AI tools have become incredibly popular over the last two years, and it's created a wide variety of offerings, free and paid alike. But they all have one thing in common:
they aren't making money.
Aside from some incredibly niche tools like T3 Chat, which already relies on other companies infrastructure, conversion rates for mass market AI tools is shockingly low. According to data from February 2025, OpenAI had a conversion rate of about 3% across all paying tiers. For reference, their weekly active user count was 400 million people. And this was before major social media booms from things like 4o image generation.
Claude has an even lower conversion rate, with Claude Pro only contributing to 15% of their overall revenue, and Perplexity having a shockingly low 0.7% conversion rate.
So what's going wrong? Let's consider a few things.
Confusing Value Propositions
Many of the top AI companies market their pro tiers as a way to "access frontier models" and "get extended usage" or use additional features like deep research, and more.
Now imagine you'd never used an AI tool before. Would you know what any of that is? Would you know the difference between an o4 and a 4o?
And as for extended limits and usage, how much is extended? They are intentionally vague with the figure as to not create a fear of limitation, but instead they create a phantom fear of hitting a limit unknown.
The main problem all of these tools have in common is they do a really bad job of showing, instead of telling. They rely solely on social media bubbles to show off their advancements, rather than showing directly in the value proposition what their best models are capable of, and what advanced voice mode is, and what else it could be capable of if you opted in.
This all leads the user to ask the famous question with any new advancement:
Why would I pay for this anyway?
It's a fair question to ask when so much of it is given away for free - and I feel like that's part of the problem. AI companies are so desperate for mass market adoption that they'll keep throwing expensive inference and diffusion onto free tiers and letting anyone use it, but they dont have a path to convert those users or a way to give them a justifiable reason to upgrade. This ends up creating a large portion of your market who wouldn't even consider the prospect of paid AI, because so much of the inference was already free, that they've just accepted that's how it should be.
Pricing for these apps can also seem steep and intimidating - the stark polarization of a $20 base tier, which will already seem high to a lot of people, placed right next to a $200 Pro tier which 95% of people won't even want, will feel very discouraging to a lot of people.
A good example of this is Manus, which launched to massive anticipation in early 2025, but the excitement fizzled as fast as it came because while it was a good tool and was genuinely innovative, people simply were not willing to pay the prices they needed to cover the inference and browser use. They quickly pivoted to attempting to give small daily limits to retain users, but by then it was too late, and now Manus is little more than its dedicated pool of users.
What can they do?
I feel like a lot can be done, the biggest ones would require breaking the whole chatbox format. But staying inside the box, I feel like the best starts would be:
Do more showing and less telling.
Your upgrade page should be more than a list of bullet points. You're selling me on the future, really sell me on it. Use graphics, demonstrations. Show me why your paid offering is going to help me.
Be more honest, and less vague.
Users want transparency, and limits should be communicated more fairly. I understand it has it's own problems, but I'd rather know and be mindful of that, than be fearful of a limit I don't know about running empty. Think about it like a fuel tank - would you rather know your fuel tank would run out soon, or just be sold on a bigger tank, with no gauge?
Be more approachable.
You shouldn't need a degree in words to talk to AI. To most people, it's an incredibly scary, confusing landscape. They need to break that mould and do more guiding to show that it can genuinely be useful and is worth investing in for our future, rather than relying on social media for press.
Overall, I find the current state of conversion very concerning and if I were a VC company I'd be demanding answers, since it's very clear this is a bubble and they have two routes. Ride the bubble until it pops, or grow the bubble. Many would like to think they'd grow the bubble, but the data simply doesnt match, and only time will tell if they can turn it around.