Ninety-two percent. That is the share of substantial civil legal problems faced by low-income Americans that receive no or inadequate legal help. Not ninety-two percent of people, but ninety-two percent of problems. Three in four low-income households encounter at least one substantial legal issue in a given year, and nearly two in five encounter five or more, such as evictions, consumer disputes, healthcare access, and family law. Nearly all of them go unresolved.
The statistic comes from the Legal Services Corporation’s 2022 Justice Gap Report, and it has barely moved in two decades. The LSC’s first study, in 2005, found that legal aid organizations were turning away one out of every two people who walked through their doors. Nearly two decades later, the turnaway rate is 49%. The math has not changed because the resources have not changed. There are roughly 1.3 million licensed attorneys in the United States. About 10,000 of them work in legal aid.
Meanwhile, the average attorney bills $349 per hour. A contested divorce runs $15,000 or more. An eviction defense is the kind of case where studies find represented tenants roughly twice as likely to win a favorable judgment and several times less likely to lose their homes. It costs between $500 and $5,000, money a tenant facing eviction does not have. Half of all Americans mistakenly believe they are entitled to a free lawyer in civil cases. They are not. The Sixth Amendment guarantee of counsel applies only in criminal proceedings. In civil court, you are on your own.
In family courts across the country, 80 to 90 percent of cases involve at least one unrepresented party. In Miami-Dade’s courts, a one-day census found 63 percent of litigants self-represented. These are not edge cases. This is the system operating as designed, and operating poorly.
The traditional model cannot scale
The conventional response to the access crisis calls for more funding for legal aid and more pro bono hours from private attorneys. Both are necessary. Neither is sufficient.
Legal aid funding has been essentially flat in real terms for thirty years. The Florida Bar Foundation has demonstrated that every dollar invested in civil legal aid returns seven in economic impact, and we are choosing not to make that investment anyway. Pro bono, admirable as it is, cannot close a gap this wide. Only half of Florida attorneys report doing any pro bono work at all, and even if every one of them doubled their hours, the math would not work. The need is measured in millions of cases. The supply is measured in thousands of volunteers.
The technology conversation usually enters here, and it usually goes wrong. The standard pitch holds that if you give lawyers AI tools and make them faster, the efficiency gains trickle down to clients. The early data supports the premise. Thomson Reuters’ 2025 Future of Professionals Report projects that AI will free up nearly 240 hours per attorney per year. A Harvard Law School Center on the Legal Profession report documented a high-volume litigation response workflow dropping from 16 hours to under four minutes. In benchmark testing, AI contract review has beaten lawyers on accuracy (94 percent against 85) in a fraction of the time. Legal aid organizations have noticed, and 74 percent are already using AI tools, nearly double the adoption rate in the broader profession.
But faster lawyers do not automatically produce cheaper legal services. According to Clio’s 2025 Legal Trends Report, firms with above-average productivity are growing revenue, not lowering prices. Among firms that adjusted pricing in response to AI, more raised their rates than lowered them. The billable hour absorbs the surplus. The incentive structure rewards throughput, not access.
Making lawyers faster is a fine goal. It is not, by itself, a solution to the access problem.
A market where none exists today
The conventional technology pitch starts with the lawyer. The model I envision starts with the client.
Consider the person facing eviction. She knows the facts of her case better than any attorney ever will. She lived them. She knows when she paid rent, what the landlord said, what condition the unit was in. What she does not know is how to organize those facts into a legal argument, which statutes apply, what procedural steps to follow, or how to draft a motion a judge will take seriously.
That gap, between knowing the facts and framing them legally, is where AI can do its most important work. Not by replacing the lawyer, but by replacing the blank page. An intelligent tool can take a client’s plain-language account of what happened and produce a structured draft of organized facts, applicable authority, and procedural requirements, formatted for filing. The client does 90 percent of the substantive work by supplying the facts, reviewing the output, and confirming the narrative is accurate. The tool does the legal packaging.
What remains is review. Not sixteen hours of case-building. Not an initial consultation, a retainer agreement, and a series of billable conferences. A focused review of a pre-assembled, AI-structured filing by an attorney who examines the legal framework, flags errors, suggests corrections, and sends it back. The client files. The attorney never enters an appearance. The client remains pro se, but assisted pro se. Not the person staring at a blank form in a courthouse hallway. Someone who arrived with a structured draft a lawyer has already reviewed, built on authority the tool has already checked against its database.
Here is where this stops being a policy argument and becomes a market.
Right now, it is not practically possible to hire a lawyer for a $25 task. The overhead alone (intake, conflicts check, file setup, billing) makes it uneconomic for the attorney and invisible to the client. There is no product at that price point. There is no market.
AI changes the unit economics. If the tool has already assembled the facts, identified the law, structured the filing, and checked the citations against the database, what remains for the attorney is judgment, and judgment at that stage can be exercised quickly. A competent lawyer reviewing an AI-packaged eviction response or a simple motion can assess form and substance in six or seven minutes. Eight to ten reviews an hour. At $25 per review, that is $200 to $250 an hour, which is competitive with or better than what most solo practitioners earn billing traditional hourly rates, without the overhead of client acquisition, intake, or case management.
For the client, $25 buys what was previously inaccessible. It buys a lawyer’s eyes on their filing before it goes to a judge. Not full representation. Not an ongoing attorney-client relationship. A discrete, bounded service (review and approval of an AI-assembled work product) delivered on demand.
For the attorney, the model offers volume-based income on pre-packaged work. No marketing. No intake calls. No accounts receivable. The platform delivers the work, the lawyer applies judgment, the client gets a reviewed filing. Everybody creates value. Everybody captures value. And the platform that connects them, which built the AI, checked the citations against the database, and matched the client with the reviewer, sustains itself on each transaction.
Most legal aid organizations serve fewer than half of those who request services. This is not a legal aid model. It is a market. The access crisis is not solved by charity. It is solved by lowering input costs to the point where market-driven solutions reach the people who need them. On demand. No taking a number and waiting in line.
The pieces exist, but the marketplace does not
This is not a new idea so much as an unbuilt one. The components are visible in isolation. Nonprofit tools like Upsolve guide low-income filers through Chapter 7 bankruptcy without an attorney. HelloDivorce offers guided divorce with optional attorney add-ons in a handful of states. Courtroom5 and Prosēi AI help pro se litigants organize their cases. On the attorney side, platforms like Harvey and CoCounsel are making lawyers faster. Each gets a piece of the model right.
None connects all three sides of the transaction. The document-assembly tools produce output without attorney review. The attorney-facing AI tools never touch the client. And the platforms that do offer attorney access (LegalZoom, Rocket Lawyer) sell it as an expensive add-on to a template, not as a six-minute review of an AI-structured filing. There is no marketplace that takes a client’s facts, packages them with verified legal authority, routes them to an attorney for rapid review, and brokers the transaction at a price both sides can live with.
The nearest cautionary example is DoNotPay, which branded itself “the world’s first robot lawyer” and drew an FTC enforcement action for overpromising what its AI could deliver. The lesson matters. A client-facing legal AI tool without attorney oversight and without reliable legal authority is not a product. It is a liability.
The hard regulatory questions
The marketplace described above does not exist in a regulatory vacuum, and anyone building it will need to confront several unresolved questions head-on.
The first is unauthorized practice of law. If the AI structures a legal filing based on the client’s facts (identifying applicable statutes, drafting arguments, formatting for a specific court), is the platform practicing law? The answer likely depends on jurisdiction, on how much legal judgment the tool exercises versus how much it merely organizes, and on whether an attorney reviews the output before it reaches the client. These lines are not clearly drawn, and the case law is thin. Florida Bar Ethics Opinion 24-1 addresses attorney use of AI but does not squarely reach the question of a client-facing tool that generates legal filings for attorney review.
The second is scope of representation. If a lawyer reviews a filing for $25 and sends it back with corrections, what is that relationship? Limited-scope representation under Rule 4-1.2(c)? Something less? A discrete consultation that never rises to an attorney-client relationship? The answer matters for malpractice exposure, conflicts obligations, and trust account requirements. Florida permits unbundled legal services, but the rules contemplate a more traditional engagement than a six-minute platform-mediated review.
The third is quality control at scale. A lawyer reviewing eight to ten filings per hour must be able to trust the AI’s work product enough to exercise meaningful judgment in the time allotted. If the underlying tool is unreliable, speed becomes recklessness. The reviewing attorney’s ethical obligations do not shrink because the platform made the task feel quick.
None of these problems is fatal. All of them demand careful design in the platform’s terms of service, in its attorney-onboarding process, in its insurance arrangements, in the tool’s own reliability. Building the marketplace is not just an engineering challenge. It is a regulatory and ethical challenge, and the builders had better take it seriously.
None of this works if the tools fabricate authority
There is a precondition, and the industry is not meeting it.
In 2025, researchers at Stanford’s RegLab and Human-Centered AI Institute published what remains the most rigorous peer-reviewed study of legal AI accuracy. They tested the leading commercial platforms, the ones attorneys are actually paying for. The results were sobering. Even the best-performing legal-specific platforms (retrieval-augmented tools marketed to practitioners, not general-purpose chatbots) produced fabricated or unsupported legal citations at least 17 percent of the time. One major platform hallucinated on more than a third of queries. A companion Stanford study of general-purpose chatbots found worse still, with hallucination rates from 58 to 88 percent on legal questions.
Judges across the country (more than 300, by one tracker’s count) have issued AI-disclosure requirements. Courts have imposed sanctions ranging from $2,500 to $30,000 for AI-generated filings containing fabricated citations. In Florida, three judicial circuits put administrative orders on the books before the state’s supreme court overtook the patchwork with a statewide rule. Amended Rule 2.515, effective June 15, 2026, makes every signer of a court filing certify that the authorities cited exist and are accurately cited.
A tool that fabricates authority one time in six cannot anchor this model. The entire value proposition depends on the reviewing attorney trusting the legal framework the AI has assembled. If the lawyer has to independently verify every citation, the six-minute review becomes a sixty-minute research project, and the economics collapse. The $25 price point vanishes. The market disappears. And the 92 percent go back to being unserved.
This is why trustworthiness is not a feature request. It is a load-bearing wall. A tool built for assisted pro se work must be built with verification at the system level, so that the citations a client sees, and the attorney reviews, have already been checked against real authority before they reach the screen. That is a harder engineering problem than generating fluent text. It is also the only engineering problem that matters.
Two pillars, one standard
Two policy positions animate my practice of law, and they complement each other. The first is that competent legal representation should not be a luxury good. The second is that the tools lawyers and their clients rely on must be worthy of that reliance. Legal services should actually be accessible, and legal AI should actually be intelligent.
These are not competing values. They are the same value. You cannot put powerful legal tools in the hands of unrepresented people and then shrug when those tools hallucinate. You cannot build a market for $25 legal review on a platform that fabricates one out of every six citations.
The technology exists to deliver legal services to more people, for less money, at higher quality, not by making lawyers faster at the same work, but by creating a new kind of transaction, one where the client supplies the facts, the AI supplies the law, and the lawyer supplies the judgment. Everybody pays less. Everybody gets more. Whether that happens turns on whether the profession, and the people building its tools, insist on getting it right. The people who can least afford a lawyer are also the people who can least afford a bad one, human or otherwise.