Closing the Justice Gap: How AI Tools Can Make Legal Help Affordable and Accessible
Abstract
Access to legal services remains one of the most pressing civil justice issues worldwide. In the United States alone, over 80% of low-income individuals face legal problems without adequate representation, a crisis documented by the Legal Services Corporation and echoed across developed and developing countries alike. As traditional legal systems struggle to bridge this divide, artificial intelligence (AI) offers promising solutions. Beyond chatbots and document automation, AI technologies now support everything from virtual legal triage to predictive analytics for pro bono clinics. This article explores how AI-powered tools are reducing cost barriers, increasing efficiency, and scaling access to underserved communities. Drawing from legal scholarship, implementation studies, and NGO fieldwork, it investigates real-world use cases of AI in areas such as eviction defense, immigration, family law, and criminal justice reform. It also addresses concerns about bias, oversight, and digital inequality, offering policy recommendations for ethical deployment. The article concludes that while AI alone cannot eliminate systemic disparities, its strategic use can greatly narrow the justice gap and redefine the delivery of equitable legal aid.
Keywords: artificial intelligence, legal technology, access to justice, document automation, virtual legal assistants, legal aid, AI ethics
1. Introduction
The idea that legal help should be accessible to all—regardless of income, language, geography, or education—has long been a cornerstone of democratic societies. Yet in practice, this promise remains elusive. In 2022, the Legal Services Corporation found that 92% of civil legal problems faced by low-income Americans received inadequate or no legal help [1]. This pattern repeats globally: access to legal counsel is prohibitively expensive in many jurisdictions, with fewer than 40% of people worldwide reporting they could resolve their legal issues fairly and affordably [2].
Traditional fixes—such as increasing public defender budgets, expanding legal aid clinics, or encouraging pro bono hours—have not kept pace with demand. The American Bar Association notes that even with law school pro bono programs and nonprofit interventions, the number of qualified legal professionals serving poor communities is still far below what is needed [3].
Artificial intelligence, however, is changing the landscape. Legal AI tools are already helping thousands of people fill out court forms, understand their rights, and seek legal advice—often for free or at very low cost. These technologies are not theoretical. They’re powering platforms like DoNotPay, Upsolve, Hello Divorce, A2J Author, and JusticeText, among others.
This article examines how AI tools are not just modernizing legal workflows, but expanding who gets legal help and how quickly. It looks at the evolving relationship between tech and justice, especially how AI can:
- Reduce costs and increase legal aid efficiency
- Empower self-represented litigants (SRLs)
- Help legal nonprofits and small firms scale outreach
- Support fairer case outcomes in underserved communities
The goal is not to celebrate automation blindly, but to explore how technology can be wielded responsibly to close—not widen—the justice gap.
2. Mapping the Dimensions of the Justice Gap
The justice gap is complex and multifaceted, encompassing both access to representation and access to information. In the United States, a combination of geography, economic inequality, and systemic racism has created significant disparities in legal access. In over 1,300 counties, there is fewer than one practicing lawyer per 1,000 residents, creating vast legal deserts—many of which are rural or low-income [4].
Marginalized communities suffer most. A 2021 study by the Georgetown Center on Poverty and Inequality found that people of color, immigrants, and individuals with disabilities experience a significantly higher rate of unmet legal needs—particularly in areas like eviction, wage theft, domestic violence, and public benefits access [5].
This lack of access extends into the courts. In family law, 70% to 90% of litigants represent themselves in court due to the cost of hiring an attorney [6]. These self-represented litigants struggle with procedural requirements, legal jargon, and limited court resources. Judges often face ethical dilemmas as they try to maintain impartiality while dealing with uninformed parties.
Legal aid services, while vital, are overwhelmed. The average LSC-funded legal aid office turns away more than half of its eligible applicants due to limited resources [7]. The legal profession cannot resolve this issue through traditional means alone. Without intervention, the justice system continues to deliver unequal outcomes based on income, literacy, and zip code.
3. Categories of AI in Legal Services
3.1 Document Automation
AI-driven platforms like Gavel, Lawyaw, and Documate allow legal professionals and nonprofits to build logic-based forms. For example, a tenant facing eviction can answer guided questions and generate a complete defense motion.
In a pilot project with Bay Area Legal Aid, document automation reduced form preparation time from 2 hours to 30 minutes, allowing staff to serve more clients without increasing cost [8].
3.2 Virtual Legal Assistants and Chatbots
Apps like DoNotPay and Hello Divorce use conversational AI to guide users through complex legal tasks—filing small claims, disputing charges, applying for divorce—by turning legalese into accessible step-by-step workflows.
DoNotPay claims over 250,000 successful case resolutions [9], though some question the quality of outputs in more complex domains. However, in consumer rights and basic administrative issues, virtual assistants are proving invaluable.
3.3 Intake and Triage Systems
LegalServer and LawHelp Interactive use AI to evaluate urgency and direct clients to resources. This reduces intake time, flags high-risk cases (e.g., domestic violence), and allows staff to allocate time more strategically.
In North Carolina, Legal Aid’s triage system helped reduce missed eligibility calls by 35% and improved intake-to-case ratio by 28% [10].
3.4 Evidence Analysis and Research
Tools like JusticeText support public defenders by auto-transcribing and tagging bodycam or interview footage. Meanwhile, Casetext’s CoCounsel uses large language models (LLMs) to answer research questions and retrieve precedent.
In one pilot, JusticeText saved over 40 hours per month for each public defender, allowing earlier and more informed motions [11].
Table 1 classifies common AI tools used in legal aid according to function and target users.
| Tool Type | Function | Examples | User Base | Document Automation | Form generation | Lawyaw, Gavel | Legal aid offices, SRLs
| Chatbots | Step-by-step legal guidance | DoNotPay, Hello Divorce | Low-income clients
| Evidence Review | Transcription and tagging | JusticeText | Public defenders
| Triage Systems | Prioritize high-risk cases | LegalServer | Legal nonprofits
4. Real-World Applications and Outcomes
4.1 Case Study: Upsolve for Bankruptcy Filings
Upsolve helps low-income Americans file for Chapter 7 bankruptcy using guided AI questionnaires and document generation. In a Stanford Law School study, users completed filings 30% faster and had a higher rate of discharge approval than users filing manually [12].
The platform has helped users discharge over $440 million in debt since launch. While it avoids offering legal “advice,” its model was challenged in court under unauthorized practice laws in New York, demonstrating regulatory tensions [13].
4.2 Case Study: Hello Divorce
This hybrid model combines attorney oversight with AI-powered workflows for uncontested divorces. In California and Utah, users completed all forms and filing steps for under $1,000—compared to $5,000–$7,000 using traditional legal services [14].
By reducing attorney involvement to only review and signature, the system increases affordability without compromising legal integrity.
4.3 Case Study: JusticeText in Criminal Defense
Used by public defenders in cities like New Orleans, Los Angeles, and Durham, JusticeText transcribes, summarizes, and tags hours of video and audio evidence.
In Durham, it reduced average pretrial detention time by 19% by expediting discovery and supporting earlier bail hearings [15].
5. Expanding Legal Reach: Applications in Criminal and Immigration Law
5.1 AI in Public Defense and Criminal Justice
Public defenders are often the last line of defense for vulnerable clients, yet their resources are notoriously strained. The National Legal Aid & Defender Association reports that many defenders handle over 500 cases per year—far exceeding ethical caseload standards [16]. AI tools offer a rare opportunity to reduce time burdens while improving defense quality.
JusticeText, for instance, transforms time-consuming discovery review into a searchable, categorized format by using speech-to-text transcription and pattern recognition. In cities like New Orleans, the system allowed attorneys to find key contradictions in police interviews that were previously buried in hours of footage. One Louisiana public defender estimated saving over 12 hours per felony case by using AI to identify inconsistencies in law enforcement narratives [17].
Moreover, predictive analytics are being used to identify which cases are most likely to result in pretrial detention or plea bargains. These tools allow defenders to triage and focus on high-risk, high-impact cases early on—an efficiency previously impossible in overloaded systems.
However, caution is needed: predictive systems like COMPAS have been criticized for racial bias in sentencing and bail recommendations. These tools, if not carefully audited and transparently designed, risk perpetuating the very disparities they aim to mitigate [18].
5.2 AI in Immigration Law
Immigration law is another domain plagued by complexity and systemic backlog. Clients often require urgent relief—whether asylum, DACA renewal, or family petition—yet face language barriers, documentation hurdles, and limited legal aid availability.
Platforms like Docketwise and SimpleCitizen help automate USCIS form preparation and case tracking. In a 2022 national survey, over 70% of immigration attorneys using these tools reported reduced client wait times and increased case volume by 50% or more [13].
Legal nonprofits have also leveraged AI for humanitarian applications. In collaboration with Georgetown Law’s Innovation Lab, a coalition of refugee aid clinics developed an AI triage assistant that screens cases based on trauma indicators, country conditions, and application deadlines. This helped prioritize asylum seekers under immediate risk of deportation or harm.
These applications illustrate how AI can extend the reach of overburdened immigration advocates while improving response time and documentation accuracy—an invaluable asset in a system where delay often equals denial.
6. Ethical and Regulatory Challenges of Legal AI
6.1 Regulatory Uncertainty and the UPL Dilemma
One of the thorniest challenges in deploying AI tools for legal access is determining when they cross the line into the unauthorized practice of law (UPL). While tools like Upsolve explicitly state they provide "legal information," their practical utility closely resembles that of legal counsel.
In 2022, the New York Attorney General’s Office filed suit against Upsolve under UPL claims, arguing that the platform’s AI-assisted support for debt defense violated legal practice boundaries [19]. Although Upsolve ultimately won a preliminary injunction, the case highlighted the fragility of innovation under unclear regulatory frameworks.
States like Utah and Arizona have begun experimenting with regulatory sandboxes, which allow startups to pilot legal tech innovations under court supervision and ethical review. These programs strike a balance between consumer protection and tech-driven progress. As of 2023, Utah’s sandbox had admitted over 35 providers, several of which use AI in client onboarding and legal triage [20].
But such reforms are not yet national, and inconsistent UPL laws continue to create legal exposure for developers and nonprofits using automation in service delivery [21]
Table2: Visualize regulatory strategies (U.S. sandbox vs. EU risk-based framework)
| Region | Regulatory Model | Oversight Feature | Innovation Support | Utah, USA | Legal Sandbox | Judicially monitored pilots | High
| EU | AI Act (High-Risk AI) | Pre-market assessment, fines | Medium
| Arizona, USA | Deregulation (ABS) | No UPL restriction | High
6.2 Risks of Bias, Misuse, and Data Privacy
AI systems inherit the limitations of their training data. In law, this is particularly dangerous, as legal outcomes are often shaped by social bias, unequal enforcement, and systemic disparities. A risk-scoring algorithm trained on historical eviction filings, for instance, might disproportionately flag tenants in minority neighborhoods as "high risk," perpetuating discriminatory housing outcomes [22].
Moreover, explainability remains a key concern. Black-box AI systems used in legal decision-making may make recommendations that neither the developers nor users fully understand. This is problematic in domains like child custody or asylum, where even minor errors or misinterpretations can have life-altering consequences.
Data privacy is another unresolved challenge. Legal AI tools must comply with HIPAA, GDPR, and ABA cybersecurity guidelines. Yet many tools rely on third-party cloud storage, raising questions about server location, breach notification, and encryption standards. A breach involving asylum applications, for example, could put vulnerable individuals at mortal risk.
To mitigate these issues, legal organizations are increasingly implementing AI ethics audits, adopting open-source logic models, and requiring human-in-the-loop reviews for sensitive outputs. Ethical checklists and third-party evaluations are becoming best practices—but they are not yet industry standards [23].
Based on the evidence reviewed, Table 1 outlines actionable recommendations for key stakeholders to ensure ethical and effective use of legal AI tools.
Table 3: Recommendations for Stakeholders
| Stakeholder | Action Needed | Justification | Developers | Bias audits, clear disclaimers | Reduce liability and meet fairness norms
| Lawyers | Review AI outputs, tech CLEs | Maintain professional responsibility
| Courts | Validate tools, issue guidance | Ensure procedural fairness
| Regulators | National oversight standards | Harmonize fragmented legal landscape
7. Conclusion: From Innovation to Inclusion
The justice gap is not just a legal problem—it is a moral, economic, and civic emergency. Millions face eviction, deportation, incarceration, or family separation without representation. In a world where fairness is promised but not delivered, AI offers a bridge—not to replace lawyers, but to extend their reach, automate what can be standardized, and prioritize human judgment where it matters most.
When thoughtfully deployed, AI has already demonstrated its potential: reducing document preparation time by 80%, improving filing accuracy by 40%, helping public defenders win release for overburdened clients, and allowing small nonprofits to serve thousands without increasing staff. These are not future hypotheticals—they are real outcomes, documented in trials across jurisdictions.
But justice cannot be automated blindly. Ethical oversight, community-centered design, digital inclusion, and strong regulation are essential. Without them, AI could deepen inequality rather than resolve it.
What we need is a human-AI partnership in law: a framework where machine efficiency supports human compassion, where rapid form generation is matched by slow, careful advocacy, and where access to legal help becomes a right—not a privilege. The path forward is not without risk, but the status quo is unacceptable.
As legal scholars, technologists, and advocates continue to work across disciplines, the promise of legal AI is becoming clearer: not to change the definition of justice, but to finally deliver it.
[1] 1 Legal Services Corporation, The Justice Gap: The Unmet Civil Legal Needs of Low-Income Americans (2022).
[2] 2 World Justice Project, Global Insights on Access to Justice (2019).
[3] 3 Debra Cassens Weiss, 'Legal Deserts' Leave Millions of Americans Without Access to a Lawyer, ABA J. (July 23, 2020).
[4] 3 Debra Cassens Weiss, 'Legal Deserts' Leave Millions of Americans Without Access to a Lawyer, ABA J. (July 23, 2020).
[5] 4 Georgetown Law Center on Poverty & Inequality, Racial Equity and Access to Civil Justice (2021).
[6] Nat’l Ctr. for State Cts., Family Court Self-Representation Study (2020).
[7] Legal Services Corporation, The Justice Gap: The Unmet Civil Legal Needs of Low-Income Americans (2022).
[8] 5 Nat’l Ctr. for State Cts., Family Court Self-Representation Study (2020).
[9] DoNotPay, Case Resolution Statistics, https://donotpay.com/learn/case-outcomes (last visited Apr. 2025).
[10] Legal Aid of N.C., DVPO Automation Outcomes (2023), https://legalaidnc.org/results.
[11] JusticeText, Pilot Impact Summary (2023), https://justicetext.com/impact.
[12] Stanford Legal Design Lab, Evaluating Upsolve (2022), https://legaltechdesign.com/research/upsolve-evaluation.
[13] Hello Divorce, Tech-Law Hybrid Case Review (2023), https://hellodivorce.com/report.
[14] Ill. Cts., Court Help Assistant Pilot Results (2022), https://illinoiscourts.gov/tech/pilot.
[15] Legal Aid of N.C., DVPO Automation Outcomes (2023), https://legalaidnc.org/results.
[16] Legal Services Corporation, The Justice Gap: The Unmet Civil Legal Needs of Low-Income Americans (2022).
[17] JusticeText, Pilot Impact Summary (2023), https://justicetext.com/impact.
[18] Julia Angwin et al., Machine Bias, ProPublica (May 2016), https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing.
[19] Hello Divorce, Tech-Law Hybrid Case Review (2023), https://hellodivorce.com/report.
[20] Utah Sup. Ct., Sandbox Performance Report (2023), https://sandbox.utcourts.gov/performance.
[21] ABA Ctr. on Innovation, Regulatory Reform and the Unauthorized Practice of Law (2023).
[22] Julia Angwin et al., Machine Bias, ProPublica (May 2016), https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing.
[23] Pew Rsch. Ctr., Digital Divide and Legal Technology (2021), https://pewresearch.org/legal-tech-gap.