AI Research for PhD Candidates: What Works and What Doesn't

PhD candidates operate at the boundary of a field's knowledge — they must know the literature comprehensively enough to identify what hasn't been done, and then make a scholarly contribution no one has made before. This creates specific tension with AI research tools: AI tools are trained on existing knowledge and excel at synthesizing what is known, while the PhD requires developing the capacity to push beyond what is known. AI assistance that shortcuts the learning required to do original research is not a tool — it's a liability.

The question for PhD candidates is not "can AI do this task?" but "if AI does this task, what scholarly development am I forgoing?" Some tasks don't build scholarly capability and are legitimate to delegate to AI. Others are central to the development of independent scholarly judgment and should not be delegated, regardless of how capable AI appears to be at them.

Where AI Provides Genuine Legitimate Support

Conceptual Orientation and Initial Literature Mapping

Before beginning systematic literature review in an unfamiliar area, having a high-level map of the intellectual terrain — the major traditions, the key debates, the main scholars — helps orient subsequent reading. AI can provide this general orientation.

Appropriate conceptual orientation uses:

• "What are the major theoretical traditions in [subfield]?"
• "What are the key debates in [research area] as of your training data?"
• "Who are the major scholars in [intellectual tradition]?"
• "What methodological approaches are commonly used in research on [topic]?"
• "What's the relationship between [theoretical framework A] and [theoretical framework B]?"

The critical discipline: AI orientation is a starting point, not a literature review. AI training data has a cutoff date, and the literature in active research areas changes substantially. More importantly, AI will describe what it has "learned" about a field in general terms that may misrepresent specific scholars' positions, elide contested interpretations, or synthesize incompatible positions in ways that create false consensus. AI orientation maps the terrain at a general level; systematic reading is what actually establishes the literature knowledge the dissertation requires.

What this is NOT: AI orientation is not a substitute for systematic literature search (which requires databases with current coverage), not an authoritative account of any scholar's actual position (which requires reading that scholar), and not the field expertise required to identify a genuine research gap (which requires knowing the literature with precision).

Writing Structure and Clarity Review

Academic writing — dissertation chapters, journal articles, conference papers, grant proposals — requires clear argument structure and precise academic prose. AI can assist with structural clarity and prose review.

Appropriate writing assistance:

• "Does the argument in this introduction build logically?"
• "Is the thesis statement clear, or is it too vague?"
• "Is this paragraph arguing one point or trying to do too many things?"
• "Is this sentence clear, or would this alternative phrasing be better?"
• "Does this section transition logically from the previous one?"
• "Is this paragraph doing what the topic sentence says it will do?"

The critical discipline: AI reviews structure and clarity; the PhD candidate maintains complete responsibility for all scholarly content, argument, evidence, and interpretation. AI cannot assess whether the argument is correct, whether the evidence supports the claim, whether the scholarship is cited accurately, or whether the contribution is original. Structural feedback from AI is useful; substantive feedback on scholarship requires human readers — advisors, committee members, peer reviewers.

Grant and Academic Writing Scaffolding

For the formatting conventions of specific academic genres — the structure of an NSF broader impacts statement, the conventions of a particular journal's methods section, the format of a fellowship application's research statement — AI can describe the genre conventions.

Appropriate genre scaffolding:

• "What are the conventions for an NSF GRFP research statement?"
• "What does a strong personal statement for a dissertation fellowship typically include?"
• "How are methods sections typically structured in [specific subfield] articles?"
• "What are the formatting requirements for an abstract in [specific format]?"

The critical discipline: genre conventions are a starting point; the content is the scholar's. Understanding what an NSF GRFP research statement typically includes doesn't relieve the candidate of developing the actual research argument. AI describes what these genres do; the candidate does them.

Administrative and Professional Writing

Fellowship recommendation requests, conference registration communications, professional email correspondence, department administrative tasks — these are legitimate uses of AI writing assistance that don't involve scholarly content.

Appropriate administrative assistance:

• Drafting professional email templates
• Structuring response communications
• Formatting administrative documents
• Travel reimbursement requests and similar administrative writing

Where AI Introduces Serious Quality and Integrity Risks

Literature Search and Citation

This is the most serious AI risk for PhD candidates: AI language models generate plausible-sounding but potentially fabricated citations. This is the single most dangerous misuse of AI in academic research.

What AI does: AI generates scholarly-sounding reference lists with author names, titles, journal names, publication years, and page numbers. Many of these references do not exist. Others exist but the content AI attributes to them is fabricated. The citations appear credible and follow standard formatting.

Why this is serious for PhD candidates:

• A dissertation that cites fabricated sources has a fundamental integrity problem
• A journal submission that cites fabricated sources will be caught in review (reviewers and editors read the sources)
• Academic careers are built on scholarly reputation; fabricated citations are career-ending integrity violations
• AI-generated citations mix real and fabricated sources, requiring verification of every individual citation

The rule: all literature searching is conducted through academic databases with verified coverage — Web of Science, Scopus, PsycINFO, JSTOR, field-specific databases. Every citation in every piece of scholarly writing is verified against the actual source. AI is not used for literature search or citation generation.

Original Scholarly Argument

The dissertation's contribution to the field is the candidate's original scholarly argument — the claim that no one has made before, supported by evidence the candidate has assembled, defended with reasoning the candidate has developed. Using AI to generate the scholarly argument bypasses the intellectual development the PhD is designed to produce.

What's at stake: the scholarly judgment required to develop an original argument — knowing the literature well enough to identify a gap, understanding what counts as evidence in the field, developing the interpretive framework that makes the evidence meaningful, defending the argument against anticipated challenges — is the capability the PhD develops. Using AI to generate this argument means the candidate has credentials but not the capability those credentials represent.

The specific risk: unlike AI-generated citations, which can be caught by checking sources, AI-generated scholarly arguments may not be catchable by external readers but will be apparent to the candidate herself when she cannot explain, defend, or build on the argument she is presenting. In dissertation defense, in job talks, in classroom teaching — the scholar who has not actually developed the argument cannot engage with it as her own.

The rule: the dissertation's scholarly argument is the candidate's intellectual work. AI can be a thinking partner for exploring ideas; it is not a source of the scholarly contribution.

Dissertation Research Itself

The empirical research — data collection, analysis, and interpretation — is the candidate's original scholarly contribution. AI-generated "findings" are not research; they are AI outputs that cannot be defended, replicated, or built upon by the scholarly community.

This includes:

• Having AI "analyze" qualitative data rather than conducting the analysis oneself
• Using AI outputs as substitute findings
• Having AI interpret research results rather than developing interpretations through analytical engagement with the data

Why this matters beyond integrity: the analytical skills developed through conducting and interpreting research — learning to see patterns in qualitative data, developing interpretive frameworks through engagement with materials, understanding what quantitative findings mean and what they don't — are the research skills that a faculty member uses throughout a career. These skills develop through doing, not through reading AI outputs.

Academic Integrity in Coursework and Examinations

Most doctoral programs have explicit academic integrity policies that govern AI use in coursework, qualifying examinations, and other assessed work. These policies vary by program and evolve; candidates are responsible for knowing their specific program's current policies.

Beyond formal policy: the coursework and examination requirements of doctoral programs are designed to develop and assess the knowledge and capabilities the PhD requires. Using AI to complete assessments that are designed to build or assess the candidate's own knowledge and capability undermines the development of those capabilities — and produces a scholar who has the credential but not what the credential is supposed to represent.

The PhD Candidate's AI Protocol

Task	AI Role	Discipline
Conceptual orientation	Starting point only	Verify everything against actual sources
Writing clarity review	Structural and prose feedback	Student maintains all content responsibility
Genre conventions	Format scaffolding	Content is the student's work
Literature search	DO NOT USE	Use academic databases with verified coverage
Citation generation	DO NOT USE	Verify every citation against the actual source
Original argument	Thinking partner only	The argument is the student's scholarly contribution
Dissertation research	DO NOT USE	The empirical work is the student's research
Coursework/examinations	Check program policy	Development of your capabilities is the purpose

What AI Cannot Do That PhD Candidates Need

AI cannot read the literature critically. AI can summarize; it cannot develop a genuine scholarly position on what existing work gets right and wrong.

AI cannot identify a real research gap. Identifying what hasn't been done requires knowing the literature with a precision that AI general knowledge cannot provide.

AI cannot conduct original research. Data collection, fieldwork, experimental design, and analysis require human judgment and human presence that AI cannot provide.

AI cannot develop scholarly judgment. The capacity to distinguish significant from trivial, to identify what an argument requires vs. what is merely interesting, to know when evidence is sufficient — these develop through the practice of scholarship, not through AI assistance.

Key Takeaways

1. AI citation generation is the single most dangerous misuse. All literature search through academic databases; all citations verified against actual sources.

2. The dissertation argument is the candidate's original intellectual work. AI as thinking partner, not argument generator.

3. The empirical research is original scholarly contribution. AI does not conduct, analyze, or interpret research.

4. AI writing assistance is for structure and clarity, not for scholarly content. The scholar is responsible for everything the text argues.

5. The development of scholarly judgment requires doing the scholarship. Capabilities that are shortcut with AI are not developed.

---

Conclusion

AI research tools for PhD candidates: appropriate uses include conceptual orientation and initial literature mapping (starting point only; verify everything against actual sources; AI's general account of a field misrepresents specific positions and may lag current debates), writing structure and clarity review (structural and prose feedback; candidate maintains all scholarly content responsibility), grant and academic writing genre conventions (format scaffolding; content is the candidate's work), and administrative and professional writing. Serious risks arise with literature search and citation (AI generates plausible-sounding but potentially fabricated citations — this is the most dangerous misuse; all literature searching through academic databases with verified coverage; every citation verified against the actual source), original scholarly argument (the dissertation's contribution is the candidate's intellectual work; AI as thinking partner for exploring ideas, not source of the scholarly argument), dissertation research itself (data collection, analysis, and interpretation are the candidate's original scholarly contribution; AI outputs are not research findings), and coursework and examinations (program policies vary; beyond policy, the development of scholarly capabilities requires doing the scholarship). What AI cannot do: read literature critically to develop genuine scholarly positions, identify real research gaps, conduct original research, or develop scholarly judgment — these require the practice of scholarship itself.

See also: Building a Personal Knowledge Base.