Policy Atlas searches two trusted databases:

1. OpenAlex, with more than 400 million academic and scientific works; and
2. Overton, with more than 20 million policy documents from governments and institutions worldwide.

By searching academic and policy databases rather than the open web, Policy Atlas grounds its AI-generated claims in more reliable evidence than general-purpose large language models (LLMs).

The tool then applies systematic review methodology to identify well-evidenced and impactful interventions. Documents are screened for relevance, findings are extracted in a structured form, and the strength of the underlying evidence is scored against established hierarchies.

Policy Atlas then produces outputs that are tailored to support policymakers with real-world policymaking tasks, specifically at high-impact decision moments - such as scoping policy options, scrutinising proposals, and making spending decisions.

• Options scoping: identifying and comparing policy intervention choices on impact, evidence strength, transferability and comparable case studies, against a do-nothing baseline.
• Theory of change modelling: mapping how an intervention is expected to lead to an outcome, as well as potential failure pathways, and surfacing the key assumptions underpinning the theory.
• Stakeholder mapping: Identifying key experts and stakeholders who could be consulted on any policy topic

• Value-for-money analysis: Helping policymakers weigh the costs and benefits of different policy intervention options.

The project began as exploratory work in Nesta's Mission Discovery team on whether AI could make evidence synthesis faster. An early prototype, LitSynth, tested whether large language models could screen and extract findings from research literature at scale.

We built a first working version and began structured user research with policymakers, followed by pilots inside government departments. Demand was immediate and clear.

More than 85 policymakers across five central government departments were onboarded to the alpha version of Policy Atlas, running over 1,000 evidence searches. We interviewed users, watched how they worked and collected feedback on every stage.

Evidence search is live and in use via our beta tool. Options appraisal is next. We are continuing to evaluate the tool through departmental pilots.

You can request access on the Policy Atlas website and we will be in touch within a couple of days to get you set up.

We are particularly keen to hear from teams working on a live decision - a spending proposal, a strategy, a green or white paper - who can support us with evaluating the tool in practice.

If your fund or foundation is interested in co-developing Policy Atlas or related approaches, we are always open to partnerships.

If you work on evidence standards, appraisal frameworks or synthesis methods we would love to explore opportunities to collaborate on improving methods and evaluations.

Several teams in the UK and internationally are working on AI tools for evidence and policy. We are eager to work alongside each other, rather than duplicate efforts. If you are building in this space, we are interested in comparing notes on architecture, evaluation and what has not worked.

We think the lack of shared standards for evaluating AI tools in policymaking is a challenge for the whole field. If you work on AI evaluation, assurance or safety in a public sector context, we would like to talk - both about Policy Atlas, and about what common standards for these tools could look like.