Take a raw capability, wrap it in the simplest possible interface, release it free to everyone, and let real usage define the category, distribution model, and product rather than defining them upfront.
OpenAI
OpenAI is the research lab turned consumer phenomenon behind ChatGPT, the fastest product ever to reach 100 million users. Its through-line is a single instinct executed three ways: take a raw research capability, wrap it in the simplest possible interface, put it in front of ordinary people for free, and let real usage define the category, the distribution model, and the product itself.
Category Creation: turning a research demo into the default consumer AI
The problem. Before late 2022, large language models were the property of researchers and a few developers with API keys. There was no mainstream category for "talk to an AI and get useful answers." GPT-3.5 was powerful, but to the public it was invisible, and OpenAI had no proven way to make an ordinary person understand what it was for.
The approach. On November 30, 2022, OpenAI shipped ChatGPT, an internally titled "Chat With GPT-3.5," as a free, open-to-everyone chat box framed deliberately as a "research preview." Rather than defining the category with marketing or a spec sheet, OpenAI let the blank text field and millions of unscripted prompts define what the product was.
How it solved it. Usage created the category almost overnight. ChatGPT hit one million users in five days and an estimated 100 million monthly users within about two months, a pace UBS called the fastest for any consumer app in history (TikTok took roughly nine months, Instagram about two and a half years). Sam Altman later wrote that the launch "kicked off a growth curve like nothing we have ever seen," and "generative AI chatbot" became a household category with ChatGPT as its default.
GTM: no waitlist, no price, no sales team
The problem. OpenAI was a research organization with no consumer go-to-market muscle: no ad budget, no sales force, no distribution deals. Its prior commercial motion, the GPT-3 API launched in June 2020, reached developers through gated access and usage-based pricing, but that model could never reach a mass audience.
The approach. For ChatGPT, OpenAI chose the opposite of gated: completely free, no waitlist, no invite code, no download, just a URL. The go-to-market was the product itself, engineered for zero-friction trial and instant shareability, so every impressive answer a user pasted into Twitter or a group chat became a distribution channel.
How it solved it. Product-led, word-of-mouth growth did what a marketing budget could not. Altman announced the launch with a single tweet ("today we launched ChatGPT"), and organic sharing drove one million sign-ups in five days with no paid acquisition. The frictionless free tier remained the top of the funnel that OpenAI later monetized through ChatGPT Plus and enterprise plans.
PMF: shipping at the lowest bar to let demand reveal itself
The problem. OpenAI genuinely did not know whether ChatGPT had product-market fit. Internally, expectations were low: the team had trained GPT-4 already and found it more exciting, and a beta with roughly 30 to 40 friends and family produced polite approval but no raving. There was real uncertainty about whether anyone outside the lab wanted this.
The approach. OpenAI tuned the base model into a helpful, conversational assistant using reinforcement learning from human feedback (RLHF), where human raters ranked responses for truthfulness and helpfulness, and then shipped it at the lowest possible bar to measure real demand. As scientist Liam Fedus put it, "We didn't want to oversell it as a big fundamental advance," and policy lead Sandhini Agarwal described it as a "research preview" meant to gather feedback, not a finished product.
How it solved it. Live usage settled the question the team could not answer internally. The explosive uptake, one million users in five days and roughly 100 million within two months, was a product-market-fit signal that dwarfed every internal prediction, and it proved the conversational, instruction-following format (not just a bigger model) was what users actually wanted. RLHF-driven conversational data, Fedus noted, delivered the leap in usefulness despite minimal underlying changes from its predecessor.