Stanford finds 47% of students unsure whether they are allowed to use AI

The teacher who gets asked whether AI is allowed on an essay has probably not been told either. Among teachers working inside schools that do have a policy, 6 percent describe it as clear ^[3], and half of American middle and high schools have no policy at all ^[2]. Multiply the two and about three teachers in a hundred work somewhere with a rule they would call clear ^[1]. Those figures come from different survey populations in Stanford's 423-page report ^[17], so three percent is an order of magnitude.

Students already want it. The 47 percent who were unsure of the rules sit inside a population where four in five already use AI for schoolwork ^[3], so the uncertainty puts usage where the institution cannot see it. What the uncertain student did next is outside what Stanford measured.

What they use it for is dull. American high schoolers reach for it to find sources (51 percent), to edit or revise essays (50 percent) and to brainstorm (50 percent) ^[5]. University students worldwide report understanding concepts (56 percent), research (52 percent), generating ideas (46 percent) and editing essays (38 percent) ^[6]. The Next Web, summarising the chapter, writes that none of that is the plagiarism scenario schools spent three years writing policies about, and that most of it is invisible to the tools built to detect it ^[7].

Essay editing is the one use with outcome evidence attached, and it points the wrong way for anyone shipping revision as the default mode. MIT researchers found assisted writers showing lower brain engagement and performing worse once the tool was taken away ^[8].

The buildable surface here is narrow: a permission state the instructor sets per assignment, visible to the student at the point of use, with a record of what the tool did. The Stanford data does not show that shipping that surface lifts adoption or improves results; it sizes the population currently guessing ^[4].

The same industry that asks students to check their school's rules also ships things schools cannot decline. A child safety group judged the AI summaries in Google Search an unacceptable risk to students, and schools cannot switch them off ^[11]. US state-level guidance recognises that AI teacher training matters and then provides neither standards nor funding for the programmes ^[12], and Google, Microsoft and OpenAI have built part of their American classroom position through teacher training schemes and industry-written courses ^[13].

The heaviest use sits well away from the places writing the rules. The UK and the US both sit at 67 percent of university students surveyed in 2025, the bottom of the same table ^[10], 28 points below Indonesia ^[2]. Year-on-year increases since 2023 have ranged between 20 and 61 percentage points depending on the country ^[18]. China requires AI instruction with minimum hours and grade-level progression in Beijing, Guangdong and Hangzhou from the 2025-26 school year, and the UAE mandated it across all grades from the same date ^[14].

Go through your own accounts. Does the institution have a written rule, and can your product show a student what that rule is where the work happens. Where both answers are no, your defaults are the policy, and the principal is the one who defends them. Where both are yes, the win is still revocable: South Korea launched AI textbooks in primary schools in March 2025 and reversed the policy after opposition from parents and teachers ^[15]. Nothing in the report suggests the reversal was about the technology working badly ^[16].