What problem does it solve?
Low income Americans who face a civil legal problem, an eviction, a custody dispute, a consumer debt claim, often go without a lawyer. The Legal Services Corporation's 2022 Justice Gap study found that low income Americans did not receive any or enough legal help for 92% of their civil legal problems. Cost is an important barrier: 46% of those who did not seek legal help for one or more problems cited cost as a reason, and separately, 53% of low income Americans in the same study doubted they could find an affordable lawyer. The result is a large population of self represented litigants who still have to find the right form, meet a real deadline and show up to a hearing they do not fully understand.
Court and legal aid websites answer some of this in writing, but people often do not know which page covers their situation or which legal term describes their problem. A generative assistant can meet the question in plain language, but the design has to be conservative: both organisations on this page state that their assistant does not give legal advice, so the assistant has to inform and refer, never advise.
- Low income Americans did not receive any or enough legal help for 92% of their civil legal problems.The Justice Gap (2022)
- Nearly one half (46%) of those who did not seek legal help for one or more problems cite concerns about cost as a reason why.The Justice Gap (2022)
How does it work?
- Answer in plain language. The person types their question in their own words; the assistant reads it without requiring legal terminology.
- Retrieve from vetted legal information only. It answers from a curated set of legal information content the organisation has written and reviewed rather than from the model's general knowledge, the way Beagle+ retrieves only from People's Law School's own site and Dial-A-Law content.
- Stay general, refuse to advise. The assistant explains the law and the process in general terms and explicitly declines to say what the person should do in their own case, a limit both organisations on this page state directly to users.
- Point to the next concrete step. It names the relevant form, self help centre or referral resource, and where a human is needed, such as a legal aid intake line or a court self help centre.
- Support the languages the population needs. Legal Aid of North Carolina describes LANC-LIA as giving multilingual answers to general civil legal questions, though its own page does not name which languages.
- Audience
- Customer facing
- Autonomy
- Assist
- Adoption
- Early adopters
- Channels
- Web chat
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
No public deployment has disclosed a measurable outcome yet.
Value drivers: Inclusion and access, Customer experience, Lower cost to serve.
Indicative value
A legal aid organisation or court self help centre answering 60,000 general legal questions a year
USD 41,667 to USD 1.1 million
Legal aid staff and hotline time released from routine questions per year
How this is calculated
Formula: questions * (minutesSaved / 60) * hourlyCost. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| General legal questions answered per year questions, questions per year | 20,000 | 100,000 | Editorial assumption. Replace with your own website and hotline question volume. |
| Staff or hotline minutes saved per question the assistant answers instead minutesSaved, minutes per question | 5 | 15 | Editorial assumption, replace with your own. Neither organisation on this page publishes a measured time saving. |
| Fully loaded cost of a legal aid staff or hotline hour hourlyCost, USD per hour | 25 | 45 | Editorial assumption. Replace with your own staff cost. |
What it leaves out: Time released, not cash saved, unless staffing changes. It leaves out the cost of writing and maintaining the underlying legal content, the value of access outside office hours, and the risk cost of a person relying on a wrong or incomplete answer instead of calling for help.
Who already uses it?
2 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
Legal Aid of North Carolina
United States · Government and public sector · 2024
LANC-LIA is a chatbot Legal Aid of North Carolina runs on its own website to give multilingual answers to general civil legal questions, focused on domestic violence, child custody, landlord tenant issues and consumer law. It answers questions, gives information on legal topics and refers people to further resources, and is built and described by the organisation to explicitly not give legal advice or answer case specific questions, only general information about the law in North Carolina. LANC launched it on July 10, 2024, developed in its Innovation Lab with LawDroid.
No outcome disclosed.
People's Law School
Canada · Government and public sector · 2024
Beagle+ is a chatbot from People's Law School, a nonprofit society in British Columbia, that guides residents to relevant, high quality legal information drawn from the organisation's own People's Law School and Dial-A-Law content. First launched in 2020 on the Rasa conversational AI framework, it was relaunched in February 2024 as Beagle+, running on OpenAI's ChatGPT with retrieval limited to the organisation's own curated content through a Pinecone vector database, so answers stay grounded in vetted legal information rather than the model's general knowledge. During development, the organisation created a 42 question test set to improve the chatbot's accuracy and helpfulness, and published the dataset to help others.
No outcome disclosed.
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- A curated, owned set of plain language legal information content, reviewed by a lawyer for accuracy
- A clear, written line between general legal information and legal advice that content owners and the assistant both follow
- A list of referral resources (self help centres, intake lines, court forms) to point to
Systems to integrate
- Content management system for the underlying legal information
- Referral or intake system for handover to a human
- Web and mobile channels, and translation for languages the served population needs
Complexity: Low
A grounded question answering assistant over an organisation's own legal information content is quick to build; both deployments on this page are described as answering general questions only, not questions about a person's case. The real work is writing and maintaining plain language legal content people can act on, and testing the refusal behaviour so the assistant never drifts into advice.
- 1
Write the content before the chatbot
Start from an existing, lawyer reviewed library of plain language legal information, the way Beagle+ draws only on People's Law School's own website and Dial-A-Law content, rather than letting the model answer from general knowledge.
- 2
Draw a hard line against legal advice
Decide explicitly what counts as advice for your jurisdiction and organisation, and test the assistant against it; both organisations on this page state directly to users that the tool answers general questions only and does not address their specific case.
- 3
Support the languages your population needs
Match the assistant's languages to who actually calls or visits; Legal Aid of North Carolina describes LANC-LIA as giving multilingual answers to general civil legal questions.
- 4
Build and publish a test set as you develop
People's Law School created a 42 question test set during development to improve Beagle+'s accuracy and helpfulness, and published the dataset to help others; do the same with real questions from your own intake data.
- 5
Give every conversation a way to reach a person
Make the route to a human, an intake line, a self help centre, a lawyer of the day, obvious and easy, especially when the assistant detects urgency or cannot help.
Guardrails
- Answers only from the organisation's own vetted legal information, never from the model's general knowledge or invented case law
- Explicit, repeated statement that the assistant gives general information, not legal advice, and cannot address the person's specific case
- No collection or storage of case specific personal data beyond what is needed to route a referral
- A visible, low friction route to a human for anything urgent, safety related or outside the assistant's content
KPIs to instrument
- Questions answered from content versus referred to a human, by topic
- User reported satisfaction and whether they found what they needed
- Share of conversations where the assistant correctly refused to give case specific advice
- Unanswered or misrouted questions, reviewed to find content gaps
Human in the loop
The assistant never tells a person what to do in their own case; a lawyer, paralegal or self help centre staff member handles anything that needs case specific judgment. Content owners review and update the underlying legal information, and both organisations on this page publish the limits of what their assistant does directly to users.
Common failure modes
- A confident answer that drifts into advice
- Under a natural sounding conversation, the assistant starts to sound like it is telling the person what to do in their case rather than explaining the law generally. Test explicitly for this drift and tune refusals, not only for factual accuracy.
- Content that has not kept up with the law
- A rule, form or deadline changes and the assistant keeps citing the old one. Assign an owner and a review cycle to the underlying legal content, the same discipline any grounded assistant needs.
- People who needed a lawyer never asked for one
- Someone with a case serious or complex enough to need a lawyer gets general information and stops there instead of being referred. Watch for topics and language patterns that suggest higher stakes and route them to a human proactively rather than waiting to be asked.
What are the risks and rules?
EU AI Act
Limited risk (transparency)
An assistant that gives general legal information and explicitly declines to advise on a person's specific case is not listed in Annex III; it carries the Article 50 duty to disclose that the person is interacting with AI. If a deployment were changed to assess a person's eligibility for legal aid itself, rather than only inform and refer, Annex III point 5(a) on essential public assistance services could apply, and the design should be reassessed.
Rules that apply
Controls to put in place
- A written, tested boundary between legal information and legal advice, reviewed by a lawyer
- Clear AI disclosure and a visible route to a human
- Content ownership and a review cycle for every topic the assistant covers
- No case specific personal data retained beyond what a referral needs
Frequently asked questions
- Can this kind of assistant give legal advice?
- No, not in either deployment on this page. Legal Aid of North Carolina's LANC-LIA states directly that it does not provide legal advice and cannot answer specific questions about a person's case, and People's Law School describes Beagle+ as guiding people to legal information rather than telling them what to do; both point people to a lawyer for case specific questions.
- What results have these assistants reported?
- Neither organisation has published usage or outcome figures for its chatbot on the pages cited here. Measure your own deflection, satisfaction and referral rates before building a business case on either deployment.
- How is this different from a general government information chatbot?
- A citizen information assistant answers "which office" and "how do I apply" questions across government services. This use case is narrower and more sensitive: it serves people who are already in or facing a legal case, so the line between general information and legal advice has to be explicit, tested and repeated to the user, not just implied by the assistant's tone.
- Is this high risk under the EU AI Act?
- Not as designed here, since it only informs and refers. It carries the Article 50 transparency duty. A version that assessed someone's eligibility for legal aid itself, rather than pointing them to where to ask, would need to be reassessed against Annex III point 5(a).
How to cite this page
Blits.ai AI Use Case Library, "AI self help assistant for self represented litigants and legal aid clients", last verified 29 September 2026, https://www.blits.ai/ai-use-cases/legal-aid-self-help-assistant. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 30 September 2026: Published after review by an automated review workflow (independent skeptic review).
- 30 September 2026: Second editorial pass after an adversarial review: rewrote the problem paragraph's claim that giving case specific advice is legal advice a tool cannot give, which was an unsourced universal legal claim, as the sourced design choice both organisations state (neither offers legal advice); rewrote feasibility.complexityNote's unsourced claim that neither deployment connects to case data or files anything to match what the sources actually say (both are described as answering general questions only); hedged the unsourced generalisation that people rarely know which page or term covers their problem to 'often do not know'; removed mobile-app from channels since neither evidence record is a mobile deployment; reordered this changelog into consistent reverse chronological order.
- 30 September 2026: Unpublished by an automated review workflow (independent skeptic review).
- 29 September 2026: First published