The source and why it matters
This review examines Girard et al. (2025), "Learning how to develop a research question throughout the PhD process: training challenges, objectives, and scaffolds drawn from doctoral programs for students and their supervisors". The article appeared in Higher Education (volume 89, pages 1001 to 1020) after online publication in July 2024, and it is open access.
The authors begin with a tension in doctoral education. Programmes have placed growing weight on generic, transferable skills that support employability. Meanwhile the central research skill of framing a question, which the authors call key to original research and hard to acquire, is still mostly passed on informally from a senior researcher to a novice. Girard and colleagues set out to provide a framework for designing programmes that help doctoral candidates, guided by their supervisors, to develop their research questions (Girard et al., 2025).
Why does this matter? The research question shapes the rest of a thesis: it decides which literature is relevant, which design and data fit, and what contribution can be claimed. Most existing guidance, however, either offers generic advice in methods textbooks or supplies tools for particular fields, such as the PICO structure for clinical questions (Richardson et al., 1995) and the FINER criteria of feasibility, interest, novelty, ethics and relevance (Hulley et al., 2013). The paper asks something different: how can the capacity to develop questions be taught?
What the paper proposes
The framework comes from two doctoral training programmes at INRAE, the French national research institute for agriculture, food and the environment. The authors helped to design and teach both over more than a decade. One course served doctoral students within a single research division, who could attend at several stages of their thesis. The other brought students and their supervisors together from across the institute at around the end of the first year. In each case the teaching relied on dialogue with peers, whether students, supervisors or trainers, drawn from varied scientific backgrounds (Girard et al., 2025).
The framework links three learning challenges to training objectives, and the authors illustrate each with scaffolds used in the courses.
| Learning challenge | Examples of training objectives | Examples of scaffolds |
|---|---|---|
| Empowering doctoral students to develop their question, guided by supervisors | Students understand what is expected of PhD research; students and supervisors clarify their respective roles | Students present before supervisors respond; mapping who is involved in supervision |
| Becoming aware of diverse research approaches and positioning oneself among them | Participants respect different research stances and can explain their project to other disciplines | Mixed-discipline groups; presenting the framing and funding context of the project |
| Expressing and reformulating the question throughout the PhD | Students articulate the academic and societal stakes of their work as it evolves | Comparing the original and current formulations of the thesis subject and explaining the changes |
Two features set the approach apart, in the authors' account. It favours speaking and argument over writing, and it treats contact with heterogeneous peers, along with the personal side of doctoral work, as a resource for learning. The authors call the resulting support distributed scaffolding, because it mixes peer, one-to-one and media-based help. The idea of scaffolding goes back to Wood et al. (1976), who used it for the way a tutor takes charge of the parts of a task a learner cannot yet manage, so that the learner can focus on, and finish, the parts within reach.
Two course practices show the approach at work. Supervisors attend, but sessions are arranged so that the doctoral student speaks first while supervisors listen before responding, which moves ownership of the question towards the student. Groups also mix disciplines and thesis stages on purpose, so participants must explain the logic of their question to people who do not share their assumptions. The authors suggest that much of the learning happens here, in having to argue for a question rather than merely state it (Girard et al., 2025).
Strengths
- It turns a tacit skill into something teachable. By naming challenges, objectives and activities, the paper converts a skill usually left to individual supervisory relationships into a design task that programmes can take on.
- It treats supervisors as learners too. Supervisors often learn the role mainly from their own experience as students and by trial and error (Amundsen & McAlpine, 2009). Placing supervisors and students in the same structured activities recognises that both shape the question. Recent Finnish data showing that the experience of supervision is linked to the wellbeing of candidates and supervisors alike strengthen this point (Tikkanen et al., 2025).
- It treats the question as evolving. The emphasis on reformulation throughout the PhD fits research on threshold concepts in doctoral learning, which traces how candidates pass through troublesome stages before they grasp what research and a doctorate require (Kiley, 2009; Trafford & Leshem, 2009).
- It is practical. The scaffolds are concrete enough for a graduate school, a research group or a student peer group to adapt.
Limitations and methodological issues
The central issue for a reader is the type of evidence behind the framework. It comes from the designers' own practice: participant observation by researchers who also acted as trainers and supervisors, programme documents, and comments from trainees during and after the courses (Girard et al., 2025). That is a sound basis for proposing a design, and insider knowledge has genuine value. It does not amount to an evaluation of effectiveness.
- No outcome measure. The paper does not compare question quality, progress or completion between participants and non-participants. It therefore cannot show that the scaffolds produce better questions, as opposed to greater confidence or satisfaction.
- An insider perspective. Because the authors built and ran the courses, their observations may lean towards what worked. Independent observers or systematic, pre-planned data collection would make the claims more secure.
- One institution and setting. Both programmes ran in a single French institute focused on agricultural, food and environmental research, where interdisciplinary and applied questions are common. Transfer to laboratory programmes in which a funded project largely fixes the question, or to professional doctorates, remains to be tested.
- Speaking over writing. The focus on oral argument is a contribution, but most doctorates require the question to be justified in writing in the end, and the link between the two is not examined.
- Course-based delivery. An Australian study of early-stage candidates found that difficulties with the research process were the challenge they named most often, and concluded that support must go beyond one-off workshops or interventions (Liardet, 2024). The INRAE design partly meets this, since one course could be attended at several stages, but the framework says less about support between sessions.
- No place for generative AI. The programmes predate the wide use of generative AI tools, which now enter question and proposal work directly (Naganuma et al., 2026). The framework does not yet say how such tools fit among its scaffolds.
None of these points undercuts the paper's aim, which is to propose a framework, not to prove it. They do govern how it should be cited. In a literature review it is fair to describe Girard et al. as offering a theoretically informed, practice-based design for teaching research question development. Saying that they showed the design improves research questions or doctoral outcomes would overstate the evidence.
How the paper sits in the wider literature
Work on how questions are built helps to place the framework. Sandberg and Alvesson (2011) found that published studies mostly arrive at questions through gap-spotting, that is, by identifying something not yet studied. Alvesson and Sandberg (2011) proposed problematisation instead: challenging the assumptions beneath existing literature to reach more interesting and influential questions. Girard and colleagues' stress on meeting other disciplines' stances could support problematisation, because encountering different assumptions makes one's own visible. Training would benefit from making that link explicit.
Research on supervision adds another layer. Lee (2008) identified several concepts of supervision: functional, enculturation, critical thinking, emancipation and relationship development. The framework's activities touch most of them, which may explain its appeal, but it also means that supervisors holding very different conceptions will bring different expectations to the same workshop. The threshold-concept literature (Kiley, 2009; Trafford & Leshem, 2009) implies that reformulation exercises may help most at the points where candidates are stuck, which argues for timing them to the candidate's stage rather than to a fixed calendar.
What newer research adds (2024 to 2026)
Research published since 2024 does not evaluate the INRAE courses, but it bears directly on their main design choices: student ownership, peer support, the timing of help, and what examiners finally look for. The table summarises eight studies.
| Study | Design and data | Key finding | What it means for the reviewed paper |
|---|---|---|---|
| Lin et al. (2026) | Academic Family Tree and OpenAlex data on 1.8 million supervisor and student pairs, 1960 to 2024 | Co-authorship intensity has an inverted U-shaped association with later productivity, impact and independence; heavy supervisor dependence undermines independence | Supports moving ownership of the question to the student, as in the "student speaks first" scaffold |
| Mireku et al. (2026) | PRISMA-guided systematic review of supervision studies in Sub-Saharan Africa, 2014 to 2024 | Limited supervisor access is the most cited challenge; supervisor training and peer support such as group writing are the most recommended practices | Supports peer-based and supervisor-training elements in a very different setting from INRAE |
| Bakhsh and Elbashir (2026) | Phenomenological interviews with eight doctoral supervisors in Pakistan, analysed with Lee's model and Gurr's developmental approach | Supervisors favour dialogic feedback, trust-building and the gradual transfer of responsibility to students | Echoes the framework's dialogue and transfer of ownership; small sample |
| Tikkanen et al. (2025) | Survey of 768 PhD candidates and 560 supervisors at a Finnish university; latent profile analysis | The supervisory experience relates to both groups' engagement and burnout | Supports treating supervisors as participants with a stake in the process |
| Liardet (2024) | Survey of 105 and interviews with 21 higher degree research candidates in Australia | The research process itself was the most frequently reported difficulty; one-off workshops are not enough | Backs repeated, stage-sensitive support rather than single courses |
| Naganuma et al. (2026) | Three-day workshop with 14 doctoral students using two custom generative AI tools to build interdisciplinary proposals | AI acted as a boundary object and supported epistemic agency; AI-originated ideas made up 20% and 17% of key concepts in the two strongest teams | Suggests AI could join the mixed-discipline dialogue, but raises questions of ownership |
| Ngulube (2025) | Qualitative content analysis of 84 PhD theses in Information Studies at an African institution | Offers a discussion and checklist of the attributes of originality | Gives a possible outcome measure for judging whether question training improves originality |
| Elmgren et al. (2024) | Six focus groups with experienced examination committee members at three Swedish universities | Examiners judge research contribution, educational achievement and academic competence, and their judgements extend to supervision and the local context | Shows why candidates must be able to justify how their question developed |
On the whole, the newer evidence supports the framework's direction. The large co-authorship study by Lin et al. (2026) found that moderate collaboration went with the best later outcomes, while heavy dependence on the supervisor undermined independence. That pattern fits the INRAE practice of letting students speak first. Supervisors in Pakistan describe a similar logic of dialogic feedback and gradual transfer of responsibility (Bakhsh & Elbashir, 2026), and a decade of studies from Sub-Saharan Africa recommends peer support and supervisor training (Mireku et al., 2026). Similar ideas thus appear across very different systems.
Other findings press on the framework's limits. Early-stage candidates report the research process as their main difficulty and need continuing rather than one-off support (Liardet, 2024), so a course model works best when it is tied to ongoing supervision. Generative AI now also takes part in the conversation: in one small workshop it helped doctoral students cross disciplinary boundaries, yet a notable share of key concepts came from the AI (Naganuma et al., 2026). That result sits uneasily with a framework built on the student's ownership of the question.
The central gap remains. None of these studies tests whether training of the kind Girard and colleagues describe produces better questions. Tools now exist that could support such a test, such as checklists of originality (Ngulube, 2025), and examiners' practices show what a well-developed question must ultimately withstand (Elmgren et al., 2024).
What it means for doctoral researchers
You can use the paper's ideas without a formal programme. The table adapts several of its scaffolds, and lessons from the newer studies, for individual use; the adaptations are ours.
| Activity | How to do it | What it develops |
|---|---|---|
| Trace your reformulations | Write your original thesis subject and your current question side by side, and explain each change | Awareness that the question evolves, and a record for your methods chapter |
| Explain it across disciplines | Present your question in five minutes to peers from another field and note what they misunderstand | Clarity and an explicit statement of assumptions |
| Name the stakes | State the academic and practical stakes, the question, and the strategy for answering it as three separate items | Coherence between purpose, question and design |
| Speak first, then listen | In supervision, present your own formulation before asking for your supervisor's view | Ownership of the question and growing independence (Lin et al., 2026) |
| Problematise | List the assumptions shared by the main studies you build on, and ask what changes if one is false | Originality beyond gap-spotting (Alvesson & Sandberg, 2011) |
| Use AI as a sparring partner | If you use a generative AI tool to suggest angles, record which ideas came from it and test each against the literature yourself | Critical judgement and a transparent account of your own contribution |
One habit underpins all of these activities: keep a dated log of your research question. Whenever it changes, note the new wording, what prompted the change (a paper, a pilot, a supervision meeting, an unexpected result, an AI suggestion you tested) and what it means for your design. The log documents your intellectual development and supplies material for the methods and reflexivity sections of your thesis. It also prepares you for examiners, who judge academic competence as well as the contribution itself (Elmgren et al., 2024), and who may ask how your question reached its final form.
For researchers in doctoral education, the paper and the newer studies point to questions worth investigating. Do such scaffolds change the quality of research questions as rated by independent assessors? How do they work for candidates who join funded projects with predefined aims? How do online or international cohorts experience the emphasis on oral argument? And how should programmes handle generative AI so that it widens a candidate's thinking without taking over the question?
Common mistakes when developing a research question
- Keeping a question so broad that no single study could answer it, or so narrow that the answer adds little.
- Choosing a method first and phrasing the question to suit it.
- Relying only on gap-spotting, so that the question asks what has not been studied rather than why it matters.
- Writing questions that can be answered yes or no, which limit what the analysis can say.
- Adopting a supervisor's or project's wording without being able to justify it in your own terms.
- Accepting a question suggested by an AI tool without tracing its assumptions or checking it against the literature.
- Failing to update the question after the literature review or pilot work, so that it no longer matches the study.
Key takeaways
- Girard et al. (2025) propose a framework of three learning challenges, matched training objectives and scaffolds for teaching research question development.
- The framework is drawn from two INRAE programmes and the authors' practice; it is a design proposal, not an outcome evaluation.
- Its strengths are including supervisors, treating the question as evolving and offering concrete activities (Amundsen & McAlpine, 2009; Kiley, 2009).
- Studies from 2024 to 2026 support its emphasis on student ownership and peer support, but show that candidates need continuing help and that generative AI now enters question work (Lin et al., 2026; Liardet, 2024; Naganuma et al., 2026).
- Combining its peer dialogue with problematisation can push questions beyond gap-spotting (Sandberg & Alvesson, 2011; Alvesson & Sandberg, 2011).
- Doctoral researchers can adapt its scaffolds now, for example by tracing and explaining each reformulation of their question.
