Science1 publisher3 min readPublished
Higher GPA predicted heavier generative-AI use in a survey of 491 online adult learners
The University of Phoenix study treated generative-AI use as the outcome and students' grades as a predictor, so it describes who picks up the tools. The authors say it cannot show what the tools do to a grade.
The Scientist · Science desk
What happened
- University of Phoenix researchers surveyed 491 nontraditional undergraduates enrolled in online general education courses about their use of generative AI in coursework.
- Of those, 212 students, or 43.2%, said they had used tools such as ChatGPT, Microsoft Copilot or Gemini for their classes.
- A multiple regression found two statistically significant predictors of how often a student used the tools: a higher GPA and more confidence in their own technology skills.
- Confidence in reading, writing and math showed no significant relationship with use frequency, and neither did hours spent on work, study or family responsibilities.
Compiled by The ScientistSomething wrong?How this is made
Why it matters
- constraint Because the model puts use on the outcome side and grades on the predictor side, it cannot support a claim about learning gains; anyone citing it to argue AI raised student achievement is citing the wrong result.
- decision Curriculum committees choosing between blanket restriction and structured guidance now have an argument from the authors, but it rests on adult learners at one online university, so a residential campus would have to measure its own distribution before borrowing the conclusion.
- exposure The measurement depends on students voluntarily admitting behaviour they said they were unsure their integrity policy permitted. The most cautious students run the greatest risk of being counted as non-users.
GPA and confidence in technology skills sit on the predictor side, and frequency of generative-AI use is what the regression tries to explain [4]. So the study describes who reaches for the tools and does not measure what happens to a student's coursework after they start. The authors are explicit that the design is cross-sectional and correlational, drawn from a nonprobability sample at one online university, and does not establish that GenAI use causes stronger academic performance [9].
The paper, "Strategic Integration or Skill Compensation? Understanding GenAI Use in Online Higher Education," appeared June 30 in Artificial Intelligence in Education [1]. Inside the group that reported use, the grades run high: 91.1% had a GPA of 3.0 or better [6], roughly 193 students [2], and 68.9% earned a 4.0 in the course where they reported using the tools [7], about 146 students [3]. Subtracting the users from the full sample leaves 279 who said they did not use GenAI [1]. The university's summary gives the direction of each predictor and the grade shares for users; it does not include coefficients, confidence intervals, or the same grade distribution for those 279. The comparison that would make 91.1% meaningful is the corresponding share among non-users.
Use was self-reported, and it was self-reported about a behaviour some of the same students described as unclear under their institution's academic integrity policies [8]. A student who suspects a tool is prohibited has a reason to say they did not use it. If that reluctance is concentrated among students doing less well, part of the association between GPA and reported use is an artifact of who was willing to say so.
Jessica Sylvester, senior manager of College Operations at University of Phoenix and the study's lead author [14], said the results "complicate the assumption that students primarily turn to generative AI because they are struggling academically or simply looking to save time" [10]. She also said the pattern "points to an important opportunity for higher education to help all learners develop the digital fluency, critical judgment and ethical awareness needed to engage with AI effectively" [11].
The open-ended answers describe how the 212 users worked. The three most common themes were clarifying complex concepts, using the tool as a "thinking partner" for brainstorming and organizing ideas, and getting academic tasks done faster [12]. The same students raised the accuracy and originality of AI-generated content, and the preservation of their own academic voice, as worries [8].
On that basis the researchers recommend moving away from restriction and toward guidance that builds GenAI literacy, critical evaluation and transparent, ethical use, including AI literacy taught across disciplines and assignments that ask students to critique and reflect on AI-generated information [13].
What to watch
- Whether the full paper reports coefficients and confidence intervals for the two non-significant predictors, which would separate a small effect from an imprecise one.
- A longitudinal study of the kind the authors recommend, following the same students across adoption instead of measuring use and grades at one moment.
- Whether anyone publishes the GPA distribution for the 279 students who said they did not use the tools.