We spent ten months building immersive 3D worlds, interactive games and generated video for studying. Then we counted what students actually asked for. They asked for flashcards.
We have published the numbers in full, including the ones that are awkward for us: What 3,735 students in 145 countries actually do with an AI tutor.
44%
of everything students generated was a flashcard deck or a mind map
22%
of uploaded files are photographs, not documents
<6%
every immersive and game format combined
Two findings, both older than the technology
Students on iTutor can generate 27 different kinds of study material. Immersive worlds. Live AI classes. Podcasts. Exam simulators. Generated video. Over ten months they made 1,688 of them, and flashcards and mind maps account for 44% of the total. Every immersive or game format put together comes to less than 6%.
It is not a quirk of one market either. Flashcards rank first in India, the United States, the Philippines and South Africa. In Egypt mind maps edge ahead, with flashcards immediately behind. Four very different places, the same answer.
The second finding surprised us more. Of 4,939 files students uploaded, 1,102 are images: a page of a textbook, a whiteboard before it gets wiped, a handwritten problem set. Not files chosen from a folder. Photographs taken seconds earlier. That behaviour comes from 269 separate students, so it is not a handful of outliers gaming the average.
Why we are publishing something inconvenient
We build the immersive things. Reporting that almost nobody picks them is not a comfortable position for a product team.
We think the alternative is worse. If the whole category keeps designing for a student who wants a simulation, and the actual student wants a cue card generated from a photo of their notes in under a minute, then everyone is building the wrong roadmap in parallel. Someone has to put the numbers out.
No individual data. Every figure in the report is an aggregate count. No student, institution, document or message is identified, no group smaller than 25 accounts is reported on its own, and no user content was read to produce it.
What we changed our own mind about
Two things we would tell anyone building in this space. Image input is not an accessibility extra; on this evidence it is close to a quarter of what arrives, and a tool that only accepts documents simply cannot see that behaviour. And novelty is not demand: the formats that look best in a launch video are the ones our users generate least.
The report carries the full method, including what the data cannot tell you. These are our users rather than a random sample of students, the sample is modest, and it skews toward markets where we have reach. We would rather state that plainly than round it off.
We intend to repeat it every year at the same address, so the figures can be compared over time.