OpenAI published research on how people actually use ChatGPT, based on 1.5 million messages. The finding worth sitting with: the fastest-growing use of the most capable consumer AI ever built is looking things up. People are using it to replace Google.
Nearly 80% of all ChatGPT usage falls into just three categories. Seeking information has grown from 14% to 24% in a year, and OpenAI describes it plainly as a close substitute for web search. Writing, meanwhile, has fallen from 36% to 24%.
- Practical guidance: 29%
- Seeking information: 24%, up from 14%
- Writing: 24%, down from 36%
Half of all messages are simply asking. Only about a third of usage is work at all. Anthropic’s own research shows the same shape: on Claude, conversation and written documents each account for roughly a third of everything produced.
This is not a failure of imagination
The easy conclusion is that people are not creative enough with the tools they have been given. We think that reading is wrong, and a little insulting. People use AI for questions because a question is the only thing it can finish by itself.
Ask it to draft an email and you still have to send it. Ask it to plan a launch and you still have to build the board, write the brief, book the review, brief the team. The AI gives you a very good answer. You do the work.
The model was never the bottleneck
Every frontier model can already reason through a product launch better than most teams. What it cannot do is open your calendar, create the tasks, load the CRM, publish the page. So it hands you instructions through a chat window and stops at the glass.
An assistant that can only talk will always be used for talking.
That is the gap the usage data is really measuring. Not a limit of intelligence, a limit of reach. The intelligence is sitting in a box beside your business instead of inside it.
What changes when the AI has hands
This is why we built Wasabe the way we did. The apps are not integrated with the AI, they are built into it: the calendar, the CRM, the docs, the site builder, the video studio. Same class of model everyone else has. It simply has somewhere to put its hands.
Ask it to plan Friday’s launch and the board fills, the brief gets written, the go/no-go lands in the calendar and the team gets the message. Not a set of instructions about the work. The work.
Almost none of this is claimed yet
Pew Research found that about half of American adults have tried an AI chatbot and roughly a quarter use one daily, overwhelmingly to ask things. Statistics Canada reports that four in five Canadian businesses still use no AI at all. This is not a saturated market. Most of the value has not been touched, because most of it was never reachable from a chat box.
The question was never how smart the model is. It is how much of your business it can reach.
