Are doctoral students using ChatGPT to think harder or to think less?

University of Phoenix

When a doctoral student sits down to review 200 papers for a dissertation literature review, the temptation to ask an AI tool for help is no longer hypothetical. It is happening in departments across the country, sometimes with institutional blessing, sometimes without clear policies, and almost always faster than universities can draft guidelines.

But does this speed come at the cost of the deep reading and critical synthesis that doctoral education is supposed to cultivate? A new scoping review by three scholars at the University of Phoenix's College of Doctoral Studies attempts to map what we currently know - and do not know - about generative AI's role in academic work.

What the literature actually shows

Patricia Akojie, Marlene Blake, and Louise Underdahl reviewed the existing scholarly literature on academic applications of generative AI tools, focusing on their use in doctoral research, academic writing, literature review processes, and knowledge development. Their article, published in the International Journal of Digital Society, uses a scoping review methodology - a systematic approach designed to map the breadth of available evidence rather than test a specific hypothesis.

The patterns they found are consistent with what many academics have observed informally. Generative AI tools, ChatGPT chief among them, are being used for literature review support, research brainstorming, and writing assistance. Students report that AI can help them synthesize large bodies of literature more quickly, identify connections between papers they might have missed, and overcome the blank-page paralysis that afflicts writers at every level.

The efficiency gains are real. A task that once consumed weeks of reading and note-taking can now be substantially accelerated. For complex scholarly work like synthesizing hundreds of sources across multiple disciplines, AI tools offer a kind of intellectual scaffolding that did not previously exist.

The integrity question nobody has resolved

The more difficult finding is what the literature reveals about academic integrity. The review identifies a significant gap between the pace of AI adoption and the development of institutional policies governing its use. Many universities have not established clear guidelines about when and how students may use generative AI in their research. Transparency requirements - disclosing that AI tools contributed to a literature review or helped draft sections of text - remain inconsistent across institutions and journals.

The core tension is straightforward. Doctoral education exists to develop independent scholarly thinking. A dissertation is supposed to demonstrate that the candidate can identify a research question, locate and evaluate relevant evidence, synthesize findings, and produce original analysis. If AI tools perform substantial portions of these tasks, the question of what the degree certifies becomes genuinely complicated.

Akojie, the lead author, framed the issue in terms of integration rather than prohibition. The review suggests that the goal should be helping researchers understand both the capabilities and limitations of generative AI technologies, not banning them outright. This aligns with the position most major research universities have adopted publicly, though implementation varies wildly.

What AI does well and where it fabricates

The review's findings on AI capabilities deserve careful parsing. Generative AI tools excel at certain academic tasks: identifying relevant literature from large databases, suggesting organizational structures for reviews, generating draft text that a researcher can then revise, and brainstorming research questions or methodological approaches.

They are notably poor at others. AI tools can fabricate citations - generating plausible-looking references to papers that do not exist. They can produce confident-sounding analysis that contains fundamental errors. They cannot evaluate the quality of a study's methodology or assess whether a finding is likely to replicate. They have no way to judge whether a claim is controversial within a field or represents settled consensus.

For a doctoral student, these limitations are especially dangerous precisely because the student may lack the expertise to catch them. A senior researcher can immediately spot a fabricated citation or a mischaracterized finding. A second-year doctoral student, still building domain knowledge, may not.

The training gap in doctoral programs

The review identifies AI literacy training as a critical need. Most doctoral programs have not incorporated formal instruction on how to use generative AI tools responsibly, how to verify AI-generated content, or how to maintain scholarly rigor when AI assists with portions of the research process.

This gap is notable because doctoral programs routinely teach other methodological skills - statistical analysis, qualitative coding, systematic review protocols - with substantial rigor. The tools that students use to conduct research are typically subjects of formal instruction. Generative AI, which is arguably reshaping research workflows more rapidly than any tool since the internet, has largely been left to students to figure out on their own.

The authors suggest that institutions need clearer policies and guidance to support responsible AI adoption in both research and teaching. This recommendation is sensible but understates the difficulty. Writing policies for a technology whose capabilities change every few months is inherently challenging. Guidelines drafted today may be outdated before the academic year ends.

A scoping review's inherent limits

The study itself has important limitations to acknowledge. A scoping review maps the breadth of a literature rather than assessing the quality of individual studies or drawing quantitative conclusions. The findings reflect what has been published on the topic, which is heavily weighted toward opinion pieces, early case studies, and policy discussions rather than rigorous empirical research with control groups and measurable outcomes.

The field is also moving so fast that any review of the literature risks being dated by the time it appears in print. Tools that existed when the review was conducted may have been superseded. Institutional policies may have evolved. Student behavior may have shifted in response to both improving technology and emerging norms.

The research comes from scholars at the University of Phoenix, an institution with a specific focus on working adult learners and online education. Whether the patterns observed in that context generalize to research-intensive doctoral programs at traditional universities is an open question.

The question that remains

The deeper issue that this review surfaces but cannot resolve is whether generative AI in doctoral education represents a productivity tool or a fundamental change in what it means to do scholarly work. A calculator does not undermine mathematical understanding if the student first learns to do arithmetic by hand. Does the same logic apply to AI-assisted literature reviews?

The honest answer is that we do not know yet. The empirical research needed to answer that question - tracking long-term scholarly output, critical thinking development, and career outcomes for students who used AI tools heavily versus those who did not - has not been conducted and will take years to mature. In the meantime, the tools are being adopted regardless, and the frameworks for their responsible use are still being written.