AI Policy Toolkit – Frequently asked questions

The AI Policy Toolkit is a practical resource for governments and stakeholders to map their AI policy landscapes, identify priorities, and explore concrete international examples to inform their own policy design and implementation.

The Toolkit is tailored for staff in ministries working on digital and AI policies, dedicated AI bodies, and line ministries engaged in AI-related work. Additionally, development agencies and international partners supporting government capacity in AI governance will find it highly useful for scoping and shaping their support. Researchers and AI policy experts might also find it useful for their work.

The Toolkit does not evaluate, score, or rank countries on their AI policies or readiness.

Developed in close consultation with government officials, the Toolkit combines features not typically found in a single platform and is distinctive in three key ways:

1. The Toolkit is modular and lightweight by design. The questionnaire is divided into pillars rooted in the OECD AI Principles, allowing officials from a specific ministry or team to complete sections independently without navigating the entire tool from start to finish. Questions under a single pillar can be completed (and pillar-specific policy priorities identified) in approximately 15-30 minutes, depending on the user’s familiarity with the national policy landscape.
2. The Toolkit adapts to a country’s specific stage of AI governance. Its structured questionnaire has been validated across multiple regional workshops, making it uniquely tailored for government use. As users answer questions within each pillar to identify priorities, certain advanced questions remain  conditional so users can focus on foundational policy options first. For example, the Toolkit will not surface a question on a high-performance computing access policy if a country does not yet have access to GPU resources suitable for AI training and inference workloads.

The Toolkit is designed for all economies, the Toolkit focuses on practical implementation rather than just diagnosis. It goes beyond merely identifying gaps to surfacing policy options and examples tailored to a country’s specific capacity and priorities. To serve the interests of more developed economies, the policy examples retrieved by the Toolkit describe what countries have done to implement specific policies ; the tool does not grade policies, recommend mandatory approaches, or issue official OECD endorsements. As such, the Toolkit is useful for countries at any stage of AI maturity-from those early in their journey looking to define policy priorities, to advanced economies focusing on concrete implementation of specific policies.

All AI outputs from the Toolkit are strictly grounded in OECD publications and the OECD.AI Policy Navigator (which catalogues over 2,000 initiatives across more than 80 jurisdictions). Unlike general-purpose chatbots, the Toolkit does not generate information from unverified training data.

All AI outputs from the Toolkit are strictly grounded in a specific, user-selected source document, not an aggregated summary across multiple documents.
This preserves source context, reduces the risk of hallucination, and makes human verification straightforward. Users can select which document the AI response draws on and verify the response against the original document page-by-page, side by side.

Unlike deep research tools, the Toolkit is not a report generation tool. It does not independently draft policies, strategies, or implementation plans.

The AI component of the Toolkit leverages OpenAI’s existing model APIs; it does not train or fine-tune a separate AI model.

No. The AI Policy Toolkit does not rank countries and does not produce a composite score. It generates a structured policy profile that tells a government what it has in place, what is missing, and what to do next.

How does the AI Policy Toolkit relate to the OECD.AI Index?

The Toolkit and the OECD.AI Index are complementary, not duplicative. The Toolkit focuses on the policy actions a government has put in place (e.g., has an AI strategy been adopted? Is there a dedicated AI body?). The OECD.AI Index measures what those actions produce (e.g., AI investment levels, R&D outputs). The two instruments are intended to be used together.

No. The Toolkit is a decision-support tool, not a tool to draft policy documents. It can help you:

• Identify which policy areas your country has and has not addressed.
• Find relevant examples of what other countries have done.

It does not:

• Generate policy text, draft strategies, or produce implementation plans.
• Endorse, rank, or recommend specific policy approaches.
• Substitute for expert analysis, stakeholder consultation, or human judgment in policy design.
• Provide legal or regulatory advice.

All outputs from the Toolkit are informational and non-prescriptive, and do not constitute official OECD recommendations.

There are two starting paths:

If you already know which policy priority you want to explore: Select a pillar and a specific policy priority of interest to retrieve relevant examples immediately. This path is well-suited for economies looking for concrete examples to guide their next steps in policy implementation.

If you are defining or updating your policy priorities: Start by mapping your country’s current AI policy landscape to identify gaps and generate a structured priority profile. The Toolkit then surfaces relevant policy examples for your identified priorities. This path is especially useful for countries early in their AI policy journey, or for those undertaking a comprehensive review.

Yes. The Toolkit is intentionally designed to support collaborative, cross-ministry participation. Each of the six pillars maps broadly to a distinct policy domain (e.g., data governance, human capital, and research investment), making it practical for different ministries/teams to complete their respective pillars independently and combine results. Coordination can be managed by sharing the full questionnaire prior to response completion and consolidating responses manually.

No. You do not need to complete all pillars in a single session. The Toolkit is modular and supports partial completion. You can complete one or more pillars, save your responses, and return to complete additional pillars at any time.

Yes. Responses are fully editable at any time. If a new law is passed, a new body is established, implementation progress is made on a policy option, or your team identifies more accurate information about an existing measure, you can update the relevant responses. The Toolkit is designed for ongoing use as a living assessment, not a one-time snapshot.

Yes. The Toolkit aims to provide a non-prescriptive guide, not a score, to indicate where the opportunity for future policy action lies. It is not meant to evaluate or penalise. An unimplemented policy option does not mean your country is behind; it may simply reflect a different context, priority order, or governance approach.

The response summary is a structured, non-scored overview of your country’s AI policy landscape across the pillars you have completed. It shows:

• What is in place: policy options you have confirmed are currently operational.
• Gaps identified: policy options marked as not in place or partially in place.
• Priority areas for further action: the gaps that the Toolkit highlights as most significant given your pillar-level response patterns.

The profile is a starting point for deliberation, not a final verdict. All responses remain editable, and the priority ordering reflects the structure of the questionnaire — it is not a normative ranking. From the profile, you can navigate directly to relevant policy examples  for guidance on that pillar.

Your responses are not made public and are not shared with other countries. They are accessible only to response providers and their colleagues, should the response provider choose to export a PDF or share a link to the results at the end of the module.

The platform collects standard analytics on platform usage (pages visited, features used, session duration, and user feedback on response quality) to support quality improvement at aggregate and anonymised level. No personally identifiable information about policy responses is linked to usage analytics.

AIEDT unit), in collaboration with member countries of GPAI, national contact points for the OECD.AI policy navigator, and regional partners.

 The Toolkit content and features were developed through an iterative, user-centred design process combining:

1. Formative user research: 14 user interviews with government officials and developmet agency staff from 14 countries/organisations across Asia-Pacific, Latin America, Europe, and the Middle East. These interviews were structured to understand their AI policy workflows, pain points, and information needs.
2. Regional co-creation workshops: 4 regional co-creation workshops in Southeast Asia (Thailand, August 2025 and March 2026), Latin America and the Caribbean (Costa Rica, December 2025), and Africa (Kenya, March 2026). These workshops included content validation and usability testing sessions with delegates from participating countries to identify feature requirements and finalise the design of the toolkit V1.0.
3. Technical development and evaluation: the AI component of the Toolkit to retrieve and generate summaries of policy documents is built on the RAG architecture, with a structured evaluation pipeline covering retrieval accuracy, answer quality, safety, and security.
4. Beta testing: a series of beta testing are planned in Q3 and Q4 2026.

Policy examples and documents provided at the end of each module are sourced from OECD publications and the OECD.AI Policy Navigator, our repository of over 2,000 initiatives across more than 80 jurisdictions, vetted by OECD analysts and our government contact points.

The quality of the AI-generated outputs used to retrieve and process these documents is vetted through an end-to-end automated evaluation testbed and partial human review. Source traceability and side-by-side verification are built into the interface to support transparency and user oversight.

For more information, please review the design and system documentation.

The Toolkit is a first release and is subject to continuous improvement. Known limitations include:

The Toolkit is not yet fully or automatically synchronised with the OECD.AI Policy Navigator. In the near future, policy examples and documents submitted to the Policy Navigator will automatically sync and surface within the Toolkit, continuously providing users with the most recent and relevant policy examples.

The coverage of policy examples is not exhaustive. To maintain a clean interface, the number of policy examples retrieved is currently limited; we plan to implement technical and design solutions to address this in future phases. For certain policy priorities, our repository may lack a sufficient number of relevant examples, particularly from lower-income and non-OECD countries. Manual curation and automated data ingestion are currently underway, and we expect these coverage gaps to narrow with each update cycle.

The AI component inherits LLM limitations. To help users retrieve and digest long policy documents, we use large language model (LLM) and retrieval-augmented generation (RAG). Like all such systems, it is vulnerable to hallucinations (inaccurate or unsupported text), particularly when retrieved content is sparse or vague. Our deliberate design choices to provide document-specific summary generation, side-by-side verification panel, and citations mitigate but do not eliminate this risk. Users should manually verify AI responses for high-stakes decisions.

The current version does not include dedicated pillars for specific sectors (e.g., AI in healthcare or agriculture). This coverage is planned for V2.0.

The platform currently operates in English. Multilingual expansion is a priority for future funding and development.

OECD.AI (YYYY), AI Policy Toolkit, accessed on DD/MM/YYYY, https://oecd.ai/ai-toolkit.

We welcome feedback from all users. You can:

• Submit feedback through the in-platform feedback form.
• Report missing policy examples or data inaccuracies via the OECD.AI Policy Navigator submission interface, which allows countries to submit new or updated policy initiatives directly.
• Suggest features or improvements by contacting the OECD AI team at aitoolkit@oecd.org.

The Toolkit follows a continuous iteration roadmap. Feedback from users will directly inform prioritisation of new features, expanded pillar coverage, and data improvements.