This entry is prompted by an article by Danson Zheng and Ian Hardy titled: “Thinking the unthinkable: Generative AI for collaborative, intra-active policy analysis” (Methodological Innovations, 2026).I call to “stop the press” because of the urgency of this topic.
The project in Zheng and Hardy, put simply, is to facilitate a form of dialogue between the researcher and Copilot, an AI-powered conversational assistant. The intended outcome is to generate novel insights into the researched topic. In particular, the goal is to assist researchers to think beyond conventional conceptual premises, thereby assisting them to address Question 4 in the WPR approach: “What is left unproblematic in this problem representation? What is silenced? Can the ‘problem’ be conceptualised differently?” (Bacchi 2026, p. 24).
This project is of particular interest to WPR researchers because WPR is selected as the “analytical method” guiding the analysis. Specifically, WPR is drawn upon to inform the analysis of a Chinese education policy document pertaining to the initiation of national competency-based education reforms within the whole Chinese education system. The Zheng and Hardy article provides a good deal of detail about exactly how the researchers approached this task. Their approach included asking Copilot specifically about its understanding of WPR to see if it matched that of the primary researcher, Danson (Zheng’s first name Danson is used throughout the article): “With this explicit guidance, Copilot responded with interpretations of the theoretical underpinnings as the way Danson understood the [WPR] approach” (Zheng and Hardy 2026, p. 7).
In a past entry (Research Hub 29 Sept 2023) I indicated that I was interested in seeing what AI had to offer WPR researchers. I continue to think this topic is vitally important given the speed at which AI is developing and its expanding place in our lives. The Zheng and Hardy article attracted my interest primarily because it directly considers if Copilot comprehends the theory informing a WPR analysis. For example, Danson (Zheng) asks Copilot about the place of Foucault in WPR: “Could you explain how Bacchi’s WPR approach is built on Foucault’s concepts?”. In this Research Hub entry I will highlight the parts of the Zheng and Hardy analysis that I found most useful and indicate areas where, I believe, more thinking is required. The commitment to produce a form of “agential engagement between the human ‘I’ and the non-human ‘GenAI’” is, I believe, well worthwhile but there are obstacles to overcome.
Copilot, Danson and WPR
The difficulties begin at the outset. Danson (Zheng) asks Copilot: “could you use the six analytical questions in Bacchi’s WPR approach to analyse” the specified Chinese education policy? As is well known, with AI usage, you will get basically what you ask for, making it crucial to pay attention to your questions. Danson asks for sixWPR questions as listed on page 6 of the Zheng and Hardy article. The questions are taken from Bacchi 2012, p. 21. I won’t repeat them since they are commonly referenced.
What I wish to point out is that something is missing from the list. At the bottom of the list of questions in the 2012 article (on p. 21), it reads “Apply this list of questions to your own problem representations”. This statement fails to appear in Danson. I’ll admit that, in the 2012 publication, it may have been possible to consider this instruction as outside of, rather than part of, WPR. However, if researchers check other renditions of the WPR Table of “questions” it is difficult to make this error of omission. For example, in both Tables in Analysing Policy (Bacchi 2009), the statement “Apply this list of questions to your own problem representations” appears clearly as part of the WPR approach (see the boxed versions in Bacchi 2009, p. xii and p.48). The latter offers this expansion of the presuppositions underpinning the instruction:
“Apply this list of questions to your own problem representations. This stage of the analysis requires a form of reflexivity, which involves subjecting the grounding assumptions in one’s own problem representations to critical scrutiny. (emphasis added)”
Bacchi 2009 appears in the reference list for Zheng and Hardy (2026) so these elaborations of the WPR framework were readily available.
Because researchers tend to ignore this undertaking to apply the WPR questions to their own problem representations (because it is not assigned a number in 2009?) my colleague, Susan Goodwin, and I made its status clear in the 2016 book, Poststructural Policy Analysis: A guide to practice (Bacchi and Goodwin 2016, p. 20). There (and again in the 2025 second edition, p. 24) the undertaking to apply the WPR questions to one’s own proposals is called “Step 7”. Finally, in my 2026 book, I change the word “Step” to “Process” to encourage continual application of the principle, which I call “self”-problematisation (see Bacchi and Goodwin 2025, pp. 42-26 and Bacchi 2026, Chapter 6). The importance of this undertaking as part of a WPR analytic strategy is, therefore, clearly endorsed in 2009, 2012 and 2016, well before publication of the Zheng and Hardy article (2026).
Danson (Zheng) explicitly asks Copilot to produce six WPR questions and so, predictably, six questions are produced. It would have been interesting to ask Copilot if it could introduce the WPR framework without specifying 6 questions. I use this example to indicate a larger point – that AI is constrained to deliver within the boundaries set by researchers, who are fallible and who are also positioned in situations where they may offer a partial perspective. We are told “Copilot responded with consistent interpretations of the theoretical underpinnings as the way Danson understood the approach” (Zheng and Hardy 2026, p. 7). A first step in considering the usefulness of the analysis produced, therefore, must involve some assessment of Danson’s (Zheng’s) understanding of WPR. Serious questions need to be asked about how such an assessment could be produced.
“Self”-problematisation and WPR
My critical intervention here is not simply to point out that an omission occurred in describing the analytical framework (WPR), e.g. leaving out Process 7 or “self”-problematisation, but to stress the centrality of “self”-problematisation to WPR and to possible collaborations with Copilot. Indeed, I will suggest that leaving out “self”-problematisation works at cross-purposes with the declared intent in Zheng and Hardy to uncover one’s deep-seated assumptions – which surely is the goal of “self”-problematisation. Hence, by omitting “self”-problematisation, the authors miss exactly what they are looking for. Let us take a closer look at this dynamic.
Zheng and Hardy (2026) are particularly concerned with Question 4 in WPR (see above and Bacchi 2026, p. 24). The authors explain that, before collaborating with Copilot, “Danson had conducted a preliminary human version of the chosen policy” (Zheng and Hardy 2026, p. 6; emphasis in original). Danson was concerned that he was being a “passive ‘follower’ of the policy as constituted” (p. 6). The authors note: “The fourth analytical question from the WPR approach, in particular, was proving challenging”:
“Consequently, to think beyond what was thinkable to him at the time, Danson thus became motivated to work with a non-human partner to ‘more deeply’ analyse the policy. Upon choosing the GenAI partner, Copilot, Danson conducted an exploratory post-qualitative collaborative policy analysis. (p. 6)”
The authors state clearly that “Question Four” in WPR motivated Danson to take on the collaboration. They specify in relation to Question 4:
“This is not an easy question, as it challenged Danson to think about what is silenced or unproblematic in the policy. By the time of the collaboration, Danson was uncertain about the analysis he had derived for this particular analytical question in his preliminary ‘human version’ of analysis with the WPR approach. (p. 7)”
Danson conducts a kind of critical interrogation with Copilot, looking to better understand Queston 4. The authors produce the methodological steps taken by Danson in collaboration with Copilot to “reflect on how Copilot’s responses prompted Danson to think differently” and I refer you to page 8 of the article for details (we pursue the question of language and translation in the last section). The primary insight generated through this
“richly collaborative intra-active series of analytical conversation challenged Dansen to reflect on how his thinking was strongly regulated by policy discourses within the Chinese policy context and to re-examine his ways of knowing and being in relation to language use and research engagement. (p. 8; emphasis added)”
You may recognise here stated goals that align with “self”-problematisation – “subjecting the grounding assumptions in one’s own problem representations to critical scrutiny” (see above), which unfortunately disappears from the WPR framework in this account.
Zheng and Hardy are certainly correct that Question 4 is challenging. In past publications I consider a range of research strategies for engaging this question.
These include:
- Working with comparative analysis
- Reading the critical literature
- Exploring WPR as a group exercise (including cross-disciplinary contributions or “lay” “experience”). (Bacchi 2026, p. 22; Research Hub entry 29 Jan 2026)
The question prompted by Zheng and Hardy’s innovative approach, I suggest, is: could generative AI operate as one possible way of disrupting taken-for-granted “truths” and raising alternative problematisations? Could GenAI assist with Question 4? I see promise for this suggestion largely through the practice of translation.
Copilot and translation
The publication of articles/chapters using WPR in languages other than English has consistently produced a stumbling block for me. There are some languages where, I believe, even the question – “What’s the problem represented to be?”- makes little sense. Clearly AI has tremendous translation skills that could assist with this situation.
Zheng and Harding provide an excellent example of the role played by translation in their interrogation of the selected Chinese education document. To understand how they proceed I need to supply some detail:
“This first lightbulb moment also became possible due to the translation between two languages by Copilot. In this collaboration, Danson communicated with Copilot for collaborative policy analysis in English. As the policy is written in Mandarin Chinese, Copilot’s effective processing of the language patterns inevitably involved a translation process. This translation process by Copilot brought unexpected benefits through disassembling/disrupting the Chinese language patterns that Danson initially ‘took for granted’. (p. 8)”
I refer you to the article for additional explanation. The authors conclude that: “GenAI’s capability in deconstructing and reconstructing language patterns, then, disrupted the discursive relations and opened up the possibility for the human researcher, Danson in this study, to think differently” (p. 9).
The goal of “thinking differently” runs as a dominant theme in Zheng and Hardy. The article title highlights this focus: “Thinking the unthinkable: Generative AI for collaborative, intra-active policy analysis”. And thinking differently is certainly one objective of Question 4 in WPR. If GenAI prompts researchers to think differently, engaging with GenAI seems to be a worthwhile project. The example of translation indicates how this outcome can be achieved.
Conclusion
Zheng and Hardy open up the question of “engagement” between human and non-human subjects (AI) in stimulating ways. They change the tone of most work on this topic where the major ethical concerns include: plagiarism, integrity, fabrication and misinformation (p. 2). Challenging the argument that qualitative research should be a distinct human practice, they support “a more ‘ethical’, intra-active way of engaging with GenAI by acknowledging both humans’ and GenAI’s capabilities”.
With all this, I struggle to see how GenAI would engage with “self”-problematisation, a central premise in WPR and the grounds of its ethical commitment. It would be useful to test GenAI by asking it to perform a WPR analysis of a selected policy document once one had confirmed that GenAI accepts the full list of questions and modes of analysis endorsed in 2009, 2012, 2016 and more recently (Bacchi and Goodwin 2025, Bacchi 2026) – i.e. including “self”-problematisation. Specifically, it would open up this conversation to important new dimensions to see if a “reflexive” GenAI is feasible and what this could look like. I take the topic further in the next entry.
References
Bacchi C. 2009. Analysing Policy: What’s the Problem Represented to Be? Pearson Education.
Bacchi C. 2012. Introducing the ‘What’s the Problem Represented to be?’ approach. In: Bletsas A and Beasley C (eds) Engaging with Carol Bacchi: Strategic Interventions and Exchanges. The University of Adelaide Press, pp.21–24.
Bacchi, C. 2026. What’s the Problem Represented to be? A new thinking paradigm. NY: Routledge.
Bacchi, C. and Goodwin, S. 2016. Poststructural Policy Analysis: A Guide to Practice. Second edition 2025. NY: Palgrave Macmillan.
Zheng, D. and Hardy, I. 2026. “Thinking the unthinkable: Generative AI for collaborative, intra-active policy analysis”. Methodological Innovations,
Carol Bacchi 