Democratic governance has always been a technology problem. From the Athenian kleroterion—the allotment machine that randomly selected citizens for civic duty—to the printing press that made mass political literacy possible, to the internet that dissolved the gatekeeping function of broadcast media, each technological inflection point has forced a renegotiation of what participation means, who gets to participate, and who controls the infrastructure of collective decision-making.
Artificial intelligence represents the most consequential such inflection point in a generation. Unlike previous technologies that expanded the channels of democratic communication, AI operates at the level of meaning itself—synthesising, filtering, prioritising, and in some cases generating the content of civic discourse. The implications are profound, and the governance frameworks being assembled in 2026 are only beginning to grapple with them.
From Experiment to Infrastructure
The trajectory of digital democracy over the past decade has been one of gradual institutionalisation. What began as a collection of civic technology experiments—online petitions, participatory budgeting platforms, digital town halls—has matured into what the OECD's June 2026 landmark report, Artificial Intelligence and the Future of Citizen Participation, describes as critical infrastructure. Governments across 22 OECD member and partner countries are now deploying AI across 50 documented use cases in participatory governance, from multilingual sense-making tools that process thousands of citizen submissions to virtual assistants that guide residents through complex consultation workflows.
The scale of adoption is striking. Go Vocal's 2026 Trends Report documents a 300% increase in Single Sign-On and eID verification adoption across civic participation platforms—a signal that digital democracy is being integrated into enterprise-grade identity and security frameworks rather than remaining a peripheral experiment. The GovTech market, already substantial, is projected to reach the trillion-dollar range over the next decade, driven by a combination of legislative mandates, budget pressures, and the genuine efficiency gains that well-implemented digital participation can deliver.
A 300% increase in eID verification adoption signals that digital democracy is maturing into critical infrastructure—but critical infrastructure demands critical scrutiny, not just critical adoption.
The OECD's typology of AI applications in civic participation is instructive. It identifies nine distinct functions: information development, sense-making, translation, transcription, virtual assistance, moderation, facilitation, simulation, and architecture. Each represents a different point of intervention in the participatory process, and each carries distinct risks. Sense-making tools—which use natural language processing to cluster and summarise large volumes of qualitative public input—are currently the most widespread, alongside virtual assistants. Both are also among the most consequential, because they operate at the point where citizen voice is translated into something that policymakers can act upon.
The Legitimacy Problem
Here lies the central tension of AI-assisted democratic governance. The efficiency case for these tools is compelling: a city conducting a major infrastructure consultation might receive 50,000 written submissions. No human team can meaningfully process that volume within a policy-relevant timeframe. AI sense-making tools can cluster submissions by theme, identify areas of consensus and contention, and surface minority viewpoints that might otherwise be lost in the aggregate. In theory, this makes participation more meaningful, not less.
But the theory depends on a set of assumptions that deserve scrutiny. When an AI system decides which themes are salient, which submissions are similar, and which viewpoints constitute a coherent position, it is making editorial judgements. Those judgements are not neutral. They reflect the training data, the design choices, and the optimisation objectives of the system's developers. A sense-making tool trained primarily on formal written English may systematically underweight submissions from communities that communicate differently. A clustering algorithm optimised for consensus may smooth over legitimate dissent. A moderation system calibrated to remove "toxic" content may disproportionately flag the language of marginalised communities whose political expression does not conform to dominant norms.
The OECD's report is candid about these risks. It identifies algorithmic bias, hallucinations, lack of transparency, and susceptibility to cyber threats as critical challenges. It also flags what it calls "exclusionary risks"—the potential for AI to widen the digital, language, and societal divide rather than narrow it. These are not hypothetical concerns. They are documented failure modes of AI systems deployed in high-stakes contexts, and democratic governance is about as high-stakes as it gets.
When AI systems synthesise public input, the question is not merely whether they are accurate—it is whether they are legitimate. Legitimacy in democratic governance cannot be delegated to a model.
When AI systems synthesise public input, the question is not merely whether they are accurate—it is whether they are legitimate. Legitimacy in democratic governance cannot be delegated to a model.
The legitimacy problem is not merely technical. It is constitutional. Democratic legitimacy derives from the consent of the governed—from the belief that collective decisions reflect, in some meaningful sense, the will of the people. When AI systems mediate between citizen voice and policy outcome, the chain of democratic accountability becomes more complex. Citizens who participate in a consultation may not know that their submissions were processed by an AI system, let alone understand how that system shaped the conclusions that policymakers received. Transparency is not just a technical requirement; it is a democratic one.
The Polarisation Paradox
One of the most striking findings in the 2026 GovTech landscape is the explicit acknowledgement that polarisation is a top global risk—and that digital democracy tools are being deployed partly as a response to it. Go Vocal's Trends Report documents that OECD data shows resident trust in local government rising from 22% to 69% when individuals feel they have genuine influence on decision-making. This is a remarkable figure, and it points to something important: the crisis of democratic legitimacy is not primarily a crisis of institutions, but a crisis of participation.
People who feel heard are more likely to trust the systems that govern them. This insight drives the hybrid participation model that is emerging as the dominant approach in 2026—blending digital scale with offline depth, using AI to manage the volume of digital engagement while preserving the deliberative quality of in-person processes. Germany's "Forum Against Fakes," which combined large-scale digital consultation with in-person assemblies, is an early example of this approach. The Netherlands' forthcoming Participatieverordening (2027) will make civic engagement a compliance requirement, creating a legislative mandate for the kind of structured participation that these platforms enable.
But the polarisation paradox cuts both ways. The same digital infrastructure that enables genuine participation can also be weaponised for manufactured consensus. Astroturfing—the use of coordinated inauthentic accounts to simulate grassroots support—is a well-documented problem in online civic spaces. AI makes astroturfing dramatically cheaper and more sophisticated. A well-resourced actor can now generate thousands of plausible-sounding submissions that cluster around a desired policy position, potentially overwhelming genuine citizen input and distorting the sense-making tools that governments rely upon.
This is not a theoretical risk. It is an active threat to the integrity of digital democratic processes, and it requires a response that goes beyond the technical. Platforms need robust identity verification—hence the significance of the eID adoption trend—but they also need adversarial testing, red-teaming, and ongoing monitoring for coordinated inauthentic behaviour. The OECD's recommendation to prioritise open-source standards and human oversight is a step in the right direction, but it is not sufficient on its own.
The Architecture of Accountability
If AI is to play a legitimate role in democratic governance, the architecture of accountability must be designed from the outset, not retrofitted after deployment. This means several things in practice.
Transparency as a Default
Citizens who participate in AI-assisted consultations should know that AI is being used, what it is being used for, and what safeguards are in place. This is not merely a matter of disclosure—it is a matter of informed consent. The OECD's framework for trustworthy AI in government emphasises the need for "fit-for-purpose" AI that includes low-tech alternatives for citizens without digital access. This principle should be extended: participation platforms should be designed so that citizens can engage meaningfully regardless of their familiarity with or trust in AI systems.
Human Oversight at the Point of Consequence
The OECD's June 2026 analysis of 50 civic AI deployments across 22 countries found that sense-making tools and virtual assistants are the most widespread applications—yet most remain in pilot or ad-hoc stages, far from the systemic integration their proponents envision.
The most critical point of intervention is not the AI system itself, but the moment when AI-processed input is translated into policy recommendation. Human oversight must be meaningful—not a rubber stamp on algorithmic output, but a genuine review that can identify and correct systematic biases, surface minority viewpoints that the AI may have underweighted, and ensure that the conclusions presented to policymakers accurately reflect the range of citizen input received.
This requires investment in human capacity, not just AI capability. The efficiency gains from AI sense-making tools are real, but they should be reinvested in the quality of human review, not simply extracted as cost savings. A consultation process that uses AI to process 50,000 submissions but then presents policymakers with a three-page summary without meaningful human review has not improved democratic governance—it has automated its degradation.
Algorithmic Auditing as Civic Infrastructure
The AI systems used in democratic governance should be subject to independent algorithmic auditing—not just for technical performance, but for democratic legitimacy. This means assessing whether the system systematically advantages or disadvantages particular communities, whether its outputs accurately represent the distribution of citizen input, and whether its design choices reflect democratic values rather than commercial or bureaucratic convenience.
Several jurisdictions are moving in this direction. The EU AI Act, fully enforceable as of August 2026, classifies AI systems used in democratic processes as high-risk, requiring conformity assessments, technical documentation, and human oversight. But the Act's focus is primarily on safety and risk management rather than democratic legitimacy per se. A more robust framework would require that AI systems used in civic participation be subject to democratic oversight—not just regulatory compliance.
The GovTech Market and Its Incentives
Any analysis of AI in democratic governance must reckon with the market structure of the GovTech sector. The companies building the platforms that governments use for civic participation—Granicus, CivicPlus, Polco, Go Vocal, and others—are commercial entities with commercial incentives. Their products are designed to be adopted, scaled, and renewed. The metrics they optimise for—engagement rates, submission volumes, platform uptime—are not the same as the metrics that matter for democratic legitimacy.
This is not a criticism of these companies individually. It is a structural observation about the relationship between commercial technology and democratic governance. When governments outsource the infrastructure of participation to commercial platforms, they also outsource some of the design choices that shape what participation looks like. The question of who controls the architecture of democratic engagement is not merely technical—it is political.
The OECD's recommendation to prioritise open-source standards is partly a response to this concern. Open-source platforms allow governments to inspect, modify, and audit the systems they use, rather than depending on proprietary black boxes. But open-source is not a panacea. It requires technical capacity that many governments lack, and it does not automatically resolve the governance questions about who makes design decisions and how those decisions are accountable to democratic processes.
The OECD's June 2026 analysis of 50 civic AI deployments across 22 countries found that sense-making tools and virtual assistants are the most widespread applications—yet most remain in pilot or ad-hoc stages, far from the systemic integration their proponents envision.
Simulation, Prediction, and the Limits of Algorithmic Governance
Among the nine AI applications identified by the OECD, simulation is perhaps the most consequential and the least discussed. AI simulation tools can model the likely impacts of policy decisions—predicting how different communities will be affected, how public services will be stressed, how infrastructure will perform under different scenarios. In theory, this capability could make democratic governance more evidence-based and more equitable.
A 300% increase in eID verification adoption signals that digital democracy is maturing into critical infrastructure—but critical infrastructure demands critical scrutiny, not just critical adoption.
In practice, simulation tools carry significant risks. They depend on historical data that may encode existing inequalities. They make assumptions about human behaviour that may not hold in novel situations. And they can create a false sense of certainty about inherently uncertain futures. When policymakers rely on simulation outputs without understanding their limitations, they risk making decisions that appear evidence-based but are actually driven by the assumptions embedded in the model.
More fundamentally, simulation tools raise questions about the relationship between prediction and democratic choice. If an AI system can predict with high confidence that a particular policy will produce a particular outcome, does that prediction constrain democratic deliberation? Should it? These are not questions that technology can answer. They are questions about the nature of democratic governance itself—about the relationship between expertise and popular sovereignty, between evidence and values, between what is likely and what is right.
Towards a Democratic AI Governance Framework
The challenge of AI in democratic governance is not primarily technical. It is institutional. The question is not whether AI can be made to work in civic participation contexts—it clearly can, and in many cases it already does. The question is whether the institutions of democratic governance can develop the capacity to govern AI, rather than simply being governed by it.
This requires several things. First, democratic institutions need technical literacy—not at the level of individual officials, but at the institutional level. Governments need the capacity to evaluate AI systems, to specify their requirements in democratic rather than purely technical terms, and to hold vendors accountable for the democratic performance of their products.
Second, democratic institutions need new accountability mechanisms. The existing frameworks—procurement rules, data protection law, administrative law—were not designed for AI systems that make consequential decisions at scale. New mechanisms are needed that can assess the democratic legitimacy of AI-assisted processes, not just their technical compliance.
Third, and most fundamentally, democratic institutions need to resist the temptation to treat AI as a solution to the crisis of democratic legitimacy. AI can make participation more efficient, more accessible, and more inclusive. But it cannot substitute for the political will to take participation seriously—to design processes that genuinely influence decisions, to invest in the human capacity to engage with citizen input, and to build the trust that makes democratic governance possible.
The OECD's finding that trust in local government rises from 22% to 69% when citizens feel genuinely heard is not primarily a finding about technology. It is a finding about the relationship between participation and legitimacy. AI can help governments listen at scale. But listening is not the same as hearing, and hearing is not the same as acting. The participation paradox of 2026 is that the tools for democratic engagement have never been more powerful—and the political commitment to democratic governance has never been more contested.
Conclusion: The Constitutional Question
The integration of AI into democratic governance is, at its core, a constitutional question. It concerns the fundamental architecture of how collective decisions are made, who has voice in those decisions, and how power is exercised and constrained. These are not questions that can be resolved by technical standards alone, however important those standards are.
The frameworks being developed in 2026—the OECD's typology of AI in civic participation, the EU AI Act's classification of democratic AI as high-risk, the emerging legislative mandates for civic engagement—represent important steps. But they are steps in a much longer journey. The destination is a form of democratic governance that is genuinely enhanced by AI: more inclusive, more responsive, more evidence-based, and more legitimate. Getting there requires not just better technology, but better institutions, better accountability mechanisms, and a deeper commitment to the democratic values that give governance its legitimacy in the first place.
The kleroterion was a technology of democratic legitimacy. So, potentially, is AI. The difference is that the kleroterion's design was transparent, its operation was public, and its outcomes were accepted as legitimate because the process was understood and trusted. For AI to play a comparable role in democratic governance, it must meet the same standard: not just technically sound, but democratically legitimate.
