"Not Me" Podcast Episode #4: OpenAI's Secret Data

Hi, it's Vlad.Remember when I told you in AI Generalist that data is the new oil? Well, OpenAI just opened their books, sort of and the numbers are wilder than a London pub at closing time.Today's episode unpacks what 10% of Earth's adults are actually doing with ChatGPT. Spoiler: they're not writing code or automating spreadsheets. They're asking for recipes, relationship advice, and how to fix their washing machines.My source material is a fresh academic paper from OpenAI, Duke, and Harvard researchers who analyzed billions of ChatGPT interactions from launch through mid-2025. Think of it like reading someone's diary, except the someone is humanity itself, and the diary is 70% about non-work stuff.What you'll hear in the showThe Great Work Myth Gets Exposed. Everyone thought ChatGPT would be the ultimate productivity weapon. Plot twist: over 70% of messages have nothing to do with work. It's like buying a Ferrari to go grocery shopping, which, honestly, is exactly what humans would do.The Real ChatGPT Use Cases (What's Overlooked)* Practical Guidance: Not "optimize my workflow" but "how do I unclog my sink without calling a plumber"* Seeking Information: Less Wikipedia, more drunk uncle at Thanksgiving who actually knows everything* Writing: But not reports and emails, think love letters, apology texts, and "how to quit my job without burning bridges"The Education Paradox Nobody's Talking About Highly educated professionals use ChatGPT for work tasks, but here's the kicker they mostly use it for "Asking" rather than "Doing." They're not automating; they're validating. It's like having a PhD student double-check your homework, except the PhD student never sleeps and costs $20/month.Vlad's Newsletter is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.The 10% Club: What It Really MeansBy July 2025, one in ten adults globally uses ChatGPT weekly. That's not adoption that's cultural infrastructure. Think email in 2005 or smartphones in 2012. We crossed the chasm while everyone was arguing about AGI timelines.What most analysts miss: this isn't evenly distributed. The motivation pattern shows three distinct user tribes:* The Validators (30%): Smart people making sure they're still smart* The Delegators (25%): "Just write this for me" crowd* The Explorers (45%): Using AI like a Swiss Army knife for lifeThree Growth Patterns That Matter for BuildersPattern 1: The After-Hours Explosion Non-work usage growing faster than work usage tells you something crucial consumer AI is eating B2B AI's lunch. Remember The Great Restructuration? Turns out restructuring starts at home, not the office.Pattern 2: The Trust Gradient Users progress from fact-checking ("Is Paris really the capital of France?") to life decisions ("Should I take this job?"). The overlooked opportunity: products that accelerate this trust curve. Think Duolingo streaks but for AI confidence.Pattern 3: The Hidden Enterprise When 70% say "non-work," they mean "not my day job." But side hustles, personal brands, and passion projects ARE work, just not W-2 work. This shadow economy runs on ChatGPT like it's electricity.Five Plays Nobody's Running YetSince you know I can't resist turning data into dollars, here are the gaps this research reveals:* ChatGPT Coaching MarketplaceConnect the Validators with the Explorers. Charge for "prompt therapy" sessions where power users teach hesitant professionals. Like Peloton instructors but for AI workflows.* Non-Work Work StackBundle tools specifically for the 70% meal planning, DIY guides, relationship scripts. Position against productivity; sell peace of mind instead.* Trust Accelerator PluginChrome extension that tracks your ChatGPT accuracy rate over time, gamifies correct predictions, and gradually suggests harder use cases. Subscription: $4.99/month for "AI confidence analytics."* Industry-Specific ValidatorsSince educated professionals mainly "Ask" rather than "Do," build vertical Q&A tools. LawyerGPT doesn't write briefs; it sanity-checks them. DoctorGPT doesn't diagnose; it suggests differentials.* The Un-Productivity SuiteEmbrace the reality: people want AI for life, not work. Dating profile optimizer, argument settler, gift idea generator all the things OpenAI is too serious to build.Personal TakeWhen I started Belkins and Folderly, everyone said B2B was where the money lived. This data suggests the opposite: B2C AI is the sleeping giant.The motivation here? We've been building AI for the workforce when we should be building it for the human force. The 70% non-work usage isn't a bug it's the feature humanity actually wanted.What's typically overlooked in this topic: AI adoption isn't about capability; it's about permission. People need social proof that it's okay to ask a machine for help with personal stuff. Once 10% do it, the next 40% follow fast.Vlad's Newsletter is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.After You Listen* Which camp are you in: Validator, Delegator, or Explorer? Reply and tell me your ChatGPT origin story.* Want the raw data breakdown? I'm building a dashboard of usage patterns by profession and geography. First 50 readers get early access.* Next episode: We're dissecting how China's AI usage patterns are completely inverting Western assumptions. Hint: they're not using it for writing.Worth Reading While the Episode Downloads* AI Generalist - The playbook that predicted this exact usage split* Main Character - Why being early to obvious trends pays better than being first to clever ones* Ideation - Build for the 70%, not the 30%Post-Credit SceneThe paper's methodology "privacy-preserving automated classification" is fancy talk for "we read your chats with a robot." But here's what they didn't measure: emotional bandwidth saved.Every non-work ChatGPT message represents a human who didn't have to bother another human. That's the real productivity gain. We're not replacing workers; we're replacing awkward conversations.Also, quick observation from London: UK ChatGPT usage spikes at 11 PM. Either everyone's a night owl or they're too embarrassed to AI during office hours. There's a startup in fixing that shame somehow.Thanks for reading and listening. Vlad This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit www.vladsnewsletter.com/subscribe