what someone actually believes and what they think everyone else believes, as a kind of data that does not exist anywhere else on the internet.

In this episode of Lead with AI, Dr. Tamara Nall speaks with Kiko Zang about why she built Chomp around private belief instead of public opinion, how the app's scoring system borrows from a decade-old piece of crowd-wisdom research, and why she now sees the resulting data as a resource for AI companies trying to understand how people actually think.

The Internet Rewards Performance, Not Honesty  

Kiko has spent most of her life moving between cultures: born in China, raised in New Zealand from age twelve, educated in the United States, and later based in Asia and Berlin before returning to the U.S. That path, she says, became a long study in how people from different backgrounds signal belonging, and how much gets lost between what someone privately thinks and what they say to fit in.

She points to the Edelman Trust Barometer, an annual survey from the public relations firm Edelman, which she says recently recorded its highest level of what researchers call insularity, meaning roughly seven in ten people are reluctant to trust or engage with anyone who does not already share their views. Kiko argues that social platforms and search algorithms are not designed to isolate people on purpose. They are designed to keep people scrolling, and the content that keeps people scrolling tends to be either something they already agree with or something that makes them angry enough to react. Over time, both patterns narrow the range of opinions any one person actually sees.

What Happens When You Guess the Room and Get It Wrong?  

Chomp's daily questions work in two steps. First, a player answers honestly. Then the app asks a second question: what percentage of everyone else does the player think will agree with them? Kiko walks through an example from the app's early data. Players were asked to rate "I love you" within the first month of dating as a green flag or a red flag, and 92.7% privately picked a green flag. But when asked how many other people would agree with them, the average guess landed close to 50%, and the small group who chose a red flag guessed almost the same thing about their own answer.

That gap, Kiko says, is the entire product. Most people overestimate how much their private opinion will look normal in public, which is exactly the distortion Chomp is trying to measure.

Reading the Room, One Green Flag at a Time  

Whether someone "read the room" correctly is a separate score from whether they landed in the majority. A player can hold a minority private opinion and still read the room well, if their guess about everyone else's answer turns out to be accurate. Kiko says most players consistently overestimate agreement no matter which side of a question they land on, something she attributes to a basic overconfidence in how representative any one person's social circle really is.

Over enough questions, Kiko says patterns emerge: where a player is conventional, where they are an outlier, and whose opinions they tend to predict correctly. The company's next planned feature is small matchmaking groups built around specific points of surprising agreement or disagreement, rather than around shared demographics or interests.

The Research Behind the Score  

Chomp's two-question format is built on a method called Surprisingly Popular, first published in the journal Nature roughly a decade ago by two researchers who later went on to teach at MIT and the University of Pennsylvania's Wharton School. Kiko explains the concept using the researchers' own textbook example: is the capital of Illinois Springfield or Chicago? People who answer Chicago tend to also guess that most other people will say Chicago. People who answer Springfield, the correct answer, guess the same thing. Because more people expect the crowd to say Chicago than actually do, the method flags Springfield as the "surprisingly popular" answer, the one that beats a simple majority vote once expectations are factored in.

Kiko says Chomp applied the same method to a question about the 2024 U.S. presidential race and, months ahead of the actual result, the surprisingly popular answer differed from the answer most people expected the public to choose. She offers this as an example of the method surfacing a belief that had not yet become an obvious majority opinion, not as a verified prediction record.

Why She Walked Away From Crypto  

Chomp's first version ran on the Solana blockchain, with players earning small crypto rewards for participating. Kiko says that version proved the core mechanic worked, drawing around 50,000 users who submitted roughly 2 million answers over a year, but she eventually concluded that the promise of a payout was competing with the emotional reason people kept coming back.

Two moments pushed her toward a full rebuild. The first came from a member of Chomp's Telegram community, who told her that a question about love and loss, prompted by the recent death of a family member, helped him feel less alone simply by seeing how many other people had answered in similar ways. The second came from Chompy, an alligator mascot Kiko created to make an earlier, more technical version of the app feel approachable. A longtime user, part of what Kiko describes as an unusually large share of women in what had otherwise been a crypto-heavy user base, later put a Chompy figure on top of her own birthday cake. Both moments led Kiko to shut down the original crypto version and rebuild Chomp as a subscription product aimed at a mainstream, and especially female, audience.

The Data AI Labs Cannot Get Anywhere Else  

Kiko says her near-term customers are market research firms and consumer brands, a comparison she draws to Wishbone, an app that let high schoolers vote on side-by-side images and became popular with brands running informal product research a decade ago. Her longer-term customer is the AI industry itself.

She argues that large language models train mostly on public text, which she estimates captures a small share of what people actually think, since most of it stays in private messages, journals, or unspoken. A former post-training lead at Meta, whom Kiko consulted early on, told her a data set needs roughly 100,000 rows to be useful for that kind of work. Chomp's first version, she says, produced about 20 times that. She points to emoji preference as one example: a model could be trained on real answers to which emoji reads as warm versus threatening, gathered directly from Chomp's format.

What Comes After the App?  

Kiko's longer-term goal is for Chomp to become a reference point for public opinion, the way people already treat search engines for facts, an app people open to check what others think before assuming they already know. She argues that search results are increasingly shaped by paid placement, and that Chomp's anonymous, incentive-aligned answers offer a cleaner signal.

Asked what she would change about the world if she could change one thing, Kiko says she would want a single, widely trusted news outlet that most people watched together, arguing that today's outlets lean further than they admit. She points to Ground News, an independent site that rates outlets by political lean, as evidence of that lean. It is a personal opinion about how news gets made, not a claim about Chomp's product, but it echoes the same interest running through the rest of the conversation: what people actually think, and how rarely that shows up in what they say.

For anyone curious about the gap between public opinion and private belief, and what that gap is worth to a company trying to build better AI, Chomp is live now at Chomp.fyi, with a waitlist and periodic data reports for anyone who signs up early.

For more conversations with the founders building the next generation of AI, subscribe to Lead with AI on your favorite podcast platform.

Quick Answers  

What is Chomp? Chomp is a daily question app that asks players to answer honestly, predict how everyone else answered, then reveals both results, turning the difference between the two into data about what people really believe.

How does Chomp's scoring work? Every question has two parts: what you personally believe, and what percentage of other people you think will agree with you. Chomp compares your guess to the real results to score how well you read the room.

What is the "Surprisingly Popular" algorithm Kiko mentions? It is a wisdom-of-the-crowd method published in Nature about a decade ago. Instead of tracking only the most popular answer, it also factors in how confidently people expected others to agree, which can surface a hidden or emerging belief that a raw majority vote misses.

Who is Chomp's data built for? Kiko says near-term buyers include market research firms and brands studying public sentiment, while the longer-term goal is supplying anonymized, aggregated belief data to AI labs training models on human judgment.

Who is Kiko Zang? Kiko Zang is the CEO and Co-Founder of Chomp. Before starting the company, she worked at Deloitte, Mattel, and IDEO, then joined the crypto exchange Orca in 2021 as its first business hire, leaving as COO in 2023 to build Chomp full time.

Follow or Subscribe to Lead with AI Podcast on your favorite platforms: Website: LeadwithAIPodcast.com | Apple Podcasts: Lead with AI | Spotify: Lead with AI | YouTube: @leadwithaipodcast | Facebook: Lead with AI | Instagram: @leadwithaipodcast | TikTok: @leadwithaipodcast | Twitter (X): @leadwithai

Follow Dr. Tamara Nall: LinkedIn: @tamaranall | Website: TamaraNall.com | Email: Tamara@LeadwithAIPodcast.com