Politics · 6 min read
How Citizens' Assemblies Turn Random Selection into Policy Advice
A walkthrough of selection, evidence, deliberation, recommendation, and the separate authority of legislatures and voters.

A temporary public designed for a bounded question
A citizens' assembly brings a relatively small group of residents together to learn, deliberate, and recommend action on a public question. Members are selected through random processes and often stratified so the final group broadly reflects demographic features of a larger population. The assembly does not replace an electorate or legislature. It creates a temporary setting in which ordinary citizens can spend more time with evidence and one another than most voters can devote to a single issue.
The model is one form of “representative deliberative process.” The OECD's 2020 comparative report examined close to 300 cases, including citizens' assemblies, juries, and panels. Its taxonomy requires more than an open town hall: participants are selected to be broadly representative, receive evidence, deliberate for sustained periods, and produce collective recommendations commissioned by a public authority.
Whether an assembly improves democracy depends on design and political context. Random selection can widen participation, but invitations still receive unequal responses. Expert testimony can inform members, but agenda setters choose which experts appear. Recommendations can clarify public judgment, but officials may ignore or selectively use them. The mechanism should be assessed stage by stage.
Recruitment combines chance with correction
Organizers commonly send invitations to a large random sample of addresses or individuals. Interested recipients volunteer to enter a second selection. From that pool, stratified sampling chooses members across variables such as age, gender, geography, socioeconomic position, or attitudes relevant to the topic. The goal is a miniature public in descriptive terms, not a statistically perfect poll.
Self-selection enters before stratification: people who reply may have more time, trust, education, or interest than those who do not. Honoraria, travel reimbursement, childcare, accessibility, and translation can reduce barriers. They cannot make participation costless or force a reluctant invitee to join. A transparent report should publish invitation and response methods, selection variables, replacements, and attrition.
Sample size imposes limits. A group of 100 can mirror several broad demographic distributions, but it cannot contain every intersecting identity or affected experience. Descriptive representation also differs from substantive representation. A member placed in an age or regional category is not authorized to speak for everyone who shares it.
Randomness serves a political purpose even when the final sample is adjusted. It reduces the advantage of campaign organizations, professional advocates, and people accustomed to public meetings. Rotation prevents members from becoming a permanent political class. Stratification then corrects some imbalances produced by chance and response. The trade-off is that every correction requires choices about which dimensions count.
Learning is structured before deliberation begins
Assemblies usually proceed through learning, consultation, deliberation, and decision. A secretariat or advisory group commissions briefing material and witnesses. Members hear experts, advocates, and people with lived experience; question them; then discuss evidence in facilitated small groups and plenary sessions. Recommendations may be adopted by vote, consensus, or a mixture.
Good design exposes disagreement rather than presenting one official answer. Materials should identify established facts, uncertainties, value conflicts, and the interests of witnesses. Participants need time to request additional information. Publishing evidence and recordings allows outsiders to examine whether the learning phase was balanced.
Facilitators are intended to distribute speaking time and keep discussion on task, not determine conclusions. Their phrasing and intervention still matter. Table composition, sequence of testimony, ballot wording, and the threshold for adoption can affect results. The OECD's evaluation guidelines consequently assess process integrity as well as participant satisfaction.
Deliberation differs from polling. A poll estimates opinions at a moment under a defined questionnaire. An assembly asks what a selected group concludes after a structured experience. Members may change their views as they learn or negotiate. That change is part of the method, but it means assembly votes should not be described as direct estimates of how the whole population would vote without the same process.
Ireland shows a chain, not direct rule by assembly
Ireland provides a well-documented national case. The 2016–2018 Citizens' Assembly consisted of a chair and 99 citizens selected to be broadly representative. It considered five topics, including the Eighth Amendment of the Constitution concerning abortion. Members met over several weekends, heard legal and medical evidence and advocacy positions, deliberated, and voted on recommendations.
The assembly recommended constitutional change. A parliamentary committee then considered its report and developed proposals. The Oireachtas passed legislation to hold a referendum. On 25 May 2018, voters approved removing the Eighth Amendment by 66.4 percent to 33.6 percent, according to the official referendum result. Parliament subsequently enacted legislation regulating abortion.
The sequence is important. The assembly did not amend the constitution or write the final law. It supplied recommendations to elected officials; parliament framed the referendum; the electorate decided the constitutional question; parliament legislated. Ireland's official Citizens' Assembly description states plainly that government and the Oireachtas are not obliged to accept assembly recommendations.
The case supports a bounded finding: an assembly can help move a politically difficult question through an institutional chain. It does not prove that the assembly caused the referendum result. Public opinion, advocacy, litigation, party strategy, media coverage, and prior constitutional debates also mattered. Nor did every Irish assembly recommendation become policy.
Subsequent referendums show the danger of treating the model as a guarantee. In March 2024, Irish voters rejected two proposed constitutional amendments on family and care after a Citizens' Assembly on Gender Equality had addressed related issues. The government's wording did not simply reproduce the assembly's recommendations, and campaign dynamics differed. The outcome demonstrates that deliberative advice, legislative translation, and mass voting are separate stages.
The strongest objections concern power around the room
Supporters argue that assemblies reduce partisan performance, give citizens time to learn, and show officials how a diverse group reasons through trade-offs. Participant surveys often report increased knowledge and trust in the process. Comparative case research suggests that high-quality deliberation is feasible under supportive conditions.
Critics ask who sets the question, appoints the advisory group, selects evidence, and decides whether recommendations matter. A government can commission an assembly to delay a decision, seek legitimacy for a preferred policy, or transfer political risk. Even a fair internal process may have little democratic value if officials offer no public response.
There is also a scale objection. One hundred unelected people cannot carry the same authorization as millions of voters. Proponents generally answer that the assembly's authority is advisory and epistemic: it supplies considered recommendations, not sovereign command. That answer is strongest when the mandate is explicit and elected institutions retain visible responsibility.
Evaluation should therefore track outputs and follow-through. Did the sample match stated criteria? Were materials balanced and public? Could members influence the agenda? Did deliberation allow dissent? Did the commissioning authority respond to each recommendation with reasons? Did wider publics understand the process? A favorable headline outcome cannot substitute for those measures.
What assemblies can establish
A well-run citizens' assembly can show what a diverse, supported group concluded after sustained learning under specified rules. It can reveal reasons, compromises, and minority positions that a conventional poll misses. It can widen access to a form of participation usually dominated by organized interests.
It cannot reveal an unmediated “will of the people.” Selection, framing, evidence, facilitation, and voting rules shape the result, as they do in legislatures and referendums in different ways. It also cannot guarantee political adoption or public agreement.
The evidence supports treating assemblies as a complement to representative and direct democracy, especially for bounded questions that benefit from learning and trade-offs. Their legitimacy rests less on the label than on transparent recruitment, plural evidence, independent facilitation, and a defined route for public response. Ireland shows both the mechanism's capacity and its boundary: an assembly can prepare advice, but elected institutions and voters remain responsible for what follows.
