OpenAI shouldn’t do credits in ChatGPT

A few days ago, Sam Altman floated adding credits to ChatGPT on X: 

‍

‍

Theoretically, we should love this. We support complex billing use cases, so the more that spreads, the more money we’ll make, right?

‍

Maybe that’s true, but I still don’t think OpenAI should add credits to ChatGPT. The reason why is a great case study in how monetization affects UX—and why different categories need different pricing models. 

‍

How monetization affects UX 

‍

Almost all of the responses to the tweet were basically “don’t do this” because it would harm the user experience: Users would get precious about their credits and evaluate which queries are worth spending a credit on and which aren’t. 

‍

That would create more hesitation to use ChatGPT, which is precisely the opposite of what OpenAI wants. As a consumer product, ChatGPT succeeds when it’s ubiquitous and people’s first resort to solve all sorts of problems. 

‍

This is how monetization/pricing concretely affects the user experience: If your pricing is at odds with how users realize the value of your product, it will make their experience of using it more annoying. 

‍

To take an even more extreme example, nobody would want to pay Figma by the frame, even if your total bill would be less. It would just be so annoying to deal with and lead to weird user behavior.

‍

But there’s another side to this: Pricing also needs to make sense for your business and how you need to pay your vendors. OpenAI’s models don’t have equal compute costs. You can see this by the many different prices on OpenAI’s own pricing: 

‍

‍

And if you use Deep Research (based on the o3-mini model), it outputs thousands of words every time you prompt it. If you constantly prompted thousands of words via o3-mini via the API, it’d be way more expensive than $20/month. 

‍

So my argument isn’t that a $20/month plan should make everything available with no limits. That’d be like saying a buffet should offer unlimited lobster and still expect to get all you can eat at a single price. Wait, that's a real place in France.

‍

‍

Anyway, my argument is that credits are a bad model for ChatGPT specifically and simple, availability-based rate limits might have the same effect, but feel better psychologically. That’s because different categories require different pricing.

Why different categories require different pricing 

‍

How you should charge your customers also depends on the category you’re in. As a general rule of thumb, the closer your product is to infrastructure, the more likely it is to be charged on usage. The closer it is to a consumer app, the more likely it is to be a subscription. 

‍

That’s because infrastructure is almost always an ingredient in another product while apps are bought by the people who use them. 

‍

Usage-based pricing is annoying to deal with if you’re the one paying the bills. If that’s your job, that’s fine because you in turn get paid to minimize that (most likely AWS) bill. But if you’re an end user, you just don’t want to think about it, pay and get it over with. 

‍

There’s also a business reason for this: Infrastructure providers usually resell other usage-based infrastructure (AWS charges a margin on data center cost which put a margin on electricity cost). 

‍

If you’re building the end user app, you stop reselling infrastructure and build something on top of the infrastructure you bought—and charge a flat subscription, meaning each user needs to be below a certain threshold of usage. 

‍

This is where ChatGPT is in a tough spot: It needs to price like a consumer app, which means by subscription. But it attracts massive power users who will abuse unlimited usage because ChatGPT is just so powerful. 

‍

OpenAI is in a pinch: It’s sitting on piles of money, but it’s burning that money fast. Unless they figure out how to massively scale their consumer app ChatGPT without going under, they’ll be commoditized. 

‍

But the answer to that monetization isn’t in credits.

‍

‍