First-generation students: access to chatbots is a given, but not their extensive use

61 interviews show that low-income American students whose parents have no formal education limit their use of ChatGPT or Claude to a few specific purposes, held back by five identified barriers.

Making language models accessible is not enough for them to be really used. This is the finding of a study conducted among American students known as FGLI — the first in their families to enter university and from modest homes — based on 61 interviews: 15 long interviews with FGLI students, 3 with other students, 3 with support program managers and 40 short interviews conducted on campus.

These students do use LLMs, but almost exclusively via a chat interface and for narrow needs. There are five obstacles why they don't go further: low perceived value, underestimated confidence in their own abilities, lack of a clear starting point, a few people around them, and limited resources.

The author proposes to think of adoption as a four-level scale — simple chatbot, ready-to-use tooled interfaces, agentic development environments, programmatic integration — and draws four principles from this to design devices that help these students climb the steps.