You hit refresh for the fifth time. The little loading icon spins, mocks you, and dies. “Something went wrong.” You check Twitter, expecting mass panic over a global server crash. Instead, you see your peers happily generating code, writing essays, and automating their lives. You are locked out in the cold, and nobody else seems to notice.
When a tool fails for everyone, it’s an outage. When it fails for only you, it’s an experiment.
Recently, a thread popped up on Hacker News titled: “Ask HN: ChatGPT free tier just stopped working, is this just me?” The top comment was a chillingly simple three words: “Works fine here.” That right there is the smoking gun. You think OpenAI’s servers are just buckling under the weight of millions of free users. You assume the engineers are scrambling to fix a bottleneck. But the truth is much more calculated.
This isn’t a server crash. It’s an A/B test. OpenAI isn’t trying to fix your broken ChatGPT because it isn’t broken—it’s working exactly as intended. By selectively segmenting users and quietly degrading the free tier experience for a small, isolated percentage, they are measuring a very specific metric: frustration. They want to know exactly how much friction it takes before a free user finally breaks down and hands over $20 a month.
In the attention economy, if a service is free, you aren’t the customer. You’re the control group.
We all celebrated when these AI tools were offered to the masses for free. We thought it was a democratization of technology. But free tiers aren’t charity. They are trust signals designed to build dependency. Once you’ve reorganized your workflow, your study habits, or your business around a tool that costs nothing, the rules change. The provider holds all the cards. They can introduce intermittent failures, slower response times, or sudden capacity limits—all under the guise of “high traffic.”
It’s brilliant, really. If they just pulled the plug on the free tier entirely, there would be a massive public backlash. But if they make it just flaky enough? If it works 80% of the time and fails when you need it most? That creates anxiety. And anxiety is the ultimate conversion driver. You don’t buy ChatGPT Plus because you want more features; you buy it because you can’t afford the risk of it failing right before your deadline.
The most expensive software is the free kind that you build your entire life around.
Stop waiting for them to fix the bug. Stop refreshing the page. Understand the game being played. If you are a free-tier user of any AI service, your experience is being actively manipulated to optimize someone else’s bottom line. Save your prompts, find local alternatives, or pay the toll. But whatever you do, don’t act surprised when the free lunch gets quietly taken off the table.
FAQ
Q: How do you know it's an A/B test and not just bad infrastructure?
A: Because global server outages are uniform. When AWS goes down, everyone knows it. When a service fails for isolated individuals while the rest of the community reports normal operation, that is the textbook definition of a segmented rollout or A/B test.
Q: What should I do if I'm locked out of the free tier?
A: Stop relying on a single point of failure. Diversify your tools, look into open-source models running locally, or accept that if a tool is critical to your work, you need to pay for the premium tier to guarantee uptime.
Q: Should we be mad at OpenAI for doing this?
A: Mad is the wrong word. You should be impressed by the psychology. It's a masterclass in dark pattern monetization. They are converting user dependency into revenue without ever having to announce a price hike.