Starting the Conversation on AI: A Session Guide for Parliamentary AI Working Groups

Download (PDF)

About this Guide

Parliaments around the world are creating internal working groups, task forces, and committees to consider what AI means for their institution. Some have a formal mandate from the Speaker, the Presidium, or the Clerk; others start as a few staff from IT, the library, and the procedural offices who begin meeting on their own. Many begin with the same question: Where do we start?

This guide gives those groups a structured way to answer it. It is a session plan: each member reads a short briefing beforehand, and the group then works through a set of questions together across two sessions, from agreeing what it is there to do to designing a small, safe test of an AI tool. It is written for the people who will lead the work inside the institution and requires no technical expertise.

What This Guide Does

This guide helps a working group identify:

1. Where people already use AI

2. Where AI could add value next

3. What foundations the parliament needs before expanding its use

A group can complete the discussion in two sessions of about 90 minutes each, without outside support, or in a single session of three hours.

The discussion goes beyond individual AI tool adoption and into how the parliament manages change, stores and retrieves records, controls access to information, and applies its existing rules. The answers can inform internal AI policies, help turn a small pilot into routine practice, and guide new ways of working.

Who Is This For

Every parliament operates differently. Some have fewer than 15 members; others have more than 400. Their staffing, budgets, procedures, and traditions also vary. This guide can support a whole parliament, a single committee, a department, or one Member’s office. When the guide refers to “your parliament,” interpret it as the group you choose to examine. Adapt, skip, or edit any part of the guide to fit your context. Rather than trying to create a perfect plan, consider this a practical first step that the group can take with confidence.

How To Use This Guide

Stage	Time	What Happens
Before Session 1	15 minutes	Each member reads Part One individually. It covers why this work matters, guiding principles, safe use today, and key terms.
Session 1Mandate and Starting Point	≈ 90 minutes	The group works through Sections 1 to 3 together, then fills in the first three rows of the summary in Section 7. One person should take notes in the response boxes as the group holds the discussion.
Between Sessions	To be decided by the group	The group could share its draft mandate with whoever it reports to and get their agreement on the mandate before the group returns for Session 2.
Session 2Starting Group and Establishing a Pilot	≈ 90 minutes	The group works through Sections 4 to 6 together, then completes the Section 7 summary and agrees on next steps.
After the Sessions	Ongoing	The group sends the individual check-in (Appendix A) to its target group, then moves into training and a pilot.

Note: All responses are for the working group’s internal use unless the group decides otherwise.

About POPVOX Foundation

POPVOX Foundation is a 501(c)3 nonpartisan nonprofit organization with a mission to help democratic institutions keep pace with a rapidly changing world. Through publications, events, prototypes, and technical assistance, the organization helps public servants and elected officials better serve their constituents and make better policy..

Part One: Before the Session

Read this section before Session 1.

Why this matters

Parliaments face a pacing problem: technology moves faster than the institutions meant to understand, use, and oversee it, and a legislature can only keep pace with what it understands. The problem has three dimensions:

1. Members and staff need to understand AI well enough to make informed policy decisions about it,

2. the institution needs modern tools to do its own work, and

3. parliament needs to keep pace with an executive branch that is adopting these tools at a faster pace.

Parliaments also hold something no one else does: the official record of how a country makes its laws. As the public moves from websites to social media to asking AI tools their questions, that record must be current, accurate, and easy for computers to read. Otherwise, AI-generated answers may rely on less authoritative sources. Lobbyists, advocacy groups, and citizens already use AI to track and respond to legislation, and the institution needs to be ready to respond at the speed questions now arrive.

A structured conversation like this one lets a parliament make choices about its future with purpose. Without one, AI still arrives: staff try tools on their own, companies pitch products, and the institution becomes strictly reactionary. A group that has mapped its own starting point can decide what innovation it will adopt, what institutional traditions it will protect, and what procedures it will leave in place.

Guiding principles

Support people.

AI tools serve the staff and Members who do the work, and those people help decide where the tools fit. The human always owns the final product.

Know what the technology can and cannot do.

Clear understanding comes before adoption, and it is the best defense against misuse. Treat every AI output as a first draft to verify, not a final answer.

Tradition protects the institution.

Parliamentary procedure protects the integrity of the institution, and some steps are slow on purpose. Name what must not change, then look for help around it.

Parliament is a learning institution.

Legislatures constantly absorb new information and adapt. That makes them well built to guide this transition rather than simply react to it.

Before anything else: using AI safely today

Staff and Members in many parliaments already use AI tools informally, often without guidance. The working group should know where this happens. Until your parliament adopts its own policy, a few simple rules protect people and the institution:

• Never put personal or sensitive information into an AI tool your parliament has not approved. This includes constituent names and case details, confidential committee material, personnel matters, and unpublished drafts.

• Verify everything before it goes into official work, especially names, dates, numbers, citations, and anything touching the official record.

• A person remains responsible for every output. AI assists; people decide.

• When unsure, ask before you paste anything into an AI tool.

Key terms

A shared vocabulary helps a working group move faster. These are the terms that come up in the sessions. For fuller definitions and more terms, POPVOX Foundation’s AI Terms for Parliamentary Work is available upon request.

Term	What It Means
AI use policy	Written guidance on how staff and Members may use AI tools, including what is off-limits and who approves new uses.
Augment vs. Automate	Augmenting helps a person do a task better or faster; automating hands the task to the system entirely. Most parliamentary uses should augment.
Personal and sensitive information	Details that identify individuals or that parliament must protect, such as constituent correspondence or confidential committee material. Never enter this into an AI tool your parliament has not approved.
Custom AI assistant	Converting paper or analog records into digital files, such as scanning a volume of Hansard.
Digitization	Converting paper or analog records into digital files, such as scanning a volume of Hansard.
Hallucination	When an AI confidently produces information that is factually incorrect, fabricated, or entirely made up but delivered in a tone that sounds completely authoritative. Hallucinations are a known and consistent limitation of all current large language models.
Human in the loop	The principle that a human remains responsible for reviewing, approving, or acting on AI-generated outputs before they are used, shared, or acted upon. The AI assists as a tool, but the human decides what to use and is responsible for the final product.
Pilot	A small, time-limited test of a tool or approach with a defined group and clear success benchmarks, decided before it starts.
Prompt	The instruction, question, or task you type into an AI tool. The quality of your prompt directly shapes the quality of the response you get.
Retrieval-augmented generation (RAG)	Instead of relying purely on what an AI learned during training, RAG allows the AI to first retrieve relevant information from a defined set of sources (such as documents you have provided, a database, or the web) and then generate a response grounded in that material.
Shadow AI	Use of AI tools at work without institutional guidance or approval. Common everywhere, and a sign that people need clear rules and approved options.
Structured data	Information organized in a consistent, machine-readable format, so computers and AI tools can read it reliably rather than guessing.

Part Two: The Working Session

The two sessions follow one possible roadmap: establish a mandate, understand where you are, protect what must not change, choose a starting group, train them, and learn by piloting. The best way to know whether a technology fits your parliament’s work is to use it, in a small, clearly defined, carefully controlled way. Move on when a section has produced a usable answer. Rough answers are fine, and the group can improve them later. If a topic needs more time than the session allows, write it down as a follow-up.

Session 1: Mandate and Starting Point (about 90 minutes)

Section	Time	What This Is	The Group Leaves With
1Establish your mandate	15 min	Agree what the group is for, who it answers to, and who belongs in it.	A shared statement of purpose and reporting lines
2Where are we now	30 min	Describe how the parliament handles records, digital tools, and AI today, including informal use.	An honest picture of current tools, use, and concerns
3What doesn't change	30 min	Name the steps, decisions, and rules that stay in human hands, then sort one real piece of work.	The steps and rules any tool must respect

Session 2: Starting Group and Pilot (about 90 minutes)

Section	Time	What This Is	The Group Leaves With
4Choose a target group	20 min	Pick who you're starting with.	One group to start with, and a champion within it
5Sketch the training	20 min	Work out what that group needs to learn and how to deliver it.	A training outline and a route to deliver it
6Plan the pilot and set benchmarks	35 min	Design a small, time-limited test and decide in advance how to judge it.	A defined pilot and a way to know if it worked
7Commit to next steps	15 min	Write the decisions on one page and agree who does what next. The group fills it in at the end of each session.	A one-page summary to share with leadership

1. Establish your mandate

15 minutes

A working group moves faster when everyone agrees on what it exists to do and who it answers to. The mandate, meaning what the group is asked to do, can be modest to start and grow as the work does. A few sentences per answer is enough to move forward.

1.1 What is this working group charged to do?

For example, the Parliament of Sint Maarten’s Digital Transformation and AI Workgroup was charged with assessing Parliament’s current digital environment, supporting a digital transformation programme, recommending frameworks for AI governance and ethics, identifying priorities for digital innovation, and promoting digital literacy.

1.2 Who does the group report to, and how often?

In Mauritius, the Clerk chairs the National Assembly’s AI Board. In the US Congress, legislative branch offices report on their AI pilot projects every quarter to the Committee on House Administration. A smaller group might agree to brief the Clerk for ten minutes once a month.

1.3 Who sits in the group today, and who is missing?

Consider whether the staff whose work may change are at the table. People closest to the work are best placed to judge where tools fit. The Parliament of Austria built its digital skills team from every part of the administration, not only IT. That surfaced skills nobody had counted on, including a stenographer who could write code and helped build the speech-to-text system.

1.4 Who else needs to stay informed as the work moves forward?

The House of Commons of Canada gave employees regular updates from the start of its AI work. Think about staff outside the group as well, Members from both government and opposition, party offices, and the IT team. The group’s work is more likely to last if people see it as serving the whole institution, not only the majority or the administration. That effort supports institutional continuity for AI integration.

2. Where are we now

30 minutes

An honest picture of where you are today makes every later decision easier. This section describes how things work now. It is not an audit, and nobody is in trouble for what they share.

2.1 How does your parliament currently manage legislative documents and records?

Are records mostly paper-based, scanned, or held in a centralized digital archive? What is searchable today? As part of the Digital Parliaments Project, Caribbean Parliaments assessed their current status and went from paper to digital workflow.

2.2 What digital tools do people rely on day to day?

List everything people use, including the informal tools: email, shared drives, video streaming, the website, and the messaging groups where much of the daily coordination happens.

Consider each group that applies to your parliament, since their tools and habits often differ:

• Members

• Staff who support Members or political parties

• Parliamentary administration

2.3 What AI tools are people already using, formally or informally, and for which tasks?

Informal use is common in every institution. Knowing where it happens shows what people need and where they need guidance first. At a 2026 IPU meeting of 23 parliaments, participants reported that much AI use in their institutions was informal and undeclared, and they treated it as a sign of what staff need. If the room goes quiet, ask it another way: if someone here were using AI, what would they most likely use it for?

Consider:

• Which tasks: drafting, summarizing, translating, research, or something else?

• Free versions, paid accounts, or tools built into software the parliament already has?

• Has anyone been told they may, or may not, use these tools?

2.4 What are the biggest challenges you face with digitization and technology adoption?

For example, several Caribbean parliaments name Hansard transcription, because standard voice-to-text tools miss local accents and dialects.

Consider:

• Equipment and internet connection

• Budget and software procurement

• Staff time and digital skills

2.5 What concerns do people in your parliament have about AI?

Common concerns include wrong answers, privacy, what AI means for people’s jobs, and how new staff will learn the institution if AI takes over the tasks they used to learn on.

2.6 What kind of adopter is your parliament?

POPVOX Foundation uses four archetypes, or types, to describe how parliaments approach AI. Read the four descriptions, then mark where your parliament is today and where the group would like it to be.

Builder: the parliament builds its own capacity.

You might be here if the parliament treats AI as a skill to build in-house, not a product to buy. It invests in its own experts and partnerships and plans for the long term, not only for quick wins.

Picture it: The Hansard team finds that standard transcription tools miss local accents. The IT unit works with a university to build a tool trained on recordings from the chamber, and the parliament commits staff to maintain it.

Configurer: the parliament adapts tools that already exist.

You might be here if the parliament has signed up to a widely available AI tool for the whole institution and adjusts it to fit legislative work. Staff get tools quickly, and the cost at the start is lower.

Picture it: The parliament turns on the AI assistant in the office software everyone already has. The IT team sets up a version that answers from the Standing Orders and writes instructions for common tasks such as drafting minutes.

Experimenter: the parliament learns by trying.

You might be here if some teams run small official tests while individual staff and Members try tools on their own. People build familiarity at low cost, but the rules are loose and the strategy is still taking shape.

Picture it: The library runs a three-month test of a research assistant. Down the hall, a committee clerk drafts minutes with a tool on her own account. Nobody has written down which approach the parliament will back.

Constrained: something in the institution holds adoption back.

You might be here if existing guidance blocks most uses, older systems are hard to connect to new tools, security rules leave little room to try anything, or budget and staff time are too tight to start.

Picture it: A researcher wants to use an AI tool to summarize public reports. The IT policy blocks every AI website, the document system cannot connect to anything new, and nobody has the authority to grant an exception.

Many parliaments fit more than one. An institution can be Constrained by its security rules while individual staff experiment on their own, or a Builder in the Hansard office and a Configurer everywhere else.

Archetype	Today	In the Future
Experimenter
Configurer
Builder
Constrained

What would need to be true to move from where you are today to where you want to be?

The Senate of Spain assessed itself and found that strong technical progress had hidden gaps in governance. Moving forward can mean catching up on rules as much as adding tools. Groups that want a deeper look can use the IPU’s Maturity Framework for AI in Parliaments, also listed in Appendix B.

3. What doesn’t change

30 minutes

Before discussing where AI could help, name what must stay fixed. Parliamentary procedure and tradition protect the institution’s integrity and the public’s trust. Naming these first makes every later idea safer, and it means the people who do the work set the boundaries.

3.1 Which steps must always rest with a person?

For example, certifying the official record, checking a transcript before publication, or approving a letter sent in a Member’s name. The Westminster Foundation for Democracy’s guidelines hold that AI should support human decisions, not replace them.

Consider steps that:

• create or change the official record

• speak in a Member’s name

• involve personal or confidential information

• involve a judgment about a person, such as a constituent case or a hiring decision

3.2 Which decisions belong only to Members, the Speaker, or the Clerk?

For example, a ruling from the Chair, how a Member votes, whether a bill or amendment is in order, and what enters the official record. AI can prepare material for these decisions. It cannot make them.

3.3 What existing rules should any new AI rules respect?

Think about the rules you already have for buying software, using personal devices and accounts, who can see which documents, protecting confidential and personal information, and required staff training.

3.4 Sort the tasks

The questions above name what must stay fixed. This exercise applies those answers to real work. It will help you think through, one step at a time, where AI should stay out, where it could help a person, and where it could do the work with a person checking the result.

Tasks are the right level for this conversation. “Should the committee office use AI?” is hard to answer. “Should AI write the first draft of the minutes?” is a question everyone in the room can picture and judge. The German Bundestag works in a similar fashion: its AI Cloverleaf Model maps where AI could help, department by department, with checks built in.

Pick one piece of work your team does from start to finish, and list the steps it takes. Then place each step in one column. Most steps will land in the first two columns. That is expected, and it is a useful message for anyone worried about their role: the aim is to take routine work off people, not to replace their judgment.

You will use this list again in Section 4 to choose where to start and in Section 6 to choose what to test.

Example: an oversight committee holds a hearing

AI Should Not Do This	AI Could Help a Person Do This (Augment)	AI Could Do This With Review (Automate)
Decide what the committee looks into	Research the background on the issue	Transcribe the hearing
Make the final choice of witnesses	Suggest a long list of possible witnesses	Gather the published evidence into one searchable file
Ask the questions and judge the answers	Draft a first set of questions	Format the final report to the house template
Agree the findings and recommendations	Draft a summary of the hearing for the report

The pattern holds across most parliamentary work: AI can prepare, draft, and record, and people decide. Now sort your own:

AI Should Not Do This	AI Could Help a Person Do This (Augment)	AI Could Do This With Review (Automate)

If you have time, repeat the exercise with a second piece of work.

Ending Session 1

If you are running two sessions, stop here. Turn to Section 7 and fill in the first three rows of the summary together. Before Session 2, share the draft mandate with whoever the group reports to and get their agreement.

4. Choose a target group

20 minutes

A good group to start with has four things: a task that causes trouble again and again, someone willing to lead, a team small enough to train together, and work where AI would help people (the middle column of your list in Section 3). If you are using this guide for a single office or committee, you are the starting group. Use this section to choose which people and tasks go first.

4.1 Which group will you start with?

For example, a single team such as Hansard, the library, or one committee. The Shura Council of Bahrain uses AI to turn spoken sessions into written records for Hansard.

4.2 Why this group?

Groups often reach for the most visible idea first. Choose the place where a win is realistic and the lessons are easy to capture. A Hansard team is a common first choice: the task repeats after every sitting, the time it takes is easy to measure, and the material is already public.

‍ 4.3 Who within that group can serve as a champion?

Champions are trusted colleagues who try the tools first, help others, and report honestly on what works. The National Assembly of Zambia named keen Members and staff as digital champions. They tried new tools first, and colleagues found them more convincing than messages from the IT team.

4.4 How will you learn what this group needs?

Appendix A offers an individual check-in the group can send after Session 2.

5. Sketch the training

20 minutes

Training works best when it is built around the target group’s own work and follows a simple sequence. Include Members where you can, not only staff. As Annelie Lotriet, Deputy Speaker of the National Assembly of South Africa, told ModParl, AI is a challenge for the whole institution, not just a project for the ICT department.

Stage	What It Covers
1Why it matters	Why these tools matter for this group's own work, shown through a real task from their day.
2Tools	Which approved tools the group will use, and what each is for.
3Good habits	Safety and skills together: keeping information private, checking answers, and writing good prompts.
4Hands-on practice	Working through real tasks from the group's own workload, with support.

Choose a route to deliver it:

Host it internally.

Best when someone in-house already uses these tools well. The group can build a session from the resources above and adapt it to its own hardware and software.

Learn from a peer parliament.

Best when a parliament nearby has already been through this. Regional networks, like the IPU’s innovation community and the Digital Parliaments Project’s quarterly calls are good places to find one.

Bring in an external organization.

Best when the group wants a facilitated session or does not yet have someone in-house to lead it. POPVOX Foundation can deliver training directly, help the group plan its own, or talk through the group’s answers from this guide.

5.1 Where are the gaps between what this group knows now and what it needs to know to use AI well?

5.2 Which delivery route fits?

5.3 What is the first step to set it up?

5.4 When could the training realistically happen?

6. Plan the pilot and set benchmarks

35 minutes

A pilot, meaning a small test, is the fastest way to learn whether a tool fits. Keep it internal, small, and time-limited, and decide how you will judge it before it starts. A pilot that shows a tool does not fit is still a success if the group records why.

The Pilot

6.1 What task or tasks will the pilot address?

Choose a narrow task from your list in Section 3. Brazil’s Chamber of Deputies, for example, uses AI to group the comments citizens submit on bills, so Members can see which issues people raise.

6.2 What tool or approach will you test?

Some parliaments start with a ready-made assistant. Others build on their own records: Italy’s Senate has described a tool that turns everyday questions into searches of its open data.

6.3 How long will it run?

Long enough to cover the full cycle of the task, such as several sittings or one committee inquiry. Three to six months is common.

6.4 What rules will keep the pilot safe?

The National Assembly of France wrote a short charter with clear rules: cloud-based AI tools may handle only public information, a person must check all AI-assisted work before anyone uses it, and staff must tell Members when they have used AI.

Benchmarks

Benchmarks are the measures you will use to judge the pilot.

What to Measure	What It Tells You	How to Measure It
Use	Whether people use the tool, and for what	Count regular users, how often they use it, and for which tasks. Note whether people move from personal accounts to the approved tool.
Time saved	Whether routine tasks take less or more time	Pick two or three recurring tasks, such as drafting committee minutes or answering a research request, and time them before the pilot and at the end.
Quality	Whether the work is as good as before, or better	Before the pilot, a supervisor reviews a small sample of the work, such as five sets of minutes, against a short checklist: accurate, complete, and in house style. At the end, the same person reviews a new sample against the same checklist and notes how many corrections each piece needed.
Confidence	Whether people feel more capable and clearer on what the tools can and cannot do	Run the same check-in (Appendix A) before and after the pilot.
Safe practice	Whether people follow the rules for safe use	Check that an approved tool and written rules exist and that people know them. Log errors caught in review as the process working.
Institutional change	Whether the pilot changed how the parliament works beyond the pilot team	Note any rule written or updated, any step in a process changed for good, and any other team that asks to join.

6.5 Which benchmarks will you use, and what does success look like for each?

6.6 How will you measure where things stand before the pilot starts?

6.7 When the pilot ends, how will you decide whether to scale, adjust, or stop, and who makes that decision?

7. Commit to next steps

15 minutes per session

Fill in this summary together. At the end of Session 1, complete the first three rows. At the end of Session 2, finish the rest. The finished summary captures the group’s decisions in one place and the group can share it with leadership as a record of its plan.

Our mandateSession 1
Where we are todaySession 1
What must not changeSession 1
Our target group and championsSession 2
Training route and timingSession 2
Our pilotSession 2
Our benchmarksSession 2
Next three actions (who and by when)Session 2
Next meetingSession 2

After Session 2

• Send the individual check-in (Appendix A) to your target group to set a baseline.

• Set up the training using the route the group chose.

• Launch the pilot, track the benchmarks, and meet again at the end to decide whether to scale, adjust, or stop.

• Share what you learn: other parliaments are asking the same questions, and your experience can help them.

Appendix A: Individual AI Check-In

The working group should send this to their target group before training begins and again when the pilot ends. We recommend keeping answers anonymous. The group can share the check-in on paper or online, whichever reaches people more easily.

Part 1: Your work

1. Which team or role best describes your work? (Keep the options broad to protect anonymity.)

2. Think about a normal week. Which three tasks take most of your time?

3. Which one task would you most like help with, and why?

4. Where would help be most useful to you? Rank your top three: finding past records, summarizing long documents, writing first drafts, transcribing audio, translating or rewriting in plain language, researching an unfamiliar topic, scheduling and admin.

Part 2: Your use today

1. How often do you use AI tools for work? (Never / Tried once or twice / Monthly / Weekly / Daily)

2. Which tools do you use, and do you use a work account or a personal one?

3. Describe one time an AI tool helped you, and one time it let you down.

4. If you do not use AI tools, what holds you back? (I don’t know how / I’m not sure it’s allowed / I don’t trust the results / I don’t have access / I don’t see the need / I worry about what it means for my role / Other)

5. Does your parliament have rules on using AI? (Yes, and I know them / Yes, but I have not read them / No / Not sure)

Part 3: Your judgment

How much do you agree? (1 = strongly disagree, 5 = strongly agree)

1. I can explain to a colleague what AI tools do well and what they do badly.

2. I know what information I must never put into an AI tool my parliament has not approved.

3. I know how to check an AI answer before I use it.

4. I can write an instruction that gets a useful result from an AI tool.

5. I know who to ask when I am unsure.

Would you use an AI tool for each of these? (Yes / Yes, with changes / No / Not sure)

1. Summarizing a bill that is already published

2. Drafting a reply to a constituent’s letter, including their name and case details

3. Summarizing papers from a closed committee session

4. Improving the wording of a press release before it goes out

5. Finding a statistic and its source for a Member’s speech

6. Tidying the wording of a colleague’s performance review

Part 4: What you need

1. What concerns you about AI in your work? (Wrong answers / Privacy / My job / Losing skills / Unfair or biased results / Nothing / Other)

2. What would help you most? Rank your top three: hands-on training, written rules, an approved tool, examples from colleagues, time to practice, a person to ask.

3. Would you take part in a small test of an AI tool? (Yes / Maybe / No)

4. What is one thing you want the working group to know?

For the Working Group: How to Apply The Results

Look At	It Tells You	Use It In
Questions 2 to 4	Where the time goes and where people want help	Choosing pilot tasks (6.1)
Questions 5 to 9	How much informal use exists and whether people know the rules	The Use and Safe practice baselines
Questions 10 to 14	Where the gaps are	Planning the training (5.1) and the Confidence baseline
Questions 15 to 20	Whether people can tell safe use from unsafe use	Training content and the Safe practice baseline
Questions 21 to 23	What worries people and who is willing	Choosing champions (4.3)

At the end of the pilot, repeat questions 5 and 10 to 20. The change between the two rounds is your comparative result.

Appendix B: Working References

IPU, Guidelines for AI in Parliaments (2024)

Comprehensive guidance on AI governance, risk, and implementation in parliaments. It recommends assembling a multidisciplinary team from across parliament to lead AI governance and establishing a working group to develop AI policy.

IPU, Maturity Framework for AI in Parliaments (2025)

A self-assessment tool that helps parliaments understand where they stand across six levels of AI maturity and plan next steps. A natural next step for groups that want a deeper look than Section 2 provides.

IPU, Use Cases for AI in Parliaments

A collection of AI use cases developed by parliaments in the IPU’s Centre for Innovation in Parliament community. Useful when a group needs ideas for Sections 3 and 6.

IPU Innovation Tracker, The AI Journey of the House of Commons of Canada

How the House of Commons of Canada has approached AI with a structured, human-centred framework that balances innovation with data protection, fiscal responsibility, and continuous communication with employees.

IPU, Examples of AI Governance in Parliaments (2025)

Takeaways from parliaments building AI governance, including the National Assembly of Mauritius, whose framework includes an AI Board chaired by the Clerk and work on Mauritian Creole transcription.

IPU Innovation Tracker, Prioritizing AI Use Cases at the German Bundestag

The Bundestag’s approach to mapping AI use cases across legislative work, supportive services, and internal administration, with a focus on the task level where AI becomes concrete. The full paper, The AI Cloverleaf Model, is by Dr. Martin Kamprath.

Westminster Foundation for Democracy, Guidelines for AI in Parliaments (2024)

Guidelines developed by 22 experts from 16 countries, covering ethical principles, privacy and security, governance and oversight, system design, and capacity building.

Parliament of Sint Maarten, Digital Transformation and AI Workgroup (2026)

The Presidium’s announcement establishing the workgroup, including its full mandate and membership. A practical model for Section 1.

Appendix C: POPVOX Foundation Resources

AI Terms for Parliamentary Work

A plain-language AI dictionary with parliamentary examples, from the basics through privacy and responsible use. Available upon request.

ORDER prompting framework

A simple structure for writing effective prompts.

Generative AI Use Policy Template

A starting point for written AI guidance.

Creating and Using Custom GPTs

A step-by-step guide to building a custom AI assistant around a role, set of documents, or recurring task.

What Is RAG and Why It Matters for Legislative AI Use Cases

AI 101 and prompt-writing recordings

Virtual training sessions a working group can use to build its own internal training.

Digital Parliaments Project community

Quarterly calls and regular peer exchanges where parliaments share lessons from their own digital transformation and AI work (For questions about this, reach out to chloe@popvox.org.)

ModParl newsletter

News and interviews on how legislatures around the world are modernizing, including their work on AI.

Continue the conversation

Once your working group has completed both sessions, POPVOX Foundation would welcome the chance to hear how it went. A member of our team may be able to talk through your answers with you and help you plan next steps.

We work only with legislatures and we are glad to hear from any working group that has used this guide.

Please contact Chloe Ladd at chloe@popvox.org with any questions.