What 3,624 students in 145 countries actually do with an AI tutor

Every AI study tool is sold on the same promise: that learning is about to become immersive, gamified, generated. We build those things. We also watched what 3,624 students in 145 countries did when we put roughly three dozen of them on one screen and let people choose.

They chose flashcards.

Finding 1: given 27 formats, students picked the two oldest

Over ten months, students generated 902 study artefacts. They had their pick of immersive 3D worlds, interactive games, generated video, podcasts, live AI classes, exam simulators and more. Flashcards and mind maps account for 51% of everything made. Every immersive or game format combined accounts for 7.5%.

India is the one country with enough students making study items to report on its own, and flashcards rank first there too.

Finding 2: a quarter of what students hand an AI is a photograph

Of 4,643 files uploaded, 1,090 are images. Not documents: pictures. A page of a textbook, a whiteboard before it gets wiped, a handwritten problem set, a worksheet the teacher printed.

This is a broad behaviour rather than a handful of outliers: the photographs come from 267 separate students, averaging four each, and the five heaviest uploaders account for only 7% of the total. It is worth stating the limit precisely, though: those 267 students are 16.8% of the 1,587 people who uploaded anything. The right claim is that 23% of uploads are photographs, not that 23% of students photograph things.

What we think this means

The two findings point the same way. Students are not asking for a new medium. They are asking for a very old one, delivered faster, from whatever is in front of them.

That is an uncomfortable result for a company that builds 3D worlds, and we are publishing it anyway, because the alternative reading is worse: that the industry keeps building for a student who does not exist. The demand we can measure is for the boring thing, produced instantly, from a photograph taken thirty seconds ago in a classroom.

Two implications we would offer for anyone building in this space. First, image input is not an accessibility feature or a nice-to-have; on this evidence it is close to a quarter of your input volume, and a tool that only accepts documents is invisible to that behaviour. Second, novelty is not the same as demand. The formats that photograph well in a launch video are the ones our users generate least.

Methodology

• Population. 3,624 individual accounts on iTutor's public platform, spanning 145 countries by signup region. Staff, internal, test and demo accounts are excluded, and so are accounts that asked to be deleted.
• Window. 3 November 2025 to 27 August 2026, inclusive. All figures are cumulative over that period.
• Finding 1 counts 902 study artefacts generated across 27 distinct formats. Each one is an artefact a student in the population above asked for; artefacts made by excluded accounts, and by school and company accounts outside the public platform, are not counted.
• Finding 2 counts 4,643 uploaded files, classified by detected file type at ingestion. "Photographs" means the image class, which covers camera captures and screenshots; we cannot separate those two, and the figure should be read as "images" rather than strictly "camera photos".
• Suppression. No group smaller than 30 students is reported on its own. In finding 1 that leaves five formats, shown separately, and one country, India, with 30 or more students making study items.
• Anonymity. Every figure is an aggregate count. No individual, institution, document or message is identified, and no user content was read to produce this report.
• What this cannot tell you. These are iTutor's users, not a random sample of students, and people who choose an AI study tool are not representative of students generally. The sample is modest and skews toward markets where we have distribution. Treat the direction as informative and the precise percentages as specific to this population.

We intend to repeat this annually at the same address, so the figures can be compared over time. If you are researching this area and want a cut of the aggregate data we have not published, write to us.

Correction, 10 October 2026: the first version of this report counted study items made by staff and test accounts, although its method said they were excluded. The figures above exclude them. The main findings hold; the share of flashcards and mind maps rose from 44% to 51%.