Speaker
Description
Introduction
The planning process is characterised by persistent challenges in attaining meaningful public outcomes. How participatory processes are operationalised is one of those enduring challenges, contributing to a lack of trust and to engagement that is often perceived as ineffective by the broader public (Crawford et al., 2020). The emergence of digital planning has not resolved these challenges. Wilson and Tewdwr-Jones (2022) indicate that while planning has focused on efficiency since the 1980s, this has not resulted in more effective engagement. These tools may also exclude seldom-heard groups and reinforce existing power structures. Generative Artificial Intelligence (AI) has emerged within this context, permeating rapidly following the launch of ChatGPT in 2022 (Peres et al., 2023). These tools produce audio, textual and visual content in seconds. In particular, Generative AI systems built on Multimodal Large Language Models (MLLMs), Diffusion Models and Generative Adversarial Networks (GANs) produce photorealistic outputs almost instantly (Wu et al., 2022; Wang et al., 2025). Such models may broaden participation by enabling individuals with varying skillsets to contribute meaningfully to planning discourse (von Brackel-Schmidt et al., 2024; Williams et al., 2024).
This research contends that Generative AI does not reduce uncertainty within participatory processes but instead relocates it in three aspects. First, uncertainty shifts away from the proposal itself and towards the conditions under which it was produced, such as the prompt, the model training data and the version of the tool used. These factors are not visible in the output and are seldom open to meaningful challenge. Second, the focus of uncertainty changes, from what will be delivered to the origin of the material and how it should be interpreted. Planning practices have established methods for addressing the former (Healey, 1997; Wilson and Tewdwr-Jones, 2022) but seldom provide comparable approaches for the latter (Liu et al., 2025). Third, uncertainty is shared unevenly. The practitioner is more likely to know the output may be incomplete or easily altered, while members of the public may assume it is more fixed or reliable than it is, creating a gap between what appears certain and what is actually uncertain. This has direct implications for the legitimacy of outcomes produced through participatory processes.
These risks are not wholly technical in nature. Outputs are susceptible to hallucinations (Quan and Lee, 2025) and exhibit geographic and cultural biases inherent within their training data, disproportionately reflecting dominant perspectives (Xia et al., 2025). Forester (1982) highlighted that the management of information constitutes a form of power within planning, and Generative AI arguably introduces a new arena in which that power is exercised. Existing research provides only partial conceptual linkages to participatory planning theory (Liu et al., 2025), remains dominated by conceptual reviews and single case studies (Du et al., 2024; Jiang et al., 2025), and focuses primarily on early consultative stages of planning processes (Othengrafen et al., 2025). Critically, no existing research compares Generative AI directly against established non-Generative AI methods under controlled conditions, meaning its participatory benefits remain unverified.
Participatory Framework
Currently, participatory planning lacks a comprehensive framework capable of evaluating these normative claims. Existing literature comprises influential yet theoretically distinct traditions, creating a fragmented body of research in which concepts are operationalised differently. Davidoff (1965) challenged the assumption of a singular public interest, establishing the case for plural and competing plans. Arnstein (1969) supplied a vocabulary of power and distinguished tokenistic practice from genuine citizen control. Forester (1982) reframed planning as a communicative and political exercise, identifying distorted communication as a structural barrier to participation. Healey (1992, 1997) conceptualised planning as a relational and iterative practice grounded in shared knowledge creation. Innes and Booher (1999, 2004, 2010) proposed collaborative rationality as a response to complexity, reframing participation as the joint construction of knowledge. Each tradition has been subject to sustained critique, particularly concerning its treatment of power and structural inequality (Westin, 2021). Collectively they provide competing answers to what meaningful participation consists of in Anglo-American contexts. Critically however, none of these authors considered these normative positions in the context of Generative AI emerging globally and within the planning discipline.
The framework responds to this fragmentation by synthesising different participatory traditions into a set of cross-cutting dimensions, following the approach set out by Jabareen (2009). That approach encompasses the mapping of relevant papers, the deconstruction of key concepts and their synthesis into an integrated structure. The dimensions were derived through iterative refinement rather than a formal coding structure, emerging from points of convergence and divergence across the five traditions. They comprise (1) Access and Representation, (2) Influence and Power, (3) Communication and Trust, (4) Deliberation and Inclusion, and (5) Legitimacy and Outcomes.
Each dimension enables a distinct implication for uncertainty to be considered. Access and Representation considers how Generative AI can lower the technical cost of producing visual material, enabling stakeholders with limited resources to enter the process more readily. Influence and Power correspond to the uneven distribution of uncertainty identified above, as practitioners and those generating the material understand how provisional an output may be, whereas members of the public do not. Communication and Trust are affected as trust no longer rests with an identifiable author but instead with a generated process whose origins cannot easily be understood or scrutinised. Deliberation and Inclusion is impacted because deliberation relies on participants questioning the material presented, yet material produced through opaque processes cannot be readily interrogated. Finally, Legitimacy and Outcomes reflects the cumulative effect of the preceding four dimensions, with legitimacy depending not only on whether individuals and groups were consulted, but on whether they held sufficient knowledge to make informed judgements.
Methodology
The research adopts a mixed-methods approach, situated within a pragmatic epistemological orientation, recognising that different forms of knowledge highlight different facets of the same question (Cherryholmes, 1992; Creswell, 2009). Northern Ireland provides the empirical setting for the research, as a subnational jurisdiction with devolved governance arrangements, resource constraints, contested institutional legitimacy and limited empirical data on the impact of Generative AI within participatory planning. The research comprises a workshop stage, outlined here, and a concurrent interview stage reported separately, and has received institutional ethics approval.
Three workshops are undertaken with three stakeholder groups operating within the planning system of Northern Ireland. These include planning professionals, elected representatives and community and voluntary organisations. Each comprises a minimum of twelve participants recruited through a purposive sampling matrix, prioritising deliberate variation rather than statistical representativeness (Patton, 2002). Sampling is structured by geography (urban and rural), and by a second dimension specific to each group, comprising sector for planners, governance tier for elected representatives and prior planning engagement for community and voluntary organisations. The workshops are structured around a single scenario set in the fictional Northern Irish town of Portneagh, a declining locality with high levels of dereliction and vacancy.
The comparison between Generative AI and analogue (non-Generative AI) tools is undertaken within participants rather than between groups. Each participant uses both toolkits and therefore acts as their own point of comparison. Each workshop utilises two toolkits, a paper-based charrette toolkit and a Generative AI toolkit, applied across two matched exercises addressing the same scenario. The order in which participants encounter each toolkit is reversed between exercises, controlling for sequence effects. Each table is supported by a facilitator and observer, with the Generative AI table additionally supported by a technology operator whose role is limited to entering the prompts specified by participants. The operator provides no evaluation, refinement or suggestion, and the same individual undertakes the role across all three workshops, ensuring observed differences are attributable to the toolkits rather than their operation. Four instruments generate the workshop data. The pre-workshop and post-workshop surveys utilise a five-point Likert scale and are analysed using the Wilcoxon signed-rank test (Wilcoxon, 1945), given the same participants complete both and the data is ordinal. A Tool Rating Card, structured around the five participatory dimensions, is completed individually following each exercise, producing a paired judgement of both tools from every participant. Controlled observation is undertaken concurrently at both tables using a structured guide informed by the framework (Merriam and Tisdell, 2016). A pilot study tests these instruments in advance, with pilot participants excluded from the primary fieldwork. Data collection commences in November 2026, with fieldwork underway at the time of the conference, initial results anticipated in early 2027 and analysis concluding across summer 2027.
Expected Results
Three expectations follow from the participatory framework. Each compare how the same participants respond to the Generative AI toolkit versus the analogue (non-Generative AI) toolkit, and each is framed so that emerging evidence can confirm or contradict it. The first concerns access. Participants without professional planning backgrounds are expected to rate the Generative AI toolkit more favourably than the analogue toolkit, reporting greater confidence in engaging with the scenario. This reflects the strongest claim made for these tools in existing literature, that Generative AI reduces technical barriers and enables more accessible experiences. This is contradicted should community and voluntary organisations rate both toolkits equally, or should greater confidence coincide with a reduced sense of influence and power. The second expectation concerns how participants interpret the material produced. Participants are expected to treat Generative AI outputs as more fixed or complete than is warranted, and to question their origins less frequently than the analogue material they produce. Evidence comes from the observations and prompt log, and the expectation is contradicted should participants question the origins of generated material as readily as they amend their own drawings. The third concerns divergence across the participatory dimensions and is critical to the research. The dimensions are expected to move in different directions rather than in unison. Access and Representation are expected to favour the Generative AI toolkit, while Influence and Power and Communication and Trust are expected to favour the analogue toolkit or show no change. The Tool Rating Card is designed to detect this pattern. This expectation is contradicted if the dimensions move together in either direction, as that would indicate a general preference for one toolkit rather than the shift described. Variation across the stakeholder groups is also informative. Should planners interrogate Generative AI material while community groups do not, this demonstrates that Generative AI reproduces rather than reduces the information gaps that persist across participatory planning traditions.
Discussion and Conclusion
The contribution of this research is twofold. Conceptually, it argues that participatory planning theory requires an account of how legitimacy is produced when participatory material is generated by systems that stakeholders cannot inspect. The foundational traditions assume participatory materials have an identifiable author whose production can be questioned. Generative AI interrupts that assumption, however it does not replace the debates established regarding who sets agendas and who exercises influence in the process. It instead reshapes them, introducing a new arena in which power is exercised and contested. The participatory framework provides an evaluative lens through which these questions can be examined empirically. Practically, the research argues that methods for assessing Generative AI must be established before adoption rather than after it, as these tools are being taken up more rapidly than the means of evaluating them are being developed. Without such mechanisms, adoption risks justification on grounds of efficiency rather than the quality of participation enabled, with Generative AI treated as a neutral technical innovation detached from debates regarding representation and power.
The Northern Ireland context provides additional depth to this research. Institutions whose legitimacy is contested have limited capacity to absorb participatory practices that appear more inclusive than they substantively are, and the research examines the technology where its risks are most consequential. The controlled workshop environment is a deliberate methodological decision. In separating the toolkits from the complexities of real planning contexts, the research establishes a comparative baseline that everyday practice cannot provide. The argument also applies in other geographic and institutional contexts where planning operates under constrained institutional trust.
References
Arnstein, S. R. (1969). A ladder of citizen participation. Journal of the American Institute of Planners, 35(4), 216–224.
Cherryholmes, C. H. (1992). Notes on pragmatism and scientific realism. Educational Researcher, 21(6), 13–17.
Crawford, J., Ellis, G., & Fox-Rogers, L. (2020). Building the capacity of the planning system to deliver public interest in Northern Ireland: A scoping paper. Queen’s University Belfast.
Creswell, J. W. (2009). Research design: Qualitative, quantitative, and mixed methods approaches. SAGE Publications.
Davidoff, P. (1965). Advocacy and pluralism in planning. Journal of the American Institute of Planners, 31(4), 331–338.
Du, J., Ye, X., Jankowski, P., Sanchez, T. W., & Mai, G. (2024). Artificial intelligence enabled participatory planning: A review. International Journal of Urban Sciences, 28(2), 183–210.
Forester, J. (1982). Planning in the face of power. Journal of the American Planning Association, 48(1), 67–80.
Healey, P. (1992). Planning through debate: The communicative turn in planning theory. The Town Planning Review, 63(2), 143–162.
Healey, P. (1997). Collaborative planning: Shaping places in fragmented societies. Macmillan Press.
Innes, J. E., & Booher, D. E. (1999). Consensus building and complex adaptive systems. Journal of the American Planning Association, 65(4), 412–423.
Innes, J. E., & Booher, D. E. (2004). Reframing public participation: Strategies for the 21st century. Planning Theory & Practice, 5(4), 419–436.
Innes, J. E., & Booher, D. E. (2010). Planning with complexity: An introduction to collaborative rationality for public policy. Routledge.
Jabareen, Y. (2009). Building a conceptual framework: Philosophy, definitions, and procedure. International Journal of Qualitative Methods, 8(4), 49–62.
Jiang, H., Li, M., Witte, P., Geertman, S., & Pan, H. (2025). Urban chatter: Exploring the potential of ChatGPT-like and generative AI in enhancing planning support. Cities, 158, Article 105701.
Liu, K., Yigitcanlar, T., Mehmood, R., Corchado, J., & Fu, X. (2025). Large language models in urban planning: A systematic review and conceptual framework. Journal of Urban Technology, 1–44.
Merriam, S. B., & Tisdell, E. J. (2016). Qualitative research: A guide to design and implementation. Jossey-Bass.
Othengrafen, F., Sievers, L., & Reinecke, E. (2025). From vision to reality: The use of artificial intelligence in different urban planning phases. Urban Planning, 10, Article 8576.
Patton, M. Q. (2002). Qualitative research and evaluation methods. SAGE Publications.
Peres, R., Schreier, M., Schweidel, D., & Sorescu, A. (2023). On ChatGPT and beyond: How generative artificial intelligence may affect research, teaching, and practice. International Journal of Research in Marketing, 40(2), 269–275.
Quan, S. J., & Lee, S. (2025). Enhancing participatory planning with ChatGPT-assisted planning support systems: A hypothetical case study in Seoul. International Journal of Urban Sciences, 29(1), 89–122.
von Brackel-Schmidt, C., Kučević, E., Leible, S., Simic, D., Gücük, G. L., & Schmidt, F. N. (2024). Equipping participation formats with generative AI: A case study predicting the future of a metropolitan city in the year 2040. International Conference on Human-Computer Interaction (pp. 270–285).
Wang, Q., Liang, Y., Zheng, Y., Xu, K., Zhao, J., & Wang, S. (2025). Generative AI for urban planning: Synthesizing satellite imagery via diffusion models. Computers, Environment and Urban Systems, 122, Article 102339.
Westin, M. (2021). The framing of power in communicative planning theory: Analysing the work of John Forester, Patsy Healey and Judith Innes. Planning Theory, 21(2), 132–154.
Wilcoxon, F. (1945). Individual comparisons by ranking methods. Biometrics Bulletin, 1(6), 80–83.
Williams, S., Beery, S., Conley, C., Evans, M. L., Garces, S., Gordon, E., Jacob, N., & Medina, E. (2024). People-powered gen AI: Collaborating with generative AI for civic engagement. An MIT Exploration of Generative AI: From Novel Chemicals to Opera (pp. 1–30).
Wilson, A., & Tewdwr-Jones, M. (2022). Digital participatory planning: Citizen engagement, democracy, and design. Routledge.
Wu, A. N., Stouffs, R., & Biljecki, F. (2022). Generative adversarial networks in the built environment: A comprehensive review of the application of GANs across data types and scales. Building and Environment, 223, Article 109477.
Xia, J., Tong, Y., & Long, Y. (2025). Advancements in the application of large language models in urban studies: A systematic review. Cities, 165, Article 106142.