Before I started my PhD, I consulted with several individuals on how I should prepare myself. What can I do? How can I put myself in the best position? I was asking anyone and everyone I could for advice. The most prominent advice or comment I heard was, “It’s important to have a healthy relationship with your advisor.”
I took that very seriously. So, of course, I put much effort into curating a healthy relationship with my advisor. I scheduled weekly meetings with her to talk about my research. At times, we would talk not just about research but about other topics such as current events and our respective life experiences.
In one of those meetings, I brought up the topic of how else I could conduct a more in-depth literature review beyond using Google Scholar. My advisor suggested I read books. I was not new to reading books, but I had become so spoiled by instant online access to academic literature that I questioned whether it made sense for me to read dozens of physical books, go to the library, find them, and then worry about returning them before the due date.
Then my advisor shared her experience of doing a PhD several decades ago, at a time when online literature review was nonexistent—let alone rich databases like Google Scholar. She mentioned that there were no web searches and that they had to go to the library and read physical books for every citation they used. This was a moment of reflection for me. The process already felt challenging, even with several tools readily available. My fellow PhD friends and I still found ways to complain about how hard the PhD journey was, despite having instant access to a wealth of information at our fingertips. I couldn’t fathom how I could have survived a PhD without the internet, without the web, and without Google Scholar.
A few months ago, I had an interesting conversation with a friend about the future of AI. We both agreed that AI is going to become as ubiquitous as Google Search is today—if not more so. Then it occurred to me: what if, in the near future, students are paralyzed without AI? It felt like a déjà vu moment.
I started thinking about how my PhD journey could have been different if AI had been as common and accessible as it is today. Or, in other words, if I had started my PhD in the fall of 2025, how would things be different? How would I have conducted my dissertation research knowing that AI was there for me to use?
So, I began reimagining the qualitative research process in the age of generative AI. When I completed my PhD, ChatGPT didn’t exist and “generative AI” was a term mostly known only to computer scientists/engineers. For researchers like me, literature reviews meant hours in the library. Okay, I’ll be frank—it meant hours on Google Scholar. I did my fieldwork in the mountains of Nepal, where people spoke their local dialects. Therefore, I conducted all of the interviews in the local language. I spent thousands of hours transcribing from the local language to Nepali, and then from Nepali to English. It was such a tedious and manual process that I could still feel myself playing back and typing out hours of audio files.
Looking back, I often wonder: how would my research process have changed if I had access to today’s AI tools? Could I have brainstormed more effectively? Synthesized literature more efficiently? Written more clearly? Would it have made the process easier or would it have risked flattening the nuance that defines qualitative inquiry?
In this article, I revisit the core steps of my dissertation process and explore how generative AI tools like ChatGPT, Claude, Elicit, and more specific AI-powered qualitative research tools like Qualz.ai, Atlas.ti, etc. could have been used strategically to support each stage. I also reflect on potential limitations and ethical concerns, especially in light of the responsibilities qualitative researchers carry.
Step 1: Selecting a Research Topic
How AI Could Help?
Every PhD student or anyone who has a PhD knows this is the first step in their research journey. Selecting a research topic is an intimate pursuit. I personally believe it has to be a perfect blend of passion, determination, and other factors, such as approval from one’s advisor. I don’t think AI could have contributed much at this step. However, AI can generate topic ideas based on emerging themes in the field I am interested in. It can help identify underexplored areas or brainstorm directions by synthesizing existing research.
Limitations:
AI lacks my lived experience and personal passion both vital to sustaining a long research project. I would not trust AI to come up with a topic that I am truly passionate about. Additionally, it might suggest overly popular topics, missing those that challenge dominant paradigms or center marginalized voices all of which could contribute to a lack of personal creativity and authenticity.
Step 2: Refining the Research Problem
How AI Could Help?
I believe this is where AI can be very helpful. I still remember trying to find as much information and literature relevant to my research topic. For instance, I spent a lot of time combing through what research had already been conducted, what findings were out there, and what novel contributions I could make in terms of knowledge production as well as practical application. LLMs can assist in narrowing broad ideas into a specific research problem. By simulating literature synthesis or prompting critical questions, they can help identify gaps or tensions worth exploring and save me a lot of time.
Limitations:
AI may offer a wealth of information quickly. However, how much can I trust this wealth of information? It could be surface-level suggestions or lack the theoretical grounding required for high-level scholarly inquiry. Therefore, I would need to validate and verify the outputs, not take them at face value. Human engagement with ambiguity and context is irreplaceable.
Step 3: Choosing a Qualitative Research Design
How AI Could Help?
Similar to the previous step, this is also an area where I can see the application of AI. I probably would have used AI to do a more thorough exploration of the pros and cons of various qualitative designs—phenomenology, ethnography, grounded theory, etc.—based on my research goals. While assessing epistemological fit or offering mentorship-level insights on feasibility and alignment must be done by the researcher, these are areas where AI could help me work more efficiently.
Limitations:
Just like many aspects of a PhD, research design is an intimate and iterative process. There is no one-size-fits-all. It’s a process that improves and becomes more focused through trial and error. Therefore, I believe methodological decisions are not just technical; they are philosophical, grounded in theory, and sometimes contextual depending on the fieldwork and what is practically feasible.
Step 4: Developing Research Questions
How AI Could Help?
I personally believe this was the most challenging, but most important step, at least it was for me coming up with a succinct research question (or questions). Behind every great research project is a powerful question, sometimes just one. Your research question shapes the entire research and PhD journey. This is where I needed to invest my creativity and ideas, and it’s where I needed to be fully engaged without the help of AI. However, after the ideation phase or once I had settled on a research question, AI could help me shape open-ended, exploratory questions that align with my design.
It could also help reframe or sharpen vague inquiries. AI could be very useful in polishing and fine-tuning the research questions I came up with. AI can help develop research questions by analyzing existing literature, identifying gaps, suggesting novel angles, and refining hypotheses based on data trends. It can also generate ideas, optimize wording, and recommend methodologies, thus speeding up parts of the research process.
Limitations:
This is one step where I would be perfectly okay not using AI at all, especially during the exploratory stage. AI might be useful to some extent, but it has significant limitations. It lacks deep contextual understanding, and grounding in theory, and it struggles with true originality. Without human oversight, AI might propose unoriginal or ethically unclear questions. While useful for brainstorming, it cannot replace a researcher’s critical thinking and creativity in developing meaningful research directions. I would be extremely cautious in how I use AI in this step. If not handled carefully, using AI at this stage could reinforce dominant framings rather than help you challenge or reimagine them.
Step 5: Conducting a Literature Review
How AI Could Help?
This is where I think I would have used AI the most. I spent so much time combing through thousands of academic journals. I devoted significant time just trying to find the appropriate, fitting, or relevant articles. I think this is where AI would have been most useful. AI-powered qualitative research platforms can significantly streamline the literature review process by efficiently scanning academic databases to retrieve relevant publications, using smart keyword optimization to ensure comprehensive coverage.
AI can summarize key findings from multiple papers, allowing me to quickly grasp essential concepts without reading every document in full. Every researcher, PhD scholar, or student has done so anyway. In addition, AI can also be extremely useful in identifying research gaps by analyzing trends, contradictions, and underexplored areas across thousands of studies. It can also organize references, suggest related works, and even generate visual concept maps to reveal connections between different research strands. It can save so much time without really compromising my critical thinking and creativity.
Limitations:
As useful as AI is for literature review, these capabilities come with important limitations. AI may struggle with nuanced interpretation of texts, could inherit biases from its training data, and might miss very recent or niche studies that aren’t well-represented in digital repositories. About two years ago, I remember AI citing an article that did not exist. AI can “hallucinate” sources or misrepresent findings. While AI dramatically accelerates the literature review process, human oversight remains crucial for critical evaluation, contextual understanding, and ensuring the quality and relevance of selected sources. Researchers should view AI as a powerful assistant that complements, rather than replaces, scholarly expertise and judgment. It doesn’t replace deep, critical engagement with the texts or the ability to evaluate methodological quality.
Step 6: Navigating IRB and Ethical Review
How AI Could Help?
IRB is crucial if the research involves human subjects. Anyone who has been through the IRB approval process knows that it is a tedious process. I would think deeply about where I could use AI in this step. AI can provide valuable assistance in navigating the IRB process by automating certain administrative tasks. It can help draft clearer consent forms, suggest standard protocol language, and flag common ethical considerations based on previous studies—potentially reducing the time spent on revisions. AI tools may also simulate how risks to participants could be framed, offering researchers a starting point for their submissions. It can also generate sample IRB language or protocols for similar studies.
Limitations:
However, IRB is also where the application of AI in the research process will be scrutinized heavily. Universities are still developing policies on AI use in research – IRB committees may examine AI-generated content carefully. While AI can streamline parts of the process, researchers must remain actively engaged in ethical decision-making, ensuring their submissions meet regulatory, research, and moral standards. Also, ethical review is deeply contextual. AI cannot assess your relationship with participants, positionality, or power dynamics. Nor can it make decisions about risk, vulnerability, or cultural appropriateness. IRBs also require human judgment grounded in local norms and values.
Step 7: Data Collection Methods
How AI Could Help?
My fieldwork was in the remote Himalayan region of Nepal where the locals only spoke their dialect and there was no internet. So, I had to conduct my surveys and interviews in the local language and travel to the field. I could not have used AI as much as I would have if internet access, language, and geography weren’t limiting factors. That said, I can see AI being useful in helping create interview guides, pilot-testing questions, and adjusting language for clarity or cultural fit.
Where applicable, AI can significantly streamline data collection for surveys and interviews by automating key tasks and enhancing efficiency. For surveys, it can help design well-structured questions, minimize bias, and optimize the flow to improve response rates. During interviews, AI can transcribe recordings in real-time. Additionally, AI can improve data quality by detecting inconsistencies, fraudulent responses, or patterns like straight-lining in surveys.
Limitations:
The advantages, however, come with some limitations. I could not have used AI for this step as much, but AI may inadvertently perpetuate biases present in its training data, particularly when dealing with diverse cultural or linguistic contexts. It also struggles with interpreting subtle human elements like tone, sarcasm, or emotional nuance in interviews, which often require human intuition. Over-reliance on automation risks making interactions feel impersonal, potentially affecting participant engagement and data richness. Ethical concerns around data privacy and consent further underscore the need for careful oversight. Therefore, AI is most effective when used as a supportive tool—handling repetitive tasks like transcription. Using AI to simulate participants risks reducing real voices to synthetic proxies. Ethical, relational, and cultural dynamics must be handled by the researcher—not outsourced.
Step 8: Data Analysis
How AI Could Help?
As mentioned earlier, my dissertation required me to go to a remote mountainous region of Nepal for fieldwork. However, once I had the final transcription ready (after thousands of hours of translating and transcription), I could see myself using AI in this step. I used tools like NVivo and Atlas.ti, but they were expensive for students and didn’t necessarily do much beyond helping me organize codes. However, with advanced AI tools specifically tailored for qualitative research like Qualz.ai, I see myself using AI for certain aspects of data analysis.
AI has the potential to offer powerful support by automating labor-intensive tasks like transcription, coding, and theme identification. It’s not only useful for accelerating analysis but also introduces consistency in coding, reducing potential human bias in initial data organization. AI can help with initial coding, theme identification, and clustering of similar responses. Initial is the keyword here. It can reduce time spent organizing data and even assist in visualizing patterns, so I could use more of my time exploring data in-depth.
Limitations:
There are, however, several limitations. These capabilities come with significant constraints. AI systems often fail to grasp the depth of human communication—missing cultural nuances, metaphorical language, or situational context that qualitative researchers carefully interpret. They may oversimplify rich narratives by forcing them into predetermined categories, and their outputs can reflect biases embedded in their training data, particularly when analyzing diverse dialects or marginalized perspectives.
Qualitative data requires interpretation. AI can’t understand irony, cultural nuance, or unspoken subtext. Over-automation risks treating rich narratives like survey data. So, researchers have to remain aware and use AI to the extent that they, the researchers, are making the crucial decisions—maintaining control and staying in the loop throughout the process. AI serves best as a preliminary processing tool, handling repetitive tasks to free researchers for the essential work of deep interpretation, contextual understanding, and theoretical development that defines rigorous qualitative research. Researchers should view AI outputs as starting points for further critical examination rather than definitive conclusions.
Step 9: Writing the Dissertation
How AI Could Help?
This is where I struggled the most. After the analysis was done and all that was left was writing, I often felt like I was stuck. Most of the time, it was just a lack of motivation as I was inching closer to the finish line. AI could have significantly supported the dissertation writing process by assisting with various stages of research and composition. Writing includes sections for literature reviews, organizing references, and generating properly formatted citations—areas where I could have used AI to save time. Additionally, AI can provide structural suggestions, improve grammar and clarity, and ensure an academic tone. For non-native English speakers like me, AI can be particularly valuable in polishing language while preserving the original meaning.
Limitations:
Writing is also thinking. Overusing AI can distance you from your voice, insights, and arguments. There are also risks of unintentional plagiarism if boundaries aren’t clear. One has to be very cautious—AI has real limits. It might give you generic arguments or shallow analysis that won’t impress your committee. Worse, it could risk originality, and some schools might even flag its use as inappropriate assistance. Remember, AI can’t replicate your scholarly voice, deep critical thinking, or the innovative ideas that make a dissertation truly yours. Since universities are still figuring out AI policies, always check with your advisor. Use AI as a tool, not a crutch. The hard work-original insights, rigorous analysis, and meaningful contributions are all reflected in your writing,and have to come from you. Stay in control, and let AI support (not replace) your unique academic journey.
Step 10: Reviewing and Editing
How AI Could Help?
Oh, reviewing and editing—the rounds and rounds of iterations. This is where AI can be very useful and save a ton of time with the least trade-offs. AI can check for clarity, cohesion, and grammar. It can simulate a “reviewer’s eye” or help you polish language for readability. AI can significantly improve your dissertation’s clarity and correctness by catching grammar errors, refining awkward phrasing, and ensuring consistent terminology. It helps maintain an academic tone and can even check citation formatting.
Limitations:
Once again, I will repeat AI has limitations and it’s not Sherlock Holmes, by the way. Don’t take AI’s suggestions at face value. It may miss field-specific nuances or suggest edits that alter your intended meaning. I think the most effective and efficient way would be to use AI as a smart proofreader but always review its suggestions critically. Your expertise should guide the final decisions to preserve your unique scholarly voice.
Doing my PhD without generative AI taught me how to sit with complexity, wrestle with ambiguity, and stay deeply connected to my data. These are skills no tool can replace. But it also took a lot of time, energy, and sometimes money—and it made the journey more stressful than it needed to be, at times. In hindsight, these challenges shaped who I am today. However, AI is here to stay.
I believe we can benefit enormously if we use AI wisely—and face serious consequences if we misuse it. I now see how AI could have made parts of the research process less isolating and more efficient—from literature synthesis to writing support. For future students, I offer this not as a prescription but as a provocation: use AI strategically, ethically, and reflexively. Let it support, not replace, your critical thinking.
And always ask yourself: What does this tool see and what might it miss?
Disclaimer: I am an AI enthusiast and believe in the potential of AI. I also believe we need to test and question AI constantly—to understand where it thrives and where it fails. This article is based solely on my own experience. It may not apply to PhD students with different research contexts, methodologies, field sites, or levels of technological access. I hope that others may find some value in this reflection on what could have been—and what might still be.