What problem does it solve?
Governments answer a large volume of parliamentary business: written and oral questions, debates, consultations and the bills themselves. Before anyone can answer a question well or draft an explanatory note, someone has to find what has already been said on the topic, by whom, and in what context, inside years of proceedings that are public but not easy to search by meaning rather than keyword. Legislative counsel then has to turn policy intent into a clause, and produce the supporting explanatory material that lets legislators and the public understand what it does.
In the United States, Warren Burke, who leads the House Office of the Legislative Counsel, testified that requests for legislation the office received were up 72% in 2025 compared with two years earlier. An anonymous congressional staffer, quoted in the same reporting, said outside groups are increasingly "going to Claude or ChatGPT to draft the legislative language itself", so the office spends more time fixing that language than it would take to draft it from scratch. The two tools on this page take a different approach: a government owns and governs the AI tool built for its own legislative and parliamentary work, whether built in house or with a partner, with a legal drafter or policy official still responsible for what is filed or said.
- Warren Burke, who leads the US House Office of the Legislative Counsel, testified that requests for legislation the office received were up 72% in 2025 compared with two years earlier.'It's Absolutely Terrifying': AI Is Reportedly Slopping Up the Bills in Congress (2026)
How does it work?
- Index the public record. The tool ingests the legislature's own data, such as debates, written questions and answers, and legislation, usually through an official parliamentary API, and keeps it current with a daily update.
- Search by meaning. A policy official describes what they need in their own words; the tool embeds the query and ranks debates, questions and answers by relevance rather than exact keyword matches.
- Summarise with citations. A language model turns the matching results into a short, cited summary, with every claim linked back to the original record so the user can check it directly.
- Draft a first version. For a narrower task, such as an explanatory note for a Bill clause, the tool is given the Bill text and the Act it amends, and generates a draft summary of the clause's effect for a drafter to open and rework.
- A named official finishes the work. The drafter or policy official edits, corrects and takes ownership of the final text; nothing the tool produces is filed or sent to a parliamentarian without that review.
- Audience
- Employee facing
- Autonomy
- Copilot
- Adoption
- Emerging
- Channels
- Internal tools
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
| KPI | Median | Reported range | Data points | Claimed by |
|---|---|---|---|---|
| Users served | Not pooled | about 400 | 1 | 1 organization |
Value drivers: Employee productivity, Speed and cycle time.
Indicative value
A government department with 40 policy and legislative staff handling parliamentary business
USD 52,800 to USD 431,200
Policy and legislative staff time cost avoided per year
How this is calculated
Formula: staff * hoursPerWeek * timeSavedShare * weeksPerYear * costPerHour. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Policy and legislative staff answering parliamentary business staff, staff | 40 | 40 | The reference department. |
| Hours per staff member per week spent on parliamentary research and drafting hoursPerWeek, hours per staff member per week | 5 | 10 | Editorial assumption for a policy team handling correspondence, debate preparation and explanatory notes. Replace with your own time study. |
| Share of that research and drafting time saved timeSavedShare, fraction of research and drafting time | 0.15 | 0.35 | Conservative against the evidence on this page, because the UK government describes Parlex only as reducing research time and New Zealand's Parliamentary Counsel Office has not published a percentage or decided whether the results warrant further investment. Replace with your own measurement once you pilot. |
| Working weeks per year weeksPerYear, weeks | 44 | 44 | Standard allowance for leave and public holidays. |
| Fully loaded cost per policy or legislative staff hour costPerHour, USD per hour | 40 | 70 | Editorial assumption for a civil service policy grade. Replace with your own cost. |
What it leaves out: Gross time value only. It leaves out the cost of running and governing the tool, the review every draft still needs from a named official, and any change in the volume or complexity of parliamentary business the department handles.
Who already uses it?
2 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
Department for Science, Innovation and Technology (Incubator for Artificial Intelligence)
United Kingdom · Government and public sector · 2025
Parlex is a semantic search and generative summary tool built by the UK government's Incubator for Artificial Intelligence for parliamentary data: Hansard debates, member profiles and written parliamentary questions. It lets policy teams, private offices and bill teams search this archive by meaning rather than keyword, and generate a cited summary of what it finds. The tool does not decide or draft an answer to a parliamentary question itself; it is a research aid that users must verify against the original record. It is one of the tools in the government's "Humphrey" AI package for civil servants, launched in January 2025, and is in beta.
- Users served: about 400, at the time of the transparency record
"It has ~400 registered pilot Civil Service users across a variety of government departments and roles."
Claimed by: organization
Parliamentary Counsel Office (New Zealand)
New Zealand · Government and public sector · 2025
New Zealand's Parliamentary Counsel Office ran a six month research and development programme with five companies to test where AI could help its work. One stream, a proof of concept built with Catalyst, explored whether a large language model could generate a useful first draft of the clause by clause explanatory note that accompanies an amendment Bill, given the Bill text and the principal Act it changes. The team tested several model sizes, hosted where possible in New Zealand for data and security control, and published a report and open source code. The office describes the result as a proof of concept only, says it can "now begin a formal evaluation of the tool", and has not decided whether the results are good enough to warrant further investment.
No outcome disclosed.
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- Full text of the legislature's proceedings, questions and answers, and legislation, ideally through an official parliamentary data API
- The Bill text and the Act it amends, structured by clause, for any drafting feature
- A style guide or drafting manual for explanatory notes, briefings or correspondence
Systems to integrate
- Parliamentary or legislature data API for debates, questions and legislation
- The department's own document and knowledge management system for prior briefings
- Legislative drafting or bill management software used by legal drafters
Complexity: Medium
Search and summarisation over public parliamentary data is the straightforward part. The harder work is connecting a private departmental knowledge base for briefings without exposing information the tool's users are not authorised to see, and building drafting features that follow the specific legal conventions and citation rules the legislature already uses.
- 1
Start with search, not drafting
Launch with semantic search and cited summarisation over public parliamentary and legislative records before adding any generation of new legal text, because a poor search result is easy to spot and a drafting error is not.
- 2
Keep every output a working note
Treat every AI output as a draft for a named person to rework, the way New Zealand's own proof of concept let drafters regenerate and compare several explanatory note drafts before choosing what to keep, rather than a text anyone can file directly.
- 3
Ground every summary with a citation
Link every generated summary back to the original debate, question or clause it is based on, so a policy official can check it in seconds instead of trusting it on faith.
- 4
Pilot with a small, named group first
Register a limited, named group of users, as the UK government did with about 400 pilot Civil Service users spread across departments and roles, and collect structured feedback before widening access to the rest of the organisation.
- 5
Decide the quality bar before you pick a model size
New Zealand's Parliamentary Counsel Office found that larger models tended to give more fit for purpose drafts than smaller ones, but cost more and made errors that were harder to spot. Decide how good a draft you need before you choose a model size, rather than defaulting to the biggest one available.
Guardrails
- AI generated text is visibly marked wherever it appears, separate from the original record
- No draft leaves the tool as a final answer; a named official signs off before it is used
- The tool states plainly that it is a research starting point, not a definitive source
- Access is restricted by security clearance and department, matching the sensitivity of the data
KPIs to instrument
- Time from a request to a usable first draft or summary
- Share of registered users who return weekly
- Editing distance between the AI draft and the final approved text
- User reported trust and accuracy from structured feedback
Human in the loop
A policy official or legislative drafter reviews, edits and takes ownership of every summary or draft before it appears in a briefing, a Bill or an answer to a parliamentarian. The tool never submits or publishes anything on its own behalf.
Common failure modes
- A summary that misses the political context
- The UK government's own transparency record for Parlex warns that the tool "may not capture nuances in debate, such as tone and broader political context" that a human researcher would catch. Keep a human judgment step before anything reaches a minister or the public record.
- Confident search results that are not actually relevant
- Semantic search can rank a loosely related debate above the one that matters most. Show the relevance ranking and the original text alongside the AI summary, not the summary alone.
What are the risks and rules?
EU AI Act
Minimal risk
A research and drafting aid that a policy official or legal drafter reviews before use is not listed in Annex III. Article 50(4) exempts AI generated text that has undergone human review and carries a named person's editorial responsibility from the disclosure duty that otherwise applies to AI generated content published on matters of public interest, which is how both tools on this page are designed to work. If a tool moved from drafting support to deciding the outcome of a request for a public service or benefit, Annex III point 5(a) could apply instead.
Guidance
- Algorithmic Transparency Recording Standard hub (UK government, Europe). UK departments publish transparency records for tools such as Parlex, including their phase, registered user numbers and underlying model.
- DSIT - Parlex (algorithmic transparency record) (Department for Science, Innovation and Technology, Europe). States that Parlex does not function in a decision making capacity and that users must verify and validate its outputs before relying on them.
Controls to put in place
- AI generated content visibly marked wherever it is shown
- Named official reviews and approves every draft or briefing before use
- Access limited by security clearance and need to know
- Source citations kept attached to every generated summary
Frequently asked questions
- Does a tool like Parlex write the answer to a parliamentary question?
- No. The UK government's own transparency record for Parlex says the tool does not function in a decision making capacity and that users must verify and validate its findings. It retrieves and summarises public parliamentary information; the civil servant still writes and takes responsibility for the answer.
- Has any government measured how much time this saves?
- Not yet with a published figure. The UK government's Parlex had about 400 registered pilot users according to its own transparency record, and New Zealand's Parliamentary Counsel Office reported that its proof of concept produced first drafts a drafter could refine, but neither has published a percentage or an hours figure. Parlex is in pilot; New Zealand's work is an internal research and development proof of concept, not yet a pilot with live users.
- Can AI draft an actual bill?
- Not on its own, and neither example here tries to. New Zealand's Parliamentary Counsel Office scoped its trial to the explanatory notes that accompany an amendment Bill, not the legal text of the Bill itself, and its own report says the office has yet to determine whether the results are good enough to warrant further investment.
How to cite this page
Blits.ai AI Use Case Library, "AI for legislative drafting and parliamentary question support", last verified 29 September 2026, https://www.blits.ai/ai-use-cases/legislative-drafting-and-parliamentary-question-support. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 29 September 2026: First published