Foundation models are already answering the public’s legal questions. This is the work of getting them to do that well across the tasks people actually need, and the shared standards, evidence, and tools that let any state build on it instead of starting alone.
Explore these reports to learn about AI strategy, development, and evaluation for legal help. Start with the first section or jump to the topic you need.
Justice professionals say where they are stuck on AI, these reports lay out the options for building and buying, and they make the case for investing in the shared layer.
The published article on how legal aid organizations are deciding to use AI, built on a survey of justice professionals. It reports what those teams weigh when they buy from a vendor, what they weigh when they build in-house, and how a commons model gives them shared infrastructure they still customize locally. Read the summary, the full text, or the PDF as it appeared in print.
What you’ll take awayOwning or fine-tuning your own model does the least to change how it behaves. Grounding it in your content, writing the prompt, and what the model makers put inside all do more.
What you'll take awaySome of this work outlasts any single state and does not belong to one. This report makes the case for national leadership on that shared layer. It says what the shared layer would own, why it has to stay neutral, and why funding it once costs less than fifty states each paying for it badly.
What you'll take awayThese reports show how to measure quality, what our measurements found, and what it would take to cover the whole task taxonomy.
Set up promptfoo and a reusable protocol like the field has done, measure any tool against the tasks you care about, and gather your own evidence before you deploy.
What you'll take awayAcross five rounds of testing, a verified list of your own resources lifted every model, while loading full articles helped only when they fit.
What you'll take awayThe full record of the testing behind the grounding finding, across five rounds. It includes the round that came back a tie because the rubric could not see the difference, the round where the earlier result reversed on high-risk questions, and an account of how much weight each finding can carry.
What you’ll take awayWe ran 60 cases through Claude, Gemini, and GPT-4o, asking each to tag the legal issue, the audience, the jurisdiction, and the urgency from a person's own words. What made it work was putting our own category lists in the prompt.
What you'll take awayTeams are measuring AI on legal help tasks and the results cannot be compared with each other. This sets out the choices, the partners, and the work to build a read the field can pool, and how the issue-area agendas and the task taxonomy fit together.
What you'll take awayStart with the inventory of what already exists, then get your own content ready for AI, then pick up the specific tools or rebuild them yourself.
You can pick up the shared vocabularies, the datasets to build and test with, and the working tools, each with an honest note on how ready it is.
What you'll take awayThis playbook turns your website, guides, and trainings into a knowledge base an AI can use safely. It covers gathering content and tiering it by authority, chunking it, building safety into the content itself, and growing from search to a knowledge graph.
What you'll take awayThis is the step-by-step companion to the content playbook. It works through every stage in detail, giving the goal, the steps to take, the tools to use, a picture of what the output looks like, and the signs you have done it well.
What you'll take awayMany teams want to build one. This short note says what decides whether a Q&A bot answers safely, and it points to the reports that go deep on each part.
What you'll take awayWe surveyed experts and built a labeled legal document set. This report says what counts as PII in legal records, how to judge whether a masking tool actually works, and how to test one without touching a real client's file.
What you'll take awayA short deterministic check catches the hallucinated links and misattributed numbers an LLM judge waves through, and it costs almost nothing to run. This report shows how to run ours and how to build your own.
What you'll take awayThe tools rest on shared standards and on a way of running cohorts, and this is where a team decides what to do first.
Shared field names and vocabularies let one team's knowledge base and directory be trusted and reused by another, and adopting them is a mapping exercise rather than a rebuild.
What you'll take awayA cohort either generates strategic clarity or carries working tools between organizations, and one that tries to do both at once tends to do neither well.
What you'll take awayFor the team that runs a public legal help website, whether it sits in a legal aid organization, a court self-help center, or a statewide website project. You locate your team by aim, content stage, and size, and that tells you what to do first, whether you are one person or five.
What you'll take awayThe projects, standards, and tools the Commons has published, gathered here.
The field's projects, datasets, benchmarks, and the legal help task taxonomy.
The 52 tasks across the justice journey that frame what AI should help with, and the rubric and evaluation work behind them.
The shared taxonomy of legal issues that ties content together across states.
The standards a knowledge base follows so its content can be trusted and reused.
The open agenda of tools the field is committing to build once and share.
The end-to-end help people need, built and shared by vertical cohorts.
The cohorts doing the building, and how to join one.