OpenAI released ChatGPT on November 30, 2022, as what the company called a free research preview built on a fine-tuned version of GPT-3.5, with no major product launch event. The interface — a plain chat box — was not novel; large language models had existed in research and developer-API form for years. What changed was legibility: ChatGPT reached an estimated 1 million users within five days and, according to a widely cited UBS analysis, roughly 100 million monthly active users by January 2023, making it at the time the fastest-growing consumer application in internet history.
The mechanism was conversational fine-tuning through reinforcement learning from human feedback (RLHF), a technique that trained the underlying GPT-3.5 model to follow instructions and refuse certain requests in ways that made outputs feel usable to non-technical people rather than requiring prompt-engineering expertise. That interface layer, more than any single capability jump, is what converted a research artifact developers had used for two years into a product a parent, student, or executive could pick up in minutes.
Microsoft, which had invested in OpenAI since 2019, moved quickly to deepen the relationship, announcing a reported $10 billion investment in January 2023 and integrating OpenAI's models into Bing search and its Office productivity suite within months. Google, whose Transformer research had underpinned much of the field, reportedly declared an internal 'code red,' accelerating the launch of its own chatbot, Bard, in a rushed February 2023 demonstration that included a factual error and wiped billions from Alphabet's market value in a single trading day.
Educators and employers scrambled fastest: New York City public schools banned ChatGPT from school devices and networks in January 2023 over plagiarism concerns, reversing the policy within months as the technology proved impossible to police reliably. Coding-help site Stack Overflow temporarily banned AI-generated answers, and customer-service, copywriting, and paralegal work began absorbing chatbot-assisted workflows faster than management literature could describe them.
Coverage in the weeks after launch emphasized ChatGPT's ability to write essays, poems, and code on demand — the parlor-trick dimension — more than the structural fact that a major AI capability had, for the first time, been placed directly and freely in front of hundreds of millions of ordinary users rather than gated behind developer APIs or enterprise contracts. That distribution decision, not any single capability improvement, is what forced competitors' hands.
Italy's data-protection authority temporarily banned ChatGPT in April 2023 over privacy and age-verification concerns, the first national government to do so, before OpenAI adjusted disclosures and controls to resolve the issue within weeks — an early sign that consumer-facing generative AI would draw the same regulatory scrutiny long applied to other mass-market data-processing products.
OpenAI followed with GPT-4 in March 2023, a substantially more capable model, and a subscription tier, ChatGPT Plus, establishing a freemium business model that most subsequent AI labs copied. Search, education, and knowledge-work tooling reorganized around the assumption that a capable conversational model was now a baseline feature rather than a novelty, with Google, Microsoft, Meta, and a wave of startups shipping competing assistants within the following eighteen months.
OpenAI's ChatGPT made large language models a consumer habit in weeks — chat as the interface, not an API key. Schools scrambled for plagiarism policies; workplaces drafted bans then quietly adopted copilots. Competitor launches compressed years of product roadmap into months.
Hallucination, copyright training-data suits, and safety versus capability races defined the hangover. Enterprise procurement shifted from experimentation to seat licenses. The launch's mechanism was distribution: a capable model behind a free, sticky UI that taught the public what 'AI' feels like.
Student essays and customer-support scripts changed in the same semester. Knowledge work discovered a probabilistic intern that never sleeps and sometimes invents citations. Adoption outran governance because the demo was free and the utility was immediate.
The product decision to launch cheaply and broadly, rather than restrict access to enterprise partners as prior model releases had done, remains the template the entire generative AI industry still follows: ship a usable consumer surface first, gather feedback and usage data at scale, and let enterprise and API revenue follow distribution rather than precede it. ChatGPT did not invent large language models; it invented the mass audience for them.
Century Signals note: OpenAI launch materials and traffic reporting; contemporaneous education/enterprise adoption coverage; early litigation and policy responses. Editorial judgment about what still structures the present — not a comprehensive history.