Artificial Intelligence and the Future of Citizen Participation

There is a growing literature on the risks and challenges that emerge from the adoption of AI in government (Berryhill et al., 2019[1]; OECD, 2024[2]; OECD, 2025[3]; Peixoto, Canuto and Jordan, 2024[4]; OECD, 2024[5]) and the threats that AI poses to democracy (Manheim and Kaplan, 2019[6]; Duberry, 2022[7]; Jungherr, 2023[8]). The OECD Expert Group on AI Futures identified 10 priority risks for enhanced policy focus, including misinformation and manipulation (OECD, 2024[9]), erosion of accountability due to lack of transparency and explainability, and invasive surveillance (OECD, 2024[5]) (see Box 3.1). As of August 2025, the OECD Observatory on AI collected nearly 8600 AI incidents and hazards since January 2022, among which more than 3,100 concerned government functions, security, and defence (OECD.AI, 2025[10]). The OECD Report Governing with Artificial Intelligence: The State of Play and Way Forward in Core Government Functions distinguishes between risks and implementation challenges. The report clusters risks in ethical, operational, exclusion, public resistance, and risk of inaction. It also mentions several implementation challenges, ranging from skill gaps to limited scalability of pilots.

However, limited attention has been devoted to how these risks and challenges translate in the context of citizen participation in policymaking (Verhulst, 2025[11]). While technological advancement is likely to address and overcome at least some of these challenges (OECD, 2025[3]), policymakers should be aware of their implications when deciding which AI tools they should use for participatory and deliberative processes. This chapter builds on the relevant categories of risks and implementation challenges outlined in the broader OECD work on AI in government to analyse more in depth how both apply to the adoption of AI in citizen participation.

Technology is not neutral (Hare, 2022[12]; Floridi, 2023[13]; Whelchel, 1986[14]), it is the product and the reflection of the values and the structures of the societies that shape it (Khun, 1968[15]; Huyskes, 2024[16]). AI systems are shaped by human choices at every stage, from model selection and training data to fine-tuning and parameter adjustments (OECD, 2025[3]). These choices can have consequences when AI tools are used to improve and enhance citizen participation processes.

Citizen participation in policymaking is meant to be a practice that preserves complexity and nuance, enabling the expression of variety of opinions and structuring societal conflict and trade-offs into democratic dialogue. Technology should empower governments and citizens by providing new opportunities for this complexity to unfold, not silence the richness of democratic debate and societal conflict through technical layers and opaque systems. AI systems carry the potential of improving government efficiency and increase its productivity, but this should not come to the cost of avoiding or reducing its efforts in engaging with citizens, in particular with those affected by the digital divide. Moreover, improper and malicious uses of AI constitute a threat to the civic space and to information integrity (OECD, 2022[17]; OECD, 2024[9]).

Incomplete or inadequate data embedded in AI models and tools (Joyce et al., 2021[18]; Eubanks, 2018[19]) can undermine the quality of citizen participation processes by resulting in skewed and imbalance outcomes, potentially eroding citizens’ trust. For example, imbalanced or incomplete training data might encode political bias (Peters, 2022[20]) and provide citizens with non-neutral assistance when researching information or drafting contributions (Tsai et al., 2024[21]).

The misuse of digital technologies by government to monitor, track or even silence their oppositions and the populations at large constitute a threat to civic space. (OECD, 2022[17]). The OECD has identified invasive surveillance to be a significant risk of AI, which can undermine the free exercise of human rights and freedoms (OECD, 2024[5]). In 2022, 6 out of the 19 AI strategies analysed in the 2022 OECD Global Report on Civic Space (Chile, Denmark, Latvia, the Netherlands, Spain, Sweden) included a discussion on the impact of AI on rights, namely personal data protection, transparency and the consequences of the non-explainability of algorithmic decision making. More than half of the strategies (53%) proposed the introduction of concrete oversight and redress mechanisms (OECD, 2022[22]).

Moreover, AI has made it easier for malicious actors to generate content and pollute the information space through deep fakes and bots (OECD, 2024[9]; Appel and & Prietzel, 2022[23]). With an increased use of AI to create and disseminate information, the entire information ecosystem including governments will face new questions regarding the accuracy and fairness of information, in the context of growing threats to information integrity, mis- and disinformation.

A polluted information space constitutes a challenge for democratic participation, during and between elections. In 2024, 122 national elections took place, including the European Parliament, the United States, and India. Research by the MIT and the Alan Turing institute calls attention on the “overblown” impact of GenAI on misinformation and the elections (Felix M. Simon, 2024[24]; Stockwell, 2024[25]), with only 19 cases of AI interference among the 112 electoral processes. Both research stress that electoral behaviour is a complex phenomenon that misinformation alone cannot change radically. Nevertheless, the Romanian elections of December 2024 were annulled by the Constitutional court following the surprise win of a candidate who used AI bots to conduct an aggressive social media campaign (Damian, 2024[26]).

Malicious actors might use AI tools to interfere with citizen participation processes. Astro-turfing techniques, such as the generation of significant amounts of misleading contributions, can result in skewed outcomes (Verhulst, 2025[11]).

Deep learning systems are often described as “black boxes” because their nature makes it difficult to explain how they arrive at a specific output. Their results stem indirectly from the training and inference processes, during which engineers iteratively adjust parameters until the model achieves high performance on predefined objectives (Clarke and Whittlestone, 2022[27]). Even experts developing state-of-the-art deep learning models lack full insight into their internal decision-making processes, making it difficult to trace the origins of particular outputs or to assess the reliability of such systems using conventional evaluation methods (OECD/CAF, 2022[28]). The inherent structure of AI systems might affect public acceptance and trust in the use of AI tools as “intermediaries” between governments and citizens in citizen participation processes.

Inadequate or incomplete training datasets and models can affect the quality of the outputs of AI tools used to support citizen participation processes. If data is partial or incomplete, AI facilitators might not reflect correctly the depth and the variety of opinions of deliberative processes. This can result in reducing or silencing minority views. Unfair AI moderation can result in censorship of specific individuals or views on a given issue. Moreover, GenAI tools used to draft consensus statements might over-simplify the nuances of debates and propose “central options” that do not reflect the complexity of the discussion. AI models might also lack understanding of context-sensitive information. When supporting citizens in drafting and submitting written contributions, AI assistance and information development tools might affect the creativity and originality of the participants’ reflection.

To adequately support citizen participation processes, AI tools should be shaped and used to preserve the quality and nuance of exchanges and contributions. Participatory and deliberative processes are not only a powerful way for governments to co-design and then deliver more effective policies, they are also, some say primarily, about fostering empathy, mutual understanding, and consensus-building in a complex environment of interests and beliefs (OECD, 2020[29]).

Many people perceive the outputs of AI systems as neutral and impartial, which might encourage users to accept them as legitimate without further scrutiny. In government, this “automation bias” occurs when public organisations or civil servants rely too heavily on AI systems for decision-making or task execution. This excessive dependence can result in users failing to recognise mistakes, accepting incorrect AI outputs, and diminishing human oversight and judgment (Passi and Vorvoreanu, 2022[30]; Klingbeil, Grützner and Schrec, 2024[31]). The presence of hallucinations, i.e. credible AI-generated facts or answers, makes overreliance on AI systems particularly risky. In the case of citizen participation processes, overreliance on AI systems might lead for instance to skewed outputs when analysing citizen inputs (Romberg and Escher, 2024[32]), or reduced quality of debate when using AI-powered moderation or facilitation tools.

Developing and deploying AI systems poses privacy and data governance challenges throughout the AI lifecycle (OECD, 2024[33]). AI systems might infringe the privacy of individuals as their personal data might be part of training datasets, or retained when using the tools, or inferred by AI systems themselves (OECD, 2025[3]). Moreover, deleting personal information from AI systems might prove technically challenging and resource intensive. These concerns could further erode citizens’ trust in government and in citizen participation processes. To address such tensions, promoting further international cooperation between data privacy and AI communities can contribute to harmonised data practices with AI development and use. For example, OECD’s Expert Group on AI, Data, and Privacy is exploring policy responses on data governance and privacy in the context of AI, involving experts from multiple sectors and disciplines around the world.

With governments’ adoption of AI systems, the risk of exclusion of people with limited access to devices, quality internet connection, and digital literacy increases (OECD, 2025[3]; ECNL, 2024[34]). If they rely solely on digital and AI tools, citizen participation processes might result exclusive by design to certain groups of citizens, such as the elderly and people living in isolated areas with limited access to the internet (OECD, 2022[35]). Moreover, imbalances in AI systems performance in different languages can affect linguistic minorities by poorly filtering their contributions and losing cultural nuances (Romberg and Escher, 2024[32]).

Large Language Models perform unevenly across languages. AI systems are mainly trained in English, Spanish, and Mandarin Chinese, defined “high resource languages” while other languages are underrepresented, or “low-resource languages” (Peixoto, Canuto and Jordan, 2024[4]; Li and al., 2025[36]). In the context of citizen participation, this means that inputs submitted in non-highly performing languages might not be processed and valued in the same way creating new democratic inequalities (Romberg and Escher, 2024[32]).

The ability of AI systems to monitor debate, make predictions, and simulate scenarios might induce governments to believe that consulting and engaging with citizens will no longer be necessary. There is a growing market of AI “personae” that companies can use for market research purposes (Leoni and Strothe, 2025[37]), it is highly plausible that similar tools will be soon available to simulate citizens (Rehan and Hassan, 2925[38]). Although synthetic citizens could be useful in very specific contexts, for example to represent future generations in the formulation of long-term policies (Swinkels, de Vette and Toom, 2025[39]), governments should not consider simulations a surrogate for their duty to involve citizens in their action (People Powered, 2025[40]).

Previous failures in AI deployment have significantly impacted reputations and eroded public trust in the government's capacity to use AI responsibly. The Netherlands' "Toeslagenaffaire" (childcare benefits scandal), where an AI system wrongfully accused 26 000 families of fraudulently claiming childcare benefits due to a skewed algorithm, who were then forced to repay those benefits, resulted in severe consequences for the people involved. Following the scandal, the government collapsed (OECD, 2023[41]). Incidents underscore the importance of building adequate guardrails for trustworthy AI adoption in government, including strong accountability and redress mechanisms, continuous monitoring and oversight, and effective risk management. Public resistance emerged as a strong concern among the cases of AI uses in government across 11 policy functions (76% of cases analysed in the OECD Report “Governing with Artificial Intelligence”) (OECD, 2025[3]).

Debates about AI often focus on deployment risks. However, delaying its responsible adoption and use, including in government‑led citizen participation, also constitutes a risk. Such delays can result in avoidable financial and non-financial costs. Reluctance in adopting AI systems typically stems from technical limitations, legacy IT, unclear privacy/data rules, insufficient awareness of practical use cases, and the fear of “getting AI wrong” (OECD, 2025[3]). Nevertheless, the use of AI in citizen participation processes can expand outreach, lower participation barriers, synthesise inputs at scale, and shorten feedback loops for better‑informed decisions. Failing to consider these tools might result in poorer process design, more limited outreach, and in some cases even in renouncing on the implementation of processes for limited resources – which would have been made possible by delegating some tasks to AI tools.

The use of AI tools to support citizen participation is still at a formative stage, characterised by experimentation, pilots, and ad hoc initiatives rather than systematised practice. While these early efforts demonstrate the potential of AI to expand and improve democratic engagement, they also reveal significant implementation challenges, including limited availability of actionable guidance, gaps in skills and capacities within the public sector, and risks linked to market concentration and vendor lock-in.

Public officials and civil servants often lack AI literacy and capabilities, including technical capabilities to develop and maintain AI systems, literacy to adopt and use AI tools, and strategic understanding of the opportunities of AI to adapt organisations and processes (UNESCO, 2022[42]). These gaps are particularly significant at the local level (UN Habitat, 2024[43]). Inadequate skills are also a major barrier to the uptake of AI in government and of its limited success when adopted (OECD, 2024[2]).

Skills gaps might also trigger other risks in the use of AI for participation, such as overreliance on AI tools, which preventing civil servants from engaging in a critical way with the technology and from flagging incorrect results (Alon Barkat and Busuioc, 2024[44]). Finally, skills gaps can often lead to outsourcing AI tools through public procurement, both increasing the costs and resulting in lock-ins (Autio, Communigs and Elliott, 2023[45]). In the context of citizen participation, this might exacerbate both the technological and ethical risks, undermining the quality of participatory and deliberative processes (Duberry, 2022[7]), while hindering the opportunities for AI to address the challenge of limited financial resources.

The majority of case studies analysed in the Typology are the product of pilot or ad-hoc initiatives (32 out of 50). More broadly, OECD analysis of the use of AI in government, consistently with the discussions held in OECD Working Parties and Networks, highlights that most initiatives remain at an early stage, often taking the form of pilots or experimental projects (OECD, 2025[3]).

In the case of citizen participation, the ad hoc nature of the adoption of AI tools intersects with the ad hoc implementation of citizen participation processes, which are often not embedded in the institutional system. Moreover, most of the cases analysed were implemented at the local level, which highlights the role of regions and cities in pushing the boundaries of democratic innovations (OECD, 2023[46]) but also comes with the challenges of coherence and scaling to succeed implementation at the national level.

This gradual and experimental approach is valuable. The OECD has long encouraged governments to adopt new methods in a controlled, iterative way to reduce risks and costs, and to allow inevitable failures to occur early, thus generating lessons that can inform future initiatives (OECD, 2017[47]; OECD, 2024[48]). At the same time, it is important to recognise that the ultimate objective of most AI projects is to move beyond experimentation, and to implement and, where appropriate, scale up successful solutions (Sarabi, 2025[49]).

The use of AI in citizen participation processes is a recent trend. Governments often lack actionable guidance on which opportunities exist, and which steps are needed to ensure a meaningful and responsible usage. This might cause risk-aversion and result in missed opportunities. National strategies for AI in government, which are now common, are often limited to high-level principles and priorities of action, lacking operational considerations (OECD, 2025[3]). Furthermore, as citizen participation processes are often implemented at the local level, it is crucial to develop actionable guidance that is relevant for multiple levels of governance while ensuring coordination and effective resource allocation.

[44] Alon Barkat, S. and M. Busuioc (2024), “Public administration meets artificial intelligence: Towards a meaningful behavioral research agenda on algorithmic decision-making in government”, Journal of Behavioral Public Administration, Vol. 7, https://doi.org/10.30636/jbpa.71.261.

[23] Appel, M. and F. & Prietzel (2022), “The detection of political deepfakes.”, Journal of Computer-Mediated Communication, Vol. 27/4.

[45] Autio, C., K. Communigs and A. Elliott (2023), Snapshot of Artificial Intelligence Procurement, https://files.thegovlab.org/a-snapshot-of-ai-procurement-challengesjune2023.

[1] Berryhill, J. et al. (2019), “Hello, World: Artificial intelligence and its use in the public sector”, OECD Working Papers on Public Governance, No. 36, OECD Publishing, Paris, https://doi.org/10.1787/726fd39d-en.

[27] Clarke, S. and J. Whittlestone (2022), A Survey of the Potential Long-term Impacts of AI - How AI Could Lead to Long-term Changes in Science, Cooperation, Power, Epistemics and Values, https://doi.org/10.1145/3514094.3534131.

[26] Damian, A. (2024), The annulment of Romania’s presidential election reflects both foreign meddling and domestic failures, https://blogs.lse.ac.uk/europpblog/2024/12/13/the-annulment-of-romanias-presidential-election-reflects-both-foreign-meddling-and-domestic-failures/.

[7] Duberry, J. (2022), Artificial Intelligence and Democracy, https://library.oapen.org/bitstream/handle/20.500.12657/91687/210733.pdf?sequence=1&isAllowed=y.

[34] ECNL (2024), Can AI tools and platforms make public engagement truly meaningful and, https://ecnl.org/news/ai-public-participation-hope-or-hype.

[19] Eubanks, V. (2018), Automating Inequality: How High-Tech Tools Profile, Police, and Punish the Poor, St. Martin’s Press.

[24] Felix M. Simon, K. (2024), AI’s impact on elections is being overblown, https://www.technologyreview.com/2024/09/03/1103464/ai-impact-elections-overblown/.

[13] Floridi, L. (2023), “On Good and Evil, the Mistaken Idea That Technology Is Ever Neutral, and the Importance of the Double-Charge Thesis”, Philosophy & Technology, https://doi.org/10.1007/s13347-023-00661-4.

[12] Hare, S. (2022), Technology is not neutral: A short guide to technology ethics, London Publishing Partnership.

[16] Huyskes, D. (2024), Tecnologia della rivoluzione: progresso e battaglie sociali dal microonde all’intelligenza artificiale, Il Saggiatore.

[18] Joyce, K. et al. (2021), “Toward a Sociology of Artificial Intelligence: A Call for Research on Inequalities and Structural Change”, Socius, Vol. 7, https://doi.org/10.1177/2378023121999581.

[8] Jungherr, A. (2023), “Artificial intelligence and democracy: A conceptual framework”, Social media+ society, Vol. 9/3.

[15] Khun, T. (1968), “LA ESTRUCTURA DE LAS REVOLUCIONES CIENTIFICAS”, Investigación Económica, Vol. 28/111/112, https://www.jstor.org/stable/42776917.

[31] Klingbeil, A., C. Grützner and P. Schrec (2024), “Trust and Reliance on AI - An experimental study on the extent and cost of overreliance on AI.”, https://doi.org/10.1016/j.chb.2024.108352.

[37] Leoni, F. and G. Strothe (2025), Personas in the Age of AI: Swiss insights, https://www.ipsos.com/en-ch/personas-age-ai-swiss-insights.

[36] Li, Z. and E. al. (2025), Language ranker: A metric for quantifying llm performance across high and low-resource languages.

[6] Manheim, K. and L. Kaplan (2019), “Artificial intelligence: Risks to privacy and democracy”, Yale JL&Tech, Vol. 21.

[3] OECD (2025), Governing with Artificial Intelligence: The state of play and way forward in core government functions, OECD Publishing.

[33] OECD (2024), “AI, data governance and privacy: Synergies and areas of international co-operation”, OECD Artificial Intelligence Papers, No. 22, OECD Publishing, Paris, https://doi.org/10.1787/2476b1a4-en.

[5] OECD (2024), “Assessing potential future artificial intelligence risks, benefits and policy imperatives”, OECD Artificial Intelligence Papers, No. 27, OECD Publishing, Paris, https://doi.org/10.1787/3f4e3dfb-en.

[48] OECD (2024), Enabling Digital Innovation in Government: The OECD GovTech Policy Framework, OECD Digital Government Studies, OECD Publishing, Paris, https://doi.org/10.1787/a51eb9b2-en.

[9] OECD (2024), Facts not Fakes: Tackling Disinformation, Strengthening Information Integrity, OECD Publishing, Paris, https://doi.org/10.1787/d909ff7a-en.

[2] OECD (2024), “Governing with Artificial Intelligence: Are governments ready?”, OECD Artificial Intelligence Papers, No. 20, OECD Publishing, Paris, https://doi.org/10.1787/26324bc2-en.

[41] OECD (2023), Global Trends in Government Innovation 2023, OECD Public Governance Reviews, OECD Publishing, Paris, https://doi.org/10.1787/0655b570-en.

[46] OECD (2023), Open Government for Stronger Democracies: A Global Assessment, OECD Publishing, https://doi.org/10.1787/5478db5b-en.

[35] OECD (2022), OECD Guidelines for Citizen Participation Processes, OECD Public Governance Reviews, OECD Publishing, Paris, https://doi.org/10.1787/f765caf6-en.

[17] OECD (2022), The Protection and Promotion of Civic Space: the Global Context and the Way Forward.

[22] OECD (2022), The Protection and Promotion of Civic Space: Strengthening Alignment with International Standards and Guidance, OECD Publishing, Paris, https://doi.org/10.1787/d234e975-en.

[29] OECD (2020), Innovative Citizen Participation and New Democratic Institutions: Catching the Deliberative Wave, OECD Publishing, https://doi.org/10.1787/339306da-en.

[47] OECD (2017), Fostering Innovation in the Public Sector, OECD Publishing, Paris, https://doi.org/10.1787/9789264270879-en.

[10] OECD.AI (2025), AIM: AI Incidents and Hazards Monitor, https://oecd.ai/en/incidents.

[28] OECD/CAF (2022), The Strategic and Responsible Use of Artificial Intelligence in the Public, OECD Publishing, https://doi.org/10.1787/1f334543-en.

[30] Passi, S. and M. Vorvoreanu (2022), Overreliance on AI: Literature review, https://www.microsoft.com/en-us/research/publication/overreliance-on-ai-literature-review/.

[4] Peixoto, T., O. Canuto and L. Jordan (2024), AI and the Future of Government: Unexpected Effects and Critical Challenges.

[40] People Powered (2025), Guide to Digital Participation Platform, https://drive.google.com/file/d/1J0BlXA5d6Fcgbk7MIZTqaq36eVgl9jaP/view?pli=1.

[20] Peters, U. (2022), “Algorithmic Political Bias in Artificial Intelligence Systems”, Philos. Technol., https://doi.org/10.1007/s13347-022-00512-8.

[38] Rehan and Hassan (2925), “Synthetic Ecosystems: Using Generative Agents to Simulate Urban Policy Outcomes in Virtual Cityscapes”, Journal of Data and Digital Innovation.

[32] Romberg, J. and T. Escher (2024), “Making Sense of Citizens’ Input through Artificial Intelligence: A Review of Methods for Computational Text Analysis to Support the Evaluation of Contributions in Public Participatio”, Digital Government: Research and Practice, pp. 1-30, https://doi.org/10.1145/3603254.

[49] Sarabi, R. (2025), How AI Can Unlock Public Wisdom and Revitalize Democratic Governance, https://carnegieendowment.org/posts/2025/07/how-ai-can-unlock-public-wisdom-and-revitalize-democratic-governance?lang=en.

[25] Stockwell, S. (2024), What has the ‘year of elections’ taught us about AI and democracy?, https://www.turing.ac.uk/blog/what-has-year-elections-taught-us-about-ai-and-democracy.

[39] Swinkels, M., O. de Vette and V. Toom (2025), “Future design in the public policy process: giving a voice to future generations”, https://doi.org/10.1080/01442872.2025.2502678.

[21] Tsai, L. et al. (2024), “Generative AI for Pro-Democracy Platforms”, An MIT Exploration of Generative AI, https://doi.org/10.21428/e4baedd9.5aaf489a.

[43] UN Habitat (2024), Global assessment of Responsiuble AI in cities, https://unhabitat.org/sites/default/files/2024/08/global_assessment_of_responsible_ai_in_citie.

[42] UNESCO (2022), UNESCO Digital Competency Framework for civil servants, https://unesdoc.unesco.org/in/documentViewer.xhtml?v=2.1.196&id=p::usmarcdef_0000383325&file=/in/rest/annotationSVC/DownloadWatermarkedAttachment/attach_import_9060015b-7ee6-4d11-a6c9-8569ecba6b71%3F_%3D383325eng.pdf&locale=en&multi=true&ark=/ark:/48223/p.

[11] Verhulst, S. (2025), Engagement Integrity: Ensuring Legitimacy at a time of AI-Augmented Participation, https://sverhulst.medium.com/engagement-integrity-ensuring-legitimacy-at-a-time-of-ai-augmented-participation-677a6ca275e0.

[14] Whelchel, R. (1986), “Is Technology Neutral?”, IEEE Technology and Society Magazine, Vol. 5, https://doi.org/10.1109/MTAS.1986.5010049.