The legal AI conversation still starts too late.
It starts with drafting. Or document review. Or filings. Or whether a lawyer verified an output before using it.
The first legal AI surface most people touch is intake.
A person searches for help. Lands on a page. Tells a story. Enters contact information. Answers a few questions. Maybe gets routed to a lawyer. Maybe gets routed to no one useful. Maybe gets sold to several firms. Maybe gets steered away from legal aid without ever being told it exists.
That layer now increasingly presents itself as AI.
The problem is not only that legal AI may generate bad text. The problem is that AI-assisted legal intake is becoming a real market before it has a real standard.
The market should stop treating that layer as ungoverned marketing plumbing.
By "national standard," I do not mean waiting for one giant federal AI law.
I mean a shared operating standard for one specific surface:
consumer-facing legal intake.
The Intake Layer Is Infrastructure
Legal intake is often treated as marketing plumbing.
That frame is wrong.
Consumer-facing intake determines:
- what facts get captured
- what the system decides the issue is
- whether urgency is recognized
- whether the person is routed to legal aid, a private attorney, a government agency, or nowhere useful
- who receives the person's information
- how many times that information is sold
- what gets retained
- and whether the user ever understands what happened
That is not peripheral to legal services.
It is infrastructure.
The legal industry already understands that legal work itself needs standards. The intake layer has lagged because it has been treated as lead generation, directory design, or growth tooling rather than as a legal-service boundary of its own.
That was always too lax. AI makes it worse.
Once intake systems start classifying matters, choosing next steps, prioritizing some users over others, and presenting generated legal information on the way to routing, the front door stops being a neutral form.
It becomes decision infrastructure.
The Current Market Has Almost No Shared Discipline
There is no coherent shared operating standard today for consumer-facing AI-assisted legal intake.
There are ethics rules for lawyers. There is growing AI activity at the state level. NCSL's state-legislation tracking now covers introduced and enacted AI measures beginning in 2025, and NCSL reported that state legislators introduced more than 1,000 AI-related measures in 2025 alone. That is a real signal of movement. It is not a substitute for a practical intake standard for legal services.
What exists instead is a messy mix of:
- lead-generation economics
- state-by-state UPL and referral constraints
- scattered law-firm advertising rules
- generic privacy disclosures
- and product teams deciding for themselves what an intake AI is allowed to do
That is not a stable operating model.
It is especially weak for the most sensitive surface in the workflow: the moment a consumer first raises a legal problem and asks for help.
The policy picture is moving. The operating discipline is still thin.
A National Standard Does Not Require Waiting for Congress
"National standard" does not need to mean "Congress passes one giant AI law and solves the problem."
That is not how this is likely to unfold.
The practical regulatory reality is more fragmented. State legislatures are moving faster than the federal government. Bar guidance remains uneven. Consumer-protection and privacy obligations are spreading through different channels.
That is exactly why a clear standard is useful now.
Firms, bar groups, legal-aid organizations, and vendors need a common operating position for one specific surface:
consumer-facing AI-assisted legal intake.
The better path is not to wait for a perfectly coherent national regime.
The better path is to define what a serious intake standard should require, let that standard harden through adoption and criticism, and make it easier for state bars, legal-aid coalitions, and regulators to converge around something real instead of reacting one scandal at a time.
What the Standard Should Actually Require
The legal industry does not need a vague "responsible AI" statement for intake.
It needs a narrower standard with operational teeth that buyers, engineers, operators, bar groups, and legal-aid organizations can actually build against.
I think any serious standard for AI-assisted legal intake should require at least five things.
1. Means-test first
If the system is collecting a legal problem from a consumer, it should check for legal-aid eligibility before defaulting into paid-firm routing.
The natural anchor is the Legal Services Corporation income-eligibility framework, with extension to state-funded and court-based means-tested assistance programs where they exist. This is not an argument for LSC-exclusivity. It is an argument that qualifying consumers should see free-or-reduced-cost options before the paid funnel takes over.
That does not mean every matter belongs with legal aid.
It means the system should not pretend the only possible destination is a paid lead funnel.
For a large share of users, the first important question is not "which firm buys this lead?" It is "does this person qualify for a public-interest path, lower-cost path, or limited-scope path before private-market intake takes over?"
If AI is going to sit at the front door, that front door should not erase legal aid by design.
2. Statutory provenance for legal information
If the intake surface is going to present legal information, explain rights, or help a user understand what kind of issue they may have, it should ground those statements in visible source material.
That does not require turning every intake page into a treatise.
It does require rejecting the worst version of the current market, where the system mixes generated legal explanations with no clear source, no date, no jurisdiction signal, and no separation between public legal information and advice.
ABA Formal Opinion 512 is useful here even though it is lawyer-facing rather than intake-specific. The ABA's message is that lawyers remain responsible for competence, confidentiality, supervision, candor, communication, and reasonable fees when AI is involved. That pushes in one direction architecturally: generated legal output should not float around ungrounded and unauditable.
Intake surfaces that explain legal issues should show where those explanations come from.
Showing where the explanations come from is the starting point. The stronger version of this pillar is that explanations are inspectable — a consumer can see not just a citation but the full retrieval trail behind any legal information the intake surface provided. Which source, which passage, which jurisdiction, what date the source was last verified against the official codification. A citation that is decorative is not a citation; a citation that is auditable is.
3. Distribution discipline
One user should not become a commodity simply because the intake layer got smarter.
The current multi-sell lead model is already bad enough without adding AI to make the capture layer look more trustworthy than it is.
A serious standard should limit how a single intake is distributed and should require clarity about who receives the information and why.
That is not anti-market.
It is anti-opacity.
If a user thinks they are contacting one legal-help destination and their information is in fact being sold broadly into an auction, the system has already failed the trust test.
AI should not make that model harder to see.
4. Economic discipline around lead fees
The legal lead market has a structural incentive problem.
When the capture layer gets paid more by maximizing resale value than by improving routing quality, the product predictably bends toward volume, ambiguity, and pressure tactics.
That is one reason so many intake experiences feel extractive.
Any standard serious enough to matter should address the economics, not only the UX.
That could take different forms. Fee caps are one path. Administrative-cost tests are another. The core idea is simpler than the mechanism:
the intake layer should not be economically rewarded for degrading fit, transparency, and trust.
5. Data minimization and no surveillance tracking on intake flows
This is the most concrete requirement and the one I think the market is still underrating.
Legal intake should collect what is necessary for the matter. It should not quietly operate as surveillance inventory.
That means:
- no third-party tracking pixels on sensitive intake flows
- no casual sharing into data-broker ecosystems
- no model-training use without explicit informed consent
- limited retention tied to the purpose of the intake
- and stronger boundaries around third-party access than generic consumer forms usually apply
Surveillance is the bipartisan AI issue because it is the point where nearly everyone eventually recognizes the same problem: too many systems collect too much sensitive information and expose it to too many actors.
Legal intake is one of the least defensible places to keep doing that.
Consumer-Side Defensibility, Not Just Lawyer-Side Defensibility
A lot of legal AI vendor positioning right now is anchored in lawyer-side defensibility — the lawyer's downstream ability to defend their work to a court or client.
Intake operates one step earlier. The defensibility standard at intake is consumer-side: can the consumer, later, verify what the system told them? Can they audit which sources backed any legal information they received? Can they see whether the system means-tested them, and what it routed them to? Can they recover the record of what they shared and what was retained?
A consumer-facing legal AI surface that cannot answer those questions is not yet a serious intake system. It is a marketing surface dressed in AI.
This Is a Consumer-Protection Issue Before It Is a Technology Debate
The market often argues about legal AI as if the hardest question is whether the model is strong enough.
That is not the first question at intake.
At intake, the first questions are:
- what is this system allowed to decide?
- what is it allowed to say?
- where does the user's information go?
- who gets access to it?
- what path does it default toward?
- what economic incentive is shaping that path?
Those are infrastructure and consumer-protection questions.
They still apply if the model is excellent.
A smoother intake surface can hide a broken operating model better than a clumsy one can.
The Standard Should Be Specific Enough to Build Against
One trap in AI governance is writing standards so abstractly that no engineer, operator, or buyer can tell what compliance would actually mean.
Legal intake does not need that kind of document.
It needs a standard specific enough to shape:
- intake design
- routing defaults
- data collection
- retention policy
- model boundaries
- and distribution economics
without pretending every legal AI issue is the same issue.
This is not an attempt to solve all of legal AI in one move.
It is an attempt to stop leaving the front door ungoverned.
The firms and platforms worth trusting will build to a standard like this before they are forced to. The useful move is not waiting for federal AI legislation. It is publishing the standard, opening the spec, and letting bar A2J commissions, legal-aid coalitions, NIST's CAISI work, and the ULC pull it forward.
Sources
- ABA Formal Opinion 512 (PDF)
- FTC Operation AI Comply press release (Sept 25, 2024)
- FTC DoNotPay matter page
- NIST Center for AI Standards and Innovation (CAISI)
- Legal Services Corporation, Income Eligibility (45 CFR Part 1611)
The standard should be concrete enough that a buyer, bar group, or legal-aid partner can ask:
- does the system means-test before monetizing?
- does it show provenance for legal information?
- does it limit distribution?
- does it constrain the economics?
- does it avoid surveillance tracking and excess retention?
If the answer is no, the product should not get to hide behind generic "responsible AI" language.
Build to the Standard Before the Standard Arrives
The legal AI market does not need more companies racing to put conversational gloss on the same bad intake economics.
It needs front-door systems that are honest about what they are doing, limited in what they are allowed to do, and designed around legal access rather than only lead capture.
The category claim is straightforward.
The most important legal AI standard is not only about document generation, agent autonomy, or what happens inside a drafting workflow.
It is also about the first surface where a person asks for help.
That surface needs more discipline than the current market gives it.
The firms and platforms worth trusting will build to that standard before they are forced to.
Sources
- ABA Formal Opinion 512 (PDF)
- ABA announcement on Formal Opinion 512
- NCSL Artificial Intelligence Legislation Database
- NCSL: New Trends Emerge as States Refine AI Legislation
FlowCounsel builds AI-enabled software for legal teams. FlowLawyers is the consumer-facing legal help platform with attorney discovery, legal-aid routing, state-specific legal information, and document tools. Neither provides legal advice. Attorney supervision of legal AI output is required.