The Impact of ChatGPT on Higher Education: A Systematic Review of Global Opportunities, Perceptions, and Challenges

Synthesis: This systematic review pools 82 empirical studies published between January 1, 2022 and October 20, 2025 to map the opportunities, perceptions, and challenges of ChatGPT in higher education. Following PRISMA guidelines, the authors screened 1046 database records in Covidence, retained 58 studies, then added 24 more through targeted Google Scholar and Consensus App searches. Two independent reviewers agreed on 58 of 63 articles, a 92% agreement rate at Cohen's κ of 0.82. Thematic content analysis organized the evidence around accuracy and reliability, ethical implications, and the forecast future role of ChatGPT in traditional teaching. Perceptions of accuracy split across 13 positive, 7 negative, and 8 mixed studies. Students and faculty reported privacy anxieties, bias concerns, and integrity worries alongside genuine enthusiasm for content creation, personalized support, collaboration, and feedback. The reviewers conclude that responsible, equitable integration depends less on the technology than on clear institutional policy, disclosure norms, verification habits, and AI Literacy instruction.

Key Findings

• The review synthesized 82 empirical studies from January 1, 2022 to October 20, 2025, sourced from five databases plus targeted searches; inter-rater agreement was 92% (κ = 0.82).
• Perceptions of ChatGPT's accuracy and reliability were divided: 13 studies reported positive views, 7 negative views, and 8 mixed views among students and faculty.
• Privacy was the most consistent ethical concern: 83% of faculty members reported concern about privacy and ethical use, and 80.5% of undergraduates would not discuss personal matters with ChatGPT.
• Reliability doubts were widespread: 73.4% of students agreed that ChatGPT raises uncertainty about the reliability of information provided, citing sporadic errors and limited contextual awareness.
• Positive evidence appeared too: up to 91% accuracy on world history items in university admission tests, and 48.2% agreement among Egyptian medical students that ChatGPT is reliable.
• Faculty and students forecast a wider instructional role across five themes: teaching and learning, content creation, personalized learning, creative capabilities, and collaboration and interaction.

How the review was conducted

The authors followed PRISMA 2020 and searched APA PsycInfo, CINAHL Ultimate, Education Source Ultimate, ERIC, and Web of Science on April 27, 2024, pairing generative AI and ChatGPT terms with higher education vocabulary. From 1046 records, 206 duplicates were removed, leaving 840 for screening; 738 failed the criteria and 102 full texts were assessed. Forty-four were excluded, most often for measuring outcomes unrelated to perceptions, attitudes, or experiences, leaving 58 studies. A targeted search added 24 studies from Google Scholar and the Consensus App, and an October 20, 2025 search added 12 more on newer model versions, reaching 82. Synthesis then applied Creswell's open, axial, and selective coding.

Accuracy and reliability: a divided literature

The review found no consensus on whether ChatGPT produces trustworthy academic output. Thirteen studies reported positive perceptions, including medical students and clinicians who rated its responses trustworthy and helpful, and first-year students in Germany and Switzerland whose AI competence predicted acceptance of ChatGPT as a reliable tool. One cross-sectional study of university admission tests reported an accuracy rate of up to 91% on world history items. Seven studies reported negative perceptions: students flagged an inability to assess source quality, cite accurately, and replace words idiomatically, and a majority agreed generative systems produce factually inaccurate outputs. Eight studies reported mixed views, pairing appreciation for idea generation with caution about accuracy; several noted that inaccuracies eroded trust. The reviewers link over-reliance to possible declines in critical thinking and independent learning.

Ethics: privacy, bias, and academic integrity

Ethical concerns clustered into Privacy, bias and fairness, and Academic Integrity. Undergraduates were unwilling to supply personal information because of confidentiality worries, students in Hong Kong were wary of AI data collection, and 83% of faculty members named privacy as a concern for their students. On bias, training data may embed societal biases that shape responses, prompting calls for universities to communicate openly about algorithms and limitations. The reviewers argue fairness depends on institutional conditions too, such as inequitable access, limited infrastructure, and underdeveloped policy environments, connecting to Equity. On integrity, the fluency and speed of ChatGPT raised concerns about plagiarism and AI-assisted cheating, and some studies feared over-reliance would weaken critical thinking and problem solving. These concerns have pushed institutions toward written guidance and Educational AI Policy.

Forecast roles and institutional response

Asked how ChatGPT might reshape conventional teaching, faculty and students described a transformative rather than incremental shift. Five themes emerged: teaching and learning, content creation, personalized learning, creative capabilities, and collaboration and interaction. Faculty in one survey rated ChatGPT and other large language models a powerful instructional tool (M = 4.55, SD = 0.965), and other studies argued chatbot tutors could act as online instructors, curriculum developers, and markers. Students valued detailed explanations, personalized feedback, grammar suggestions, and real-time assistance. Reported use among medical and pharmacy students stayed under half: 44.5% for drug information, 38.9% for homework, and 39.3% for writing research articles. Studies cautioned that declining critical thinking and risks from AI-assisted grading require institutional safeguards.

What this means for practice

• Publish clear course and university rules on when and how ChatGPT may be used, stating the rationale when use is restricted.
• Require disclosure of the AI tool and the model or version used so assessment decisions remain defensible.
• Teach AI Literacy directly: bias identification, privacy awareness, citation of AI-generated content, fact-checking, and the limits of generated output.
• Verify AI outputs against learning objectives and keep AI supplementary, so it enhances rather than replaces critical thinking.

Limitations

• The review covers only January 1, 2022 to October 2025; GPT-5 appeared inside the window but had no included classroom evaluations.
• Included studies concentrate heavily in Asia, limiting generalization to South America, Africa, and Oceania, where empirical work on generative AI remains limited or early.
• Descriptive statistics and thematic content analysis identify trends but cannot establish the magnitude, direction, or statistical significance of relationships.
• The review used no guiding theoretical framework such as the Technology Acceptance Model or TPACK, and examined only ChatGPT, excluding Gemini, Claude, Copilot, and Perplexity.

Citation

Apata, Olukayode Emmanuel; Kwok, Oi-Man; Ajose, Segun Timothy. (2026). The Impact of ChatGPT on Higher Education: A Systematic Review of Global Opportunities, Perceptions, and Challenges. Journal of Computer Assisted Learning, 42, e70309. https://doi.org/10.1002/jcal.70309