New guidelines launched to prepare government datasets for AI

The UK government has published new guidance aimed at helping public sector organisations prepare their data for effective and responsible use with artificial intelligence, as departments accelerate experimentation with AI-enabled tools and services.

The guidance, Guidelines and best practices for making government datasets ready for AI, sets out a practical framework for improving the quality, governance and usability of public sector data, reflecting growing recognition that data readiness - rather than algorithms - is now the main constraint on AI adoption in government.

Published by the Government Digital Service, the document is intended to support data leaders, digital teams and policy officials across central government, local authorities and arm’s-length bodies. It builds on lessons from recent AI pilots, many of which have struggled because underlying datasets were incomplete, poorly documented or not designed for reuse.

At the heart of the guidance is a four-pillar model for “AI-ready” data. The first pillar focuses on technical optimisation, encouraging organisations to ensure datasets are stored in modern, interoperable formats and can be accessed reliably through APIs or secure platforms. This reflects the growing need to integrate AI tools with live operational systems rather than static data extracts.

The second pillar addresses data and metadata quality. Departments are urged to treat datasets as products rather than by-products of services, with clear ownership, consistent definitions and rich metadata that explains how and why the data was collected. Without this context, the guidance warns, AI systems risk producing misleading or biased outputs.

A third pillar looks at organisational and infrastructure readiness, emphasising the importance of governance, stewardship roles and sustainable funding. The guidance highlights the need for cross-organisational collaboration, particularly where AI use cases depend on linking data across departmental or sector boundaries.

The final pillar covers legal, security and ethical considerations, reinforcing existing obligations around data protection, transparency and public trust. It encourages teams to assess privacy risks early, consider bias and representativeness, and ensure datasets are used in line with clear legal powers and ethical frameworks.

Alongside the principles, the document includes an action plan and self-assessment checklist to help organisations evaluate whether a dataset is suitable for AI use and what remediation may be required.

The publication aligns with wider government work on responsible AI, including the AI Playbook and data standards promoted by the Government Digital Service and the Open Data Institute.