How students actually use AI: what 574,740 conversations showed

The report in one table

What it found

• Four styles, evenly spread. The report sorted conversations into four styles of use. Each made up between 23% and 29% of the total.
• Nearly half were direct. About 47% of conversations were what the report calls Direct: the student was seeking an answer, with minimal engagement.
• The AI did the higher-level work. The report also classed what the chatbot was doing in its replies.

These six figures come from the report's chart. They add up to 98.2%. The categories come from Bloom's taxonomy, a scale teachers use to rank thinking skills, and creating and analysing are among its higher levels. Together they were 70% of what the AI did in its replies. The report itself warns that students may be handing over the higher-order thinking.

Why a company would publish this

A company that reports close to half the student use of its product as answer-getting is not flattering itself. That counts for something. It is still the company describing its own product, with its own tool and its own categories. Nobody outside has checked the sorting.

What it did not show

• Not whether students learned. The report counts kinds of conversation. It has no test scores and no grades. Anthropic says so.
• Not cheating. A direct request can be a student checking a fact they already understand. The report cannot tell that apart from a student copying an answer.
• Not all students. These were early users of one chatbot, at universities, over 18 days. Some accounts will have been staff. School pupils are not in the data.
• Not exact figures. An automated tool did the sorting, and Anthropic notes that it makes classification errors.

We found no newer report on student use from the same company. It did publish a report on how teachers use the chatbot, in August 2025, based on about 74,000 conversations.

What surveys of younger students say

For teenagers, the best figures come from surveys, which ask people what they do. In a Pew Research Center survey of 1,458 US teens aged 13 to 17, published in February 2026, 54% said they had used chatbots for help with schoolwork. 10% said they do all or most of their schoolwork with chatbots, 21% some and 23% a little. 59% said classmates use AI to cheat at least somewhat often.

A survey shows what people say, not what they do. But the two sources agree on the broad picture: use for study is common, and a good share of it is answer-getting.

Why the direct style matters

The usage report cannot say that direct use hurts learning. Experiments can test that, and several have. In a field trial with about 1,000 students, AI that gave answers went with exam scores 17% lower, and AI that gave hints largely avoided it. In three experiments with 1,222 adults, people who asked for direct answers gave up more afterwards.

So the report tells you how common the direct style is. The experiments tell you why that is worth changing.

What to do with this

Look at your own last ten chats. Count how many started with "write", "solve" or "give me" before you had tried anything. Then switch the order: your attempt first, the AI second.

Questions