Research Article
Impact of ChatGPT on High School Grade Point Average: A Comparative Study by Subject, School Type, and Socioeconomic Factors
Synthesis: Cho (2025) reviews studies linking ChatGPT use to high school GPA in essay writing and mathematics, then compares it by school type, household income, parental education, and teacher and peer context. The reviewed estimates are consistently positive — 0.31 GPA points for weekly writing use (Kim et al., 2024) and 0.18 for mathematics homework in a randomized trial (Micabalo et al., 2024) — but the pattern is distributional. Private school users out-gained public school users (0.38 against 0.28 in writing), students above median income out-gained lower-income students (0.42 against 0.15 over three months), and children of college-educated parents out-gained children of parents who finished high school. Cho calls this a "triple divide" of institutional, socioeconomic, and pedagogical readiness: generative AI widens gaps unless infrastructure, training, and AI literacy are funded together.
Key Findings
- Writing gains reached the transcript, not just the draft. Students using ChatGPT on at least two writing assignments a week for a semester gained 0.31 GPA points, mostly among weaker writers (Kim et al., 2024).
- Mathematics moved less than writing. In a randomized trial of 1,200 students from 20 schools, ChatGPT in mathematics homework raised GPA by 0.18 points, mostly in algebra and geometry and least in advanced placement courses (Micabalo et al., 2024).
- The same tool paid off more in better-resourced schools. Private school students who used ChatGPT regularly gained 0.38 points in writing and 0.23 in mathematics, against 0.28 and 0.15 for public school users.
- Household income separated users by a wide margin. In the OECD pilot survey of 7,800 students in 9 countries, above-median-income students gained 0.42 GPA points after three months of use, versus 0.15 for low-income students, who more often misread AI feedback.
- Parental education predicted how the tool was used. Students with a bachelor's-educated parent gained 0.34 in writing and 0.19 in mathematics; those whose parents finished high school gained 0.21 and 0.11, accepting answers without challenge (Ni et al., 2025).
- Teacher readiness tracked with outcomes. Schools running regular AI workshops and incentives reported gains of 0.29 points, against 0.12 where integration was optional or unregulated (UNESCO, 2025).
- The gain came with a cost to originality. Coherence and clarity rose in AI-assisted essays while originality fell slightly, and plagiarism detectors missed AI-generated text (Khalifa and Albadawy, 2024).
Writing and Mathematics Reward Different Kinds of Help
Essay writing fits most smoothly into existing instruction: ChatGPT scaffolds brainstorming, outlines, transitions, vocabulary, and grammar, and critiques drafts before a teacher sees them. The gains ran through a live teacher loop of feedback and iterative revision, not generation alone.
Mathematics is harder because it demands symbolic reasoning and multi-step solutions. Step-by-step explanations help students who struggle with notation, and explaining reasoning back to the tool gives a metacognitive workout that mirrors peer tutoring. The risk is concrete: it occasionally supplies incorrect steps or misapplies formulas, so students who do not cross-check learn wrong procedures confidently; the schools described pair it with error checking such as Wolfram Alpha or Desmos.
Institutional and Family Readiness Set the Return
The clearest institutional finding is the private–public gap. In a cross-sectional analysis of 3,400 students in private schools across 15 US states, AI usage raised average GPA by 12%, and Cho credits the advantage less to the tool than to what surrounded it: workshops on ethical use, AI coordinators, collaboration, and bandwidth. Public schools are described as short on infrastructure and training, producing unsupervised use and more dishonesty. Where public systems invested, results followed: the "AI for All" initiative in Texas is credited with a 17% gain in writing performance.
The socioeconomic findings point at access but land on interpretation: they predict whether a student can judge which suggestion to keep and notice an error. Students without that support used the tool more passively, and unmentored bootcamps or subsidized devices were judged unlikely to close the gap. That is a differential effect across learner groups: identical access produced unequal learning.
Peers, Guardrails, and the Risk of Overreliance
Peer norms shape adoption as much as policy: students take up ChatGPT when they see peers using it and fear being left out, and younger students lean more on peer signals. Structured peer work cuts the other way: a peer assessment cycle built around ChatGPT in STEM courses improved knowledge construction, critical thinking, problem solving, and creativity (Wu et al., 2024).
The counterweight is overreliance. Cho distinguishes revising with the tool from substituting for it, tying the second to formulaic writing and shallow understanding that does not survive the course. Because detection is unreliable, the review argues for assessment that asks for process and voice, and for integrity rules that describe permitted use.
What this means for practice
- Instructors: Keep the human revision loop around AI drafts; a ChatGPT pass alone raised polish without raising learning. In mathematics, add a checking step before answers enter notes.
- Administrators: Fund what the private-school advantage rested on — teacher training, an AI coordinator role, bandwidth, and devices — before buying licenses.
- Designers: Build prompts that ask students to explain, predict, or check rather than accept; the metacognitive pattern is the one tied to durable gains.
- Policymakers: Treat AI literacy as an equity measure: interpretation skill, not device distribution, drove the income gap.
Limitations
- This is a narrative review with no primary data and no reported search strategy; every GPA figure is borrowed from studies that differ in design, sample, and outcome measure.
- Most GPA comparisons are between users and non-users within existing schools. Only the Micabalo et al. (2024) study is described as randomized, so the other gaps are associations, not causal effects.
- Several central figures rest on one secondary source (Ma, 2021) or on agency reports (OECD 2024; UNESCO 2025) whose sampling is undescribed, leaving them unverifiable.
- GPA cannot separate genuine learning gains from completion effects or grading leniency, and nothing is measured about retention or transfer.
Citation
Cho, Y. (2025). Impact of ChatGPT on High School Grade Point Average: A Comparative Study by Subject, School Type, and Socioeconomic Factors. Educational Innovations and Emerging Technologies, 5(2), 22–28.