'The evidence base is almost non-existent': the university buys AI before knowing if it helps to learn

OpenAI claims to have sold at least 700,000 licenses to institutions. Educational researchers, on the other hand, struggle to produce any solid studies on the effects of learning.

Higher education institutions sign large contracts with AI suppliers without independent research on the pedagogical effectiveness of these tools. OpenAI claims at least 700,000 licenses sold to institutions, including California State University, Arizona State University, and Maine University. CSU signed a contract in 2025 of $17 million.

Justin Reich, who runs MIT's Teaching Systems Lab, summarizes the situation with one sentence: the evidence base is almost non-existent. The problem is not only the lack of resources — the federal government has emptied the Institute for Education Sciences of its substance — but also the pace of the sector. Stacey Alicea of the Research Partnership for Professional Learning points out that the models change every one to three months, making the classic research protocols obsolete even before the results are published.

When work exists, their quality is concerned: Patrick O'Neill, of Ivy Tech Community College, noted in several peer-reviewed articles advocating the effectiveness of AI of poor methodologies and poorly used statistics. More than a thousand tools are now being marketed to students. Carly Robinson, of Stanford's Systems Change initiative, does not close the door — AI can improve learning — but believes that we are still far from it.