Citizens' assemblies after two decades: what happened to their recommendations

Between 2004, when 160 randomly selected residents of British Columbia sat down to redesign their province's electoral system, and 2026, citizens' assemblies have moved from academic curiosity to a routine instrument of government in parts of Europe, Canada and Australia. This study asks a narrow question that enthusiasm often skips: what actually happened to their recommendations? It reviews seven well-documented national and sub-national assemblies, the OECD's 2020 comparative dataset of 289 deliberative processes, and the main academic attempts to measure impact. The pattern is consistent but uncomfortable for both champions and sceptics. Assemblies that answer a bounded question commissioned by an authority that has pre-committed to a route for decision — a referendum, a parliamentary committee, a published response — often see their proposals enacted. Assemblies with very broad remits, or whose sponsors made no binding commitment, mostly see their proposals filtered, delayed or quietly shelved. Measured impact also depends heavily on how impact is measured.

Key findings

• Uptake is common for local, bounded processes. Of the 55 cases in the OECD database with implementation data (1986–2019), authorities implemented more than half of the recommendations in 76% and all of them in 36%; in 11% none were implemented. The sample is small and self-reported.
• Selective uptake is the norm, not the exception. A study of 571 proposals from 39 Spanish local participatory processes found roughly a third fully implemented, a third partially implemented and a third rejected; cheap proposals, proposals that did not challenge existing policy and proposals backed by both politicians and civil servants did best.
• The clearest national successes ran through a referendum. Ireland's assemblies on marriage equality (2013) and abortion (2017) fed into referendums that passed with 62.07% (2015) and 66.40% (2018). But assemblies also produced proposals voters rejected: British Columbia's STV (57.7% in 2005, short of a 60% threshold; 39.1% in 2009) and Ontario's MMP (36.9% in 2007).
• Broad climate remits produced weak follow-through. France's Citizens' Convention for Climate graded the government's response at 3.3 out of 10 in February 2021; a WWF France count reported in August 2026 found 24 of its 149 proposals fully applied. Evaluators of Climate Assembly UK concluded that its very wide scope weakened its impact on Parliament and government.
• Germany's first Bundestag assembly had no legislative effect. The 160-member Bürgerrat on nutrition delivered nine recommendations in February 2024; a year later the responsible committee found no majority for anything beyond taking note, and none came to a vote.
• Permanence helps procedure more than outcomes. Ostbelgien's standing citizens' dialogue guarantees committee hearings, a political response and a later review for every assembly, but its public record does not classify which recommendations were implemented.
• Measurement is the weak link. Most studies infer impact by matching the text of recommendations to later policy documents. Vrydagh (2022) shows this "congruence" method both over- and under-states influence unless it accounts for decision-makers' prior positions and is checked through interviews.

Research question and scope

The question is: when a randomly selected citizens' assembly makes recommendations to a public authority, what happens to them, and which design features are associated with uptake? "Uptake" here means that a recommendation is adopted as law, regulation, budget or official policy, or is put to the public in a binding vote. Softer effects — changes in elite discourse, learning among members, shifts in public opinion — are discussed only where they bear on uptake.

The scope is limited to representative deliberative processes: bodies of citizens chosen by lot, usually stratified by age, gender, region and education, who learn, hear witnesses and deliberate before agreeing collective recommendations. Open public meetings, online consultations and participatory budgeting (covered in a separate study) are excluded. The case set covers the processes the brief identified — British Columbia (2004), Ontario (2006–07), Ireland (2012–14 and 2016–18), France (2019–21), the United Kingdom (2020), Ostbelgien (since 2019) and the German Bundestag (2023–24) — plus the OECD's cross-national data.

Background

The modern citizens' assembly combines two older ideas. One is the jury: a small group of lay people, chosen by lot, given evidence and time, trusted to reach a judgement. The other is the deliberative poll and citizens' jury developed from the 1970s onwards, which showed that ordinary people can handle complex policy information when given structured support. British Columbia's 2004 Citizens' Assembly on Electoral Reform made the format famous by giving a lottery-selected body a real constitutional task and a direct path to the ballot.

Since then the numbers have grown quickly. The OECD's 2020 report, the first systematic comparison, drew on 289 cases (282 in OECD countries) from 1986 to October 2019. Processes ran at every level of government — 52% local, 30% regional, 15% national and 3% international — and the report identified twelve distinct models, from short citizens' juries to permanent bodies such as the Ostbelgien model. Participants gave on average 3.7 days of their time spread over 6.6 weeks. A separate European inventory, the POLITICIZE dataset, coded 105 deliberative mini-publics held in 18 European countries between 2000 and 2020.

Public opinion is supportive in principle but conditional. A survey of 14,043 respondents in 15 Western European countries found the strongest support among the less educated, those with a low sense of political competence and those with anti-elite attitudes — and found that support depended partly on what respondents expected assemblies to decide. That conditionality matters for impact: governments that ignore assemblies risk disappointing exactly the citizens the format is meant to re-engage.

Method and sources

This is a structured literature and document review, not new fieldwork. For each case we traced the recommendation, the commissioning authority's formal commitment, the decision route and the documented outcome, using official records (assembly reports, parliamentary pages, electoral results) wherever possible. Cross-national patterns come from the OECD dataset and from peer-reviewed studies that code the fate of individual proposals. Three cautions apply throughout:

1. Selection. Well-documented cases are disproportionately those that were large, national or successful enough to attract researchers.
2. Attribution. A recommendation that becomes law may have been on the government's agenda anyway; a rejected recommendation may still shape a later reform.
3. Self-reporting. The OECD implementation data were supplied by organisers and commissioning authorities, who have reasons to report success.

One source — the WWF France count of the French convention's proposals — could be read only through press coverage (Reporterre, August 2026); its classification method has not been checked.

Finding 1: what the cross-national data say

The OECD dataset

The OECD collected implementation data for 55 of its 289 cases. Its headline figures are encouraging:

The OECD itself is careful: it describes the data as "limited", calls for more research on whether authorities "cherry-pick" proposals that suit their existing agenda or cost less, and stresses that more evidence is needed before the conclusion can be called robust. Response practices were also uneven. In 66% of the 104 cases with data, authorities discussed the recommendations face to face with participants; in 42% they published a response through official or media channels without meeting participants; in 24% they followed up directly with participants after the public response. The commissioning context helps explain the positive skew: every OECD case was commissioned by a public authority with the power to act, and most were local processes on concrete questions. The often-cited example, the 2014 Melbourne People's Panel of 43 residents on the city's ten-year, AUD 5 billion financial plan, saw the council accept 10 of 11 recommendations and annex the panel's text to the final plan.

Proposal-level evidence

A stronger design comes from Font, Smith, Galais and Alarcón (2018), who followed 571 individual proposals from 39 participatory processes in Spanish municipalities, including but not limited to deliberative ones. About 36% were fully implemented, and roughly 32% each were partially implemented or rejected. Only four of the 39 processes implemented everything and three implemented nothing; the rest picked and chose. The strongest predictor was internal support: a proposal backed by both politicians and civil servants had a predicted 46% chance of full implementation, against 17% for one lacking that support. Costly proposals and those that challenged existing policy were much more likely to be abandoned — non-challenging proposals had a 42% chance of full implementation, challenging ones 26%. Process quality mattered too: proposals from processes meeting all three of the authors' quality criteria had a 45% chance of full implementation.

Why the measurement matters

Jacquet and van der Does (2021), reviewing 35 years of research on the consequences of mini-publics, found that the literature concentrates on proximate effects — on members and on the immediate policy decision — and has said little about wider or longer-term consequences. Vrydagh (2022) goes further and questions the dominant measurement technique. Most impact studies compare the wording of citizen recommendations with later policy documents. Using the Brussels "Make Your Mobility" citizens' panel as a test case, he shows that this desk-based congruence approach lacks reliability: it can credit a panel with ideas decision-makers already held, and it can miss less visible influence. He proposes a "sequential impact matrix" that accounts for decision-makers' pre-existing preferences, and a mixed method that adds interviews with the officials who followed up the process. The implication is that headline uptake rates, including the OECD's, should be read as rough indications rather than measurements.

Finding 2: electoral reform — when the route is a referendum

British Columbia, 2004–2009

The British Columbia Citizens' Assembly had 160 members — one man and one woman from each of the province's 79 electoral districts plus two Aboriginal members — with Jack Blaney as chair. It spent January to March 2004 learning, held public hearings in May and June, deliberated from September to November and submitted its final report on 10 December 2004, recommending the single transferable vote (BC-STV). The government had committed in advance to put the recommendation to voters, but set a double threshold: 60% of votes province-wide and a simple majority in 60% of districts.

At the referendum of 17 May 2005, 57.7% voted for BC-STV — above a majority but below the 60% bar. A second referendum in 2009 drew only 39.1% support, and a third, government-designed referendum in 2018 saw 61.30% favour keeping first-past-the-post. The assembly therefore achieved the hardest part of uptake — a binding public vote — and still produced no change. Its case shows that a referendum route transfers the decision to an electorate that has not deliberated, and that supermajority thresholds can defeat a proposal that a majority supports.

Ontario, 2006–2007

Ontario copied the model with 103 members, one per riding. The assembly recommended mixed-member proportional representation by 94 votes to 8. At the referendum held with the provincial election on 10 October 2007, only 36.9% voted for change; MMP won a majority in 5 of 107 ridings, all in Toronto, against a threshold of 60% province-wide and majorities in 64 ridings. Stephenson and Tanguay's survey analysis attributes the defeat less to opposition to proportional representation than to low attention and limited information: better-informed voters were far more supportive, but even simulated full information would have left support well short of the threshold.

Finding 3: Ireland — the strongest case for uptake

Two bodies, two designs

Ireland's Convention on the Constitution (2012–14), chaired by Tom Arnold, was a mixed body of 66 randomly selected citizens and 33 politicians plus a chair. It considered most topics in a single weekend, produced nine reports and 38 recommendations, 18 of which would have required constitutional change, and received a government response to every report. The government accepted several institutional recommendations, including a secret ballot to elect the Ceann Comhairle (the speaker of the lower house) and an Electoral Commission, and put two proposals to referendum on 22 May 2015. Marriage equality passed with 62.07% of the vote; lowering the age of candidacy for the presidency from 35 to 21 was rejected by 73.06%.

The Citizens' Assembly of 2016–18, chaired by Supreme Court judge Mary Laffoy, had 99 citizens and no politicians. It spent five weekends on the Eighth Amendment, the constitutional provision that had effectively banned abortion. At its final meeting on 22–23 April 2017, members voted 79 to 12 that the provision should not be retained in full, and 51 to 38 to replace it with a clause authorising parliament to legislate. In a further ballot, 64% recommended that termination without restriction as to reasons should be lawful, with a range of gestational limits. Unlike the Convention, the Assembly's report went to a special joint parliamentary committee, whose cross-party majority broadly endorsed legal abortion up to 12 weeks. The referendum of 25 May 2018 passed with 66.40%, and the Health (Regulation of Termination of Pregnancy) Act 2018 followed in December.

What made the difference

Farrell, Suiter and Harris (2019) argue that the significance of the Irish experience lies in repetition: two successive mini-publics feeding into referendums suggested a degree of "systematisation" of deliberation in constitutional politics. Three design features stand out:

• A pre-committed route. The parliamentary resolutions establishing both bodies obliged the government to respond to each recommendation and to state a timetable for any referendum it accepted.
• A parliamentary bridge. For the Eighth Amendment, a joint committee of both Houses translated the assembly's report into a legislative position, giving elected politicians ownership.
• Topics already on the agenda. Marriage equality and abortion were salient, values-based questions on which party leaders were divided or reluctant to lead. The assemblies arguably gave politicians cover to act — a function that is real but not the same as independent agenda-setting.

The limits

Irish uptake has not been uniform. The government's own wording of the March 2024 referendums on family and care, which followed the Citizens' Assembly on Gender Equality, was rejected by large margins on 8 March 2024 — 67.69% No on the family amendment and 73.93% No on the care amendment, according to the official results. Later assemblies show the same unevenness: the atlas's own tracking shows high self-reported uptake for the biodiversity assembly and a still-pending response to the Dublin assembly's mayoral proposal.

Finding 4: climate assemblies and broad remits

France: "without filter" meets politics

The Citizens' Convention for Climate (Convention Citoyenne pour le Climat) was asked how to cut France's greenhouse-gas emissions by at least 40% by 2030 compared with 1990, "in a spirit of social justice". It met from October 2019 and, after delays caused by COVID-19, approved 149 measures in June 2020. President Emmanuel Macron had promised to pass its proposals to parliament, referendum or regulation "without filter"; he immediately set aside three — a change to the constitution's preamble, a 4% tax on corporate dividends and a lower motorway speed limit — and committed to the other 146. A climate bill was submitted on 10 February 2021, and the Climate and Resilience Law followed in August 2021. The constitutional referendum the president had floated never took place because the necessary parliamentary support was lacking.

At their final session in February 2021, members graded the government's follow-up at an average of 3.3 out of 10 (median 3), and 2.5 on whether it would achieve the 40% target in a spirit of social justice. Giraudet and colleagues (2022), a team of researchers who observed the proceedings, concluded that the steering bodies' substantial input did not undermine the citizens' independence, and that the process built consensus among members but did little to build support in the wider public: awareness of the convention rose from 22% to 42% between survey waves, but 74% of aware respondents doubted its representativeness. A WWF France count reported in August 2026 found 24 proposals fully applied, 92 partially and 33 not at all (the last figure includes the three set aside in 2020); the count was published around the fifth anniversary of the Climate and Resilience Law. Classification of "partial" application is a judgement call, but every available count points the same way: the "without filter" promise was not kept.

United Kingdom: an advisory assembly to Parliament

Climate Assembly UK was commissioned by six House of Commons select committees to ask how the UK should reach net zero by 2050. Its 108 members met over six weekends from late January to mid-May 2020 and published The path to net zero on 10 September 2020. Recruitment produced 1,748 positive responses to random postal invitations, a 5.8% response rate; members were stratified by age, gender, ethnicity, education, region, urban or rural location and level of concern about climate change, and were paid £150 per weekend. The committees asked the government to respond, but there was no pre-committed decision route. Elstub, Carrick, Farrell and Mockler's evaluation found that attempts to match the assembly's scope to the scale of climate change compromised members' learning, the coordination of recommendations, members' endorsement of them and their impact on Parliament and government; they suggested more democratic agenda-setting as a remedy.

Finding 5: Germany and Ostbelgien — institutional homes, uncertain effects

The Bundestag's nutrition assembly

The German Bundestag decided on 10 May 2023 to set up its first own citizens' assembly, on "Nutrition in Transition" (Ernährung im Wandel). Its 160 members, drawn by stratified lot from 62 municipalities, met from 29 September 2023 to 14 January 2024 in three in-person weekends and six online evenings, and handed nine recommendations to the Bundestag President on 20 February 2024 — the first being free lunches for all children in schools and day care. The Bundestag debated them, and the agriculture committee held four public expert hearings. Its report of 11 February 2025 recorded that there was no majority for any committee decision beyond taking note, that the early end of the parliamentary term prevented consideration of the remaining recommendations, and that none had been implemented; it left the next Bundestag free to take them up. Trade press reported in November 2025 that the Bundestag administration had dissolved the staff unit for citizens' assemblies, despite a coalition-agreement commitment to continue the format.

Ostbelgien's permanent dialogue

Ostbelgien — the German-speaking Community of Belgium — took a different route: a decree adopted in February 2019 created a standing Citizens' Council of 24 former participants, which chooses topics and convenes assemblies of about 25 to 50 residents (in practice around 30). Niessen and Reuchamps (2022) describe the core commitment: recommendations are discussed in a joint committee with members, MPs and the responsible minister, who must then say whether and how they will be implemented and justify any rejection. The public record shows the procedure working on schedule. Assemblies on care (recommendations presented October 2020), inclusive education (2021), housing (2022), digital skills (2022), integration (2023), pupils' competences (2024) and older people (December 2025) each received a political response within months and, for the first six, a later review session. An assembly on spatial planning meets in autumn 2026. What the record does not show is a count of implemented recommendations: the review asks how far recommendations "inspired" politicians' work, a softer test than enactment.

A comparison of the cases

What the evidence does not show

• It does not show that assemblies cause the reforms they recommend. In Ireland, change on marriage and abortion had broad social momentum; the assemblies plausibly accelerated and legitimised it. Establishing a counterfactual is not possible with the available data.
• It does not show that high uptake equals good policy. Uptake measures whether authorities agreed, not whether the policy worked. None of the studies reviewed here evaluates the downstream policy outcomes of assembly-originated measures.
• It does not show a representative picture of all assemblies. The OECD implementation data cover 55 of 289 cases; proposal-level studies such as Font et al. come from a single country and mix deliberative and non-deliberative processes.
• It does not show that low uptake means failure. British Columbia's and Ontario's assemblies produced coherent proposals and a public vote; their defeat is evidence about thresholds and referendum campaigns as much as about the assemblies.
• It does not establish the effect of permanence. Ostbelgien and similar standing bodies are too recent, too small and too weakly evaluated to show whether institutionalisation raises uptake.

Implications for institution design

The cases point to a short list of features associated with uptake. None guarantees it.

1. Agree the decision route before the assembly starts. The highest uptake came where the sponsor committed in advance to a specific next step — a referendum, a joint committee, a published response with reasons. Promises of unconditional adoption ("without filter") are worse than precise, limited commitments, because they cannot be kept.
2. Keep the question bounded. Climate Assembly UK's evaluators and the French experience suggest that very wide remits dilute learning and diffuse accountability. A single clear question, or a short list, produces recommendations that can be costed and acted on.
3. Build a bridge to elected representatives. Ireland's joint committee and Ostbelgien's joint sessions give politicians a formal role in translating recommendations. Mixed bodies (Ireland's Convention, Brussels' deliberative committees) are an alternative, at some cost to citizen independence.
4. Require reasons for rejection. The OECD's own good-practice principles and Ostbelgien's decree both emphasise a public, reasoned response. Reason-giving does not force adoption, but it makes cherry-picking visible.
5. Plan for the wider public. Where the final decision is a referendum or depends on public acceptance, the assembly's legitimacy outside the room matters. The French and Ontario cases show that low awareness and doubts about representativeness can sink proposals; publicity, broadcast proceedings and voter information (as in Oregon's Citizens' Initiative Review) are part of the design, not an afterthought.
6. Fund evaluation with a counterfactual in mind. Pre-registration of what counts as uptake, tracking of each recommendation and interviews with officials, as Vrydagh proposes, would turn anecdote into evidence.

Open questions

• Does institutionalisation (permanent councils, statutory routes) raise uptake, or merely make non-uptake more orderly?
• Do assemblies change what governments do, or mainly when and how they do things already under consideration?
• How much does the public's view of an assembly depend on knowing about it, and can that awareness be built cheaply?
• Are referendums the best complement to assemblies, given that the deciding electorate has not deliberated?
• What happens to assemblies when the political sponsor loses power, as in Germany after 2025 and Madrid after 2019?
• Can a cross-national dataset of recommendation-level outcomes, coded consistently, be built and kept current?

Sources

1. OECD (2020). Innovative Citizen Participation and New Democratic Institutions: Catching the Deliberative Wave. Paris: OECD Publishing. Executive summary; chapter 4 (response and implementation, Figures 4.8–4.9).
2. Paulis, E., Pilet, J.-B., Panel, S., Vittori, D. & Close, C. (2021). The POLITICIZE dataset: an inventory of deliberative mini-publics (DMPs) in Europe. European Political Science 20(3), 521–542.
3. Vrydagh, J. (2022). Measuring the impact of consultative citizen participation: reviewing the congruency approaches for assessing the uptake of citizen ideas. Policy Sciences 55(1), 65–88.
4. Font, J., Smith, G., Galais, C. & Alarcón, P. (2018). Cherry-picking participation: explaining the fate of proposals from participatory processes. European Journal of Political Research 57(3), 615–636.
5. Jacquet, V. & van der Does, R. (2021). The consequences of deliberative minipublics: systematic overview, conceptual gaps, and new directions. Representation 57(1), 131–141.
6. Citizens' Assembly on Electoral Reform (British Columbia). Citizens' Assembly in Action. University of British Columbia archive.
7. Legislative Assembly of British Columbia. 2005, 2009 and 2018 electoral referenda.
8. Stephenson, L. B. & Tanguay, B. (2009). Ontario's Referendum on Proportional Representation: Why Citizens Said No. Montreal: Institute for Research on Public Policy.
9. The Citizens' Assembly (Ireland). Information note on differences between the Convention on the Constitution and the Citizens' Assembly.
10. The Citizens' Assembly (Ireland). Results of the ballots on the Eighth Amendment, 22–23 April 2017.
11. Farrell, D. M., Suiter, J. & Harris, C. (2019). 'Systematizing' constitutional deliberation: the 2016–18 citizens' assembly in Ireland. Irish Political Studies 34(1), 113–123.
12. Department of Housing, Local Government and Heritage (Ireland). Referendum Results 1937–2024 (official results booklet, linked from the Referendum Returning Officer's site); figures cross-checked against ElectionsIreland.org, Summary of Referendums. 2015 marriage equality 62.07% Yes, presidential age 73.06% No; 2018 termination of pregnancy 66.40% Yes; 2024 family 67.69% No, care 73.93% No.
13. The Irish Times (9 March 2024). As it happened: Ireland votes No on family and care referendums.
14. Giraudet, L.-G., Apouey, B., Arab, H. et al. (2022). "Co-construction" in deliberative democracy: lessons from the French Citizens' Convention for Climate. Humanities and Social Sciences Communications 9, 207.
15. Reporterre (August 2026). Seules 24 des 149 mesures de la Convention citoyenne pour le climat sont vraiment appliquées (reporting a WWF France analysis).
16. Climate Assembly UK (2020). The path to net zero: About Climate Assembly UK. House of Commons.
17. Elstub, S., Carrick, J., Farrell, D. M. & Mockler, P. (2021). The scope of climate assemblies: lessons from the Climate Assembly UK. Sustainability 13(20), 11272.
18. Deutscher Bundestag (2025). Abschlussbericht zum Bürgerrat „Ernährung im Wandel" übergeben. Textarchiv, 11 February 2025.
19. Lebensmittel Praxis (November 2025). Bürgerbeteiligung: Bundestag löst Stabsstelle für Bürgerräte auf.
20. Niessen, C. & Reuchamps, M. (2022). Institutionalising citizen deliberation in parliament: the permanent citizens' dialogue in the German-speaking Community of Belgium. Parliamentary Affairs 75(1), 135–153.
21. Bürgerdialog in Ostbelgien. Stand der Dinge (status of all citizens' assemblies).
22. Pilet, J.-B., Bol, D., Vittori, D. & Paulis, E. (2023). Public support for deliberative citizens' assemblies selected through sortition: evidence from 15 countries. European Journal of Political Research 62(3), 873–902.