How ChatGPT went to market as a free research preview and let X do the rest

The X angleUsers shared ChatGPT screenshots on X within hours of launch, and Sam Altman used X in December 2022 to announce that it had crossed one million users.

Most go to market strategy writing assumes a launch is something a company engineers: a positioning document, a waitlist, a press embargo, a sequence of announcements. ChatGPT is the case that breaks the template. OpenAI released it at the end of November 2022 as a free research preview, framed as an experiment to gather feedback, and the go to market plan that followed was largely written by its users on X. For founders studying go to market strategy examples in AI, it remains the reference point, partly because it is so hard to copy and partly because the parts that can be copied are often misread.

The product and the bet

ChatGPT was not a new model so much as a new interface. OpenAI had offered its GPT models through an API and a developer playground for some time. What changed in November 2022 was packaging: a dialogue-optimized model from the GPT-3.5 series, fine-tuned with human feedback, placed behind a plain chat box that anyone with an account could use for free.

The bet was about framing as much as technology. By calling the release a research preview, OpenAI lowered expectations, invited people to probe the system's weaknesses and gave itself room to be wrong in public. A product launch promises something. A research preview asks for help. That distinction shaped how early users talked about it: they were testers sharing discoveries, not customers judging a purchase.

Who it was for

Officially, ChatGPT was for everyone and for no one in particular. There was no defined ideal customer profile, no vertical and no use case on the landing page. In practice, the first wave of users was the technology audience that already followed OpenAI and AI research on X: developers, researchers, founders and writers. That group had two useful properties. They knew how to stress test a model, and they had large, engaged followings.

A conventional go to market strategy for AI startups would warn against launching without a segment. ChatGPT got away with it because its first segment found itself, and because the product was general enough that every new user type discovered a fresh use case to post about.

The launch

The launch itself was understated. OpenAI published a blog post and opened access. There was no keynote, no pricing page and no enterprise sales motion. The interface was spare, and access was free.

What followed is well documented. Screenshots of ChatGPT writing poems, debugging code, drafting essays and explaining concepts filled timelines within hours. In early December 2022, Sam Altman posted on X that ChatGPT, launched the previous Wednesday, had crossed one million users. Coverage at the time, including reporting that cited OpenAI president Greg Brockman, put that figure at roughly five days after launch, and some outlets noted it may have counted sign ups rather than active users. Either way, it was a striking number for a product with no marketing budget attached to it.

The free research preview also had a cost. Demand outstripped capacity, and the service was frequently at capacity in its first weeks. When OpenAI later announced its paid tier, CNN noted that many people who wanted to try the tool had been locked out or put on a waitlist. Scarcity was not a deliberate tactic, but it added urgency and talk value.

How X carried it

ChatGPT's distribution on X ran on three engines, and only one of them was controlled by OpenAI.

The screenshot loop

The core unit of distribution was the screenshot. A chat transcript is a near perfect social object: it is self-contained, readable in a feed, surprising and easy to attribute. Each screenshot was simultaneously a product demonstration, a testimonial and a call to try it. Because outputs varied, people kept posting new ones, and the novelty did not wear off as fast as a single launch video would.

The research preview framing amplified this. Users posted failures as eagerly as successes: wrong answers, odd refusals and jailbreak attempts. For a typical startup, public failure screenshots would be a crisis. For ChatGPT, they kept the conversation going and reinforced the idea that this was something new enough to explore.

The founder as scoreboard

Altman's role on X was narrower than many assume. He did not run a thread campaign or a content calendar. His most cited post was a milestone: a short note that the product had crossed one million users. That kind of post works because it gives the audience a scoreboard. It turned scattered individual excitement into a collective story with a number attached, and it gave journalists a quotable, attributable figure from the company's chief executive.

Press that followed the feed

Mainstream coverage arrived quickly, but it followed the X conversation rather than leading it. Reporters could see the screenshots, embed them and point to the milestone post. For founders, the lesson is about sequencing: in this case, social proof on X came first and press amplified it, rather than the reverse.

What came next

The business model arrived after demand was proven, not before. On February 1, 2023, OpenAI announced ChatGPT Plus as a pilot subscription. The offer was framed around reliability and priority: general access even at peak times, faster responses and early access to new features. Crucially, the free tier stayed. OpenAI said the subscription would help support free access for as many people as possible, and that it was exploring lower cost plans, business plans and data packs. Plus opened in the United States first, and on February 10, 2023, OpenAI expanded it to customers outside the US.

That pricing structure is worth describing qualitatively, because the shape matters more than the number. It was a freemium model where the paid tier sold relief from the free tier's main pain point, capacity, rather than gating the core experience. Users who had been turned away at busy hours were the natural first buyers.

Product iteration then moved in several directions at once:

• Model upgrades. GPT-4 became available to Plus subscribers in March 2023, giving the paid tier a capability reason to exist on top of the capacity reason.
• Mobile. On May 18, 2023, OpenAI announced the ChatGPT app for iOS. It launched in the US first, was free to use, synced history with the web and included voice input through Whisper, OpenAI's speech recognition model. Plus subscribers got their paid features on mobile too.
• Business and platform offers. The early hint about business plans became a full product line over time, extending a consumer product into a b2b go to market strategy.

The pattern across all three was the same. Each step converted attention that already existed into a more durable relationship: a subscription, an app on the home screen, a workspace account.

The go to market lessons

ChatGPT is a misleading role model if read literally. Few startups will have a product that is both this novel and this broadly useful, or a parent brand that already commands attention among AI researchers on X. The analysis below separates the parts of the playbook that transfer from the parts that do not.

1. Framing can be a distribution tool

Calling the release a research preview did real go to market work. It invited exploration, made failures shareable instead of damaging, and set a bar that the product could clear. Founders can borrow this by launching with an honest frame (beta, preview, experiment) that makes early users feel like collaborators. The frame has to be true, though. A polished product dressed up as an experiment reads as a stunt.

2. Design the output to be posted

The chat transcript was inherently shareable. That is partly luck, but it is also design: a clean interface, readable text and outputs that make sense out of context. A useful question for any go to market plan on X is simple. What does a user screenshot after their first good session, and does it explain the product without a caption?

3. Let the founder hold the scoreboard

Altman's milestone post shows a modest but repeatable tactic. Founders do not need to be prolific posters to matter on X. A small number of credible, specific updates (a usage milestone, a capacity change, a new release) can anchor the conversation and give press something attributable. The key condition is verifiability: the number has to be one the company is comfortable standing behind.

4. Charge for the pain the free tier creates

Plus did not take anything away from free users. It sold priority during the exact moments the free tier failed. For a saas go to market strategy, this is a cleaner model than feature gating early on, because the upgrade reason is felt directly rather than explained on a pricing page.

5. Launch before the business model, but not forever

OpenAI launched without pricing and introduced it about two months later. That window let demand reveal what people valued. The analysis here is that the delay worked because the company could absorb the cost of free usage. A seed-stage startup with tight compute budgets should shorten that window considerably and decide in advance which signal will trigger the paid tier.

6. Do not mistake the channel for the strategy

It is tempting to conclude that X launched ChatGPT. More accurately, X amplified a product that was already remarkable to its first users. The channel compounded the product; it did not substitute for it. Founders reviewing their own go to market strategy sample should be wary of borrowing the launch style without the underlying surprise.

The launch in one table

Reduced to its parts, the ChatGPT launch can be summarized as follows. This is analysis of the public record, not a description of OpenAI's internal planning.

What the case means for a startup today

The ChatGPT launch sits at one extreme of the range of go to market strategy examples: minimal launch apparatus, maximal product novelty, and a social channel doing nearly all of the distribution. The more typical startup sits closer to the other end, with a narrower product, a defined buyer and the need to earn every share.

Still, the transferable core is clear. Ship something people want to show others. Frame the release honestly so that early flaws become part of the story. Use the founder's account sparingly, for facts that matter. Introduce pricing that sells relief from real friction. And treat X as the place where early proof is visible, not as a substitute for building that proof. For more cases in this series, see the full examples library, including how Perplexity launched into the same market only days later.