Who Is Liable When an AI Referral Service Gives Bad Legal Advice?

Direct Answer to the Liability Question

Liability for an AI referral service usually depends on what the service actually does, not on the label “AI.” A website that merely supplies names, directories, or general information is less likely to owe the duties of a lawyer, broker, fiduciary, or regulated financial adviser than a service that evaluates a person’s case, recommends a specific professional, accepts a fee, controls the selection process, or makes an affirmative promise about the provider’s competence. The strongest claim generally arises when the referral platform knowingly recommends an unsuitable provider, conceals material conflicts, misrepresents credentials, or receives a commission without adequate disclosure. Ordinary mistakes do not automatically produce liability, although they can support negligence, deceptive-practices, contractual, or fiduciary claims when the applicable legal elements are present.

Also worth reading: How Does AI Match Clients With the Right Legal Service Providers in 2026? · How Do You Compare Legal Service Fees Before Choosing a Lawyer or AI-Assisted Service in 2026? · What Is the Definitive AI Legal Service Pilot Checklist for Enterprise Adoption?

As of October 1, 2026, there is still no generally applicable U.S. statute that makes every AI-generated referral legally attributable to the company operating the tool. Courts instead examine the entity’s conduct, the governing state law, disclaimers, advertising, and the reasonable expectations created by the service. The German decision reported by the Transparency Coalition concerning Google and an AI hallucination illustrates why provider liability can vary by jurisdiction: German court reasoning about a particular statement by Google cannot simply be imported into an American case involving a lawyer-referral platform. A defensible answer therefore separates the AI model, the referral company, the selected professional, and any platform that processes payment or credentials.

How Liability Is Allocated Among the Parties

The first defendant to analyze is the model developer. Its duties may arise from product liability, negligence, contract, privacy law, or an AI-specific statute, but a private referral service ordinarily cannot establish developer liability merely by showing that the model produced a mistaken result. The second actor is the referral company. It may be liable if its own ranking rules, sales representations, screening procedures, or conflict disclosures are misleading, or if employees or contractors disregarded accepted safety controls. The third is the referred lawyer or adviser, who remains responsible for professional judgment and must verify facts before acting.

A payment processor or marketplace may add another layer, but payment processing alone usually does not convert it into the professional offering legal advice. Similarly, a law firm can be responsible for a bad recommendation made by its own intake system, yet the platform supplying that recommendation may be independently responsible if it held itself out as screening candidates for competence and conflicts. Contract terms matter, although an enforceable limitation is unlikely to excuse knowing deception, gross negligence, or conduct that a statute prohibits disclaiming. Practical responsibility can be more complicated than legal responsibility: the model may have generated the text, the broker selected the source, and the lawyer failed to check it, but only some of those actors may have a legally recognized duty to the consumer.

The allocation also changes with the subject matter. Referring someone to a personal-injury lawyer creates different risk than referring a business to an employment-law specialist, securities adviser, tax preparer, or AI vendor. Referral sources already play a familiar commercial role in pension administration, where administrators often use brokers, financial advisers, and other referral channels. In that setting, the relevant questions include credential verification, fee disclosure, conflicts, and whether the administrator retained meaningful discretion rather than accepting an automatically ranked provider.

## Negligence, Advertising, and Fiduciary-Duty Claims

Negligence requires more than showing that a referral was unhelpful. A claimant generally must prove a duty, breach, causation, and damages, although the precise formulation varies by state. A referral company may owe a duty if it presents itself as checking quality, conflicts, licensing, or case fit. Evidence could include the referral score, the information collected, the screening criteria, the time before referral, and warnings that users should conduct a separate search. If the platform ranks a suspended lawyer highly without an explanation, liability becomes more plausible than when a user independently imports an obsolete address into an otherwise neutral directory.

Fiduciary claims require a special relationship involving loyalty, trust, confidence, and discretionary authority. A paid marketplace is not automatically a fiduciary. A service that tells a consumer it will identify the “best” lawyer, while accepting a commission from that lawyer and receiving confidential case details, faces a stronger argument for scrutiny. The company does not cure that problem simply by calling itself an “AI Legal Services Broker.” Courts examine substance: who selected the provider, who earned the fee, whether alternatives were disclosed, and whether the system was designed to benefit the platform as well as the user.

State consumer-protection statutes may be more important than professional rules when the service is not itself a law firm. An unqualified representation that a provider is “AI vetted,” “top rated,” or authorized to handle a specific matter can be treated as an objective claim that must be substantiated. The FTC’s Guides Concerning the Use of Endorsements and Testimonials in Advertising, 16 C.F.R. Part 255, are relevant when testimonials, influencer reviews, or provider-supplied endorsements are used. Because this is legal-content work for a general audience, the key phrase is “AI referral service liability,” but the final legal analysis remains jurisdiction-specific.

Why Generative AI Does Not Itself Answer the Case

Generative AI can summarize intake information, compare published credentials, identify missing questions, and draft a profile. Those functions can improve consistency, but the output may contain fabricated statutes, nonexistent cases, incorrect licensing information, or unsupported competency claims. The system’s ability to produce fluent language makes those errors dangerous; fluency is not evidence that the model performed the required legal checks. A referral workflow should therefore preserve source documents, show the basis for each score, and require a person to confirm the recommendation before a consumer relies on it.

Liability becomes harder where vendors use separate foundation models, retrieval systems, ranking tools, and databases. If the foundation model supplied text but the broker supplied a deliberately false credential label, the broker may be directly responsible for that label even if the model generated the surrounding description. A model developer’s terms may disclaim accuracy, but those terms bind only parties covered by the contract and do not defeat independent statutory duties. Likewise, human review can reduce risk without eliminating it. A reviewer who rubber-stamps thousands of matches is not equivalent to a reviewer who confirms the provider, matter type, jurisdiction, conflict status, and fee arrangement for each referral.

Professional-responsibility literature is increasingly concerned with human oversight, verification, and disclosure when lawyers use generative AI. The references to Thomson Reuters, Harvey, the ABA-related OPR material, and Circular 230 commentary show that AI use is moving into routine legal operations. Referral platforms are not exempt from that trend. Before launch, a broker should ask whether recommendations concern licensed legal services, whether they enter regulated areas such as tax advice, and whether the platform is making a professional judgment or merely operating a search tool.

Practical Steps Before a Consumer Relies on a Match

The consumer should first determine what the service promised. A directory saying “browse providers” generally calls for a different response from a paid broker saying it reviewed credentials and selected a suitable lawyer. The consumer should compare the provider’s license with the relevant state bar, confirm the exact office and individual attorney, and search for public disciplinary history. For federal matters, the lawyer should separately confirm experience with that court and agency; for international matters, the consumer should ask whether counsel is authorized where the work will occur.

Next, the consumer should request a written explanation of the match, including the date it was made and the data used. A useful record names the provider, the referral basis, the service fee, and any relationship between platform and provider. The consumer should ask whether the lawyer has taken the matter, whether conflicts have been checked, and what information the platform retains. A referral is not an engagement, so a prospective client should obtain a separate fee agreement, scope document, and representation consent.

If the recommendation caused a concrete loss—for example, a filing deadline passed, a provider was not licensed on the relevant date, or a disclosed fee was materially misstated—the consumer should preserve the page, receipt, terms, intake responses, recommendation, communications, and timeline. The appropriate response may start with the platform and then involve the state bar’s grievance process, an insurer, a payment dispute, or court. Deadlines may be short in litigation, tax, securities, or criminal matters, so preserving evidence and seeking prompt legal advice is more useful than debating the technology label first.

A service operator should also create an incident process. On discovering a bad license, provider complaint, cyberattack, model error, or undisclosed conflict, it should suspend the affected listing, preserve logs, notify affected users where legally required, and investigate ranking and approval controls. A documented process can demonstrate reasonable care, but it is not a defense to intentional misconduct and should not be used to conceal recurring failures.

Common Mistakes That Weaken Safety and Liability Defenses

The most obvious mistake is treating a disclaimer as a cure-all. Language such as “not legal advice” can clarify a purely informational tool, but it may contradict prominent statements that the service has vetted or matched the “best” providers. A second error is hiding the referral compensation. If a broker receives $500 from a law firm for each matter, for example, that payment should be disclosed in language a reasonable consumer can understand. Commissions are not inherently improper, yet they create incentives that independent screening and written conflict protocols are designed to address.

Another mistake is using a legal-sounding title without defining the service. Calling an unlicensed matching platform an “AI legal broker” does not determine its regulatory status. The operator should avoid implying that the platform itself gives legal advice, represents the consumer, or guarantees an outcome. It should also avoid measuring providers mainly by revenue, paid placement, or an opaque model score. If a provider purchases enhanced placement, that fact should appear beside the “AI recommended” label rather than being buried in terms of service.

Technical mistakes include failing to remove withdrawn or suspended listings, relying on an outdated model, exposing personal intake data in prompts, and treating fabricated citations as harmless hallucinations. The Cambridge University Press & Assessment discussion of AI agents emphasizes potential liability, cybercrime, ethics, and safety, while reports about Anthropic and rogue agents show that autonomy can move risk beyond a simple chatbot interaction. A referral system should therefore use least-privilege access, audit logs, human approval, data deletion schedules, and an incident-response contact. None of these controls is proof that harm cannot occur, but each can establish that the operator took a reasonable, documented step.

How a Regulated Broker Can Reduce Risk

Risk reduction begins by defining the service in one page. The description should say whether the company only lists providers, collects intake information, recommends counsel, facilitates a consultation, or negotiates engagement terms. It should identify the decision-maker, the applicable jurisdictions, and the source and age of provider data. A provider profile should show license date, office, practice areas, fee information, conflicts status, and any paid relationship. The interface should not display a high-confidence recommendation when required fields are absent.

The company should adopt a repeatable approval process. A lawyer or compliance manager can check a sample or every newly approved provider against primary sources, after which automated systems may monitor changes. A stronger model applies a human review to material matches, especially sanctions, urgent matters, high-value transactions, vulnerable consumers, and unfamiliar jurisdictions. The company should keep a versioned ranking policy and test whether protected characteristics or commercial payments distort results. As of October 1, 2026, the company should also monitor enacted state AI laws and professional rules rather than assume that federal model-specific guidance resolves every referral question.

Insurance must match the actual business. General liability coverage may not respond to professional malpractice, privacy incidents, cyberattacks, or the sale of financial services. Operators should ask an insurer or broker whether errors-and-omissions coverage is available and whether the service is conducted by lawyers, a legal services company, or a technology intermediary. A technology policy can cover the platform, while professional coverage addresses services supplied by licensed professionals. The policy should be reviewed annually and after any material change to autonomous matching, agentic transactions, or data use.

Contract language is useful only when it matches conduct. The operator can state that a referral is informational, that no lawyer-client relationship begins until the law firm accepts the engagement, and that users should independently verify credentials. It can also disclose fees, retention periods, and complaint channels. It should not promise that AI is accurate, that users will recover damages, or that the platform is a law firm unless those statements are true and legally supportable.

When to Act and What It May Cost

A consumer should act immediately after discovering an urgent legal deadline, suspected unauthorized practice, identity theft, or an undisclosed adverse decision involving the referred provider. Waiting may allow a limitation period, grievance window, insurance notice period, or professional response deadline to expire. A person should not assume that contacting a referral platform preserves every legal right; preservation notices, evidentiary steps, and court deadlines may require separate professional advice.

A platform should act before a public complaint when it learns that a provider’s license was suspended, a profile contains a fabricated credential, a serious data breach occurred, or automated recommendations are repeatedly producing the same category of error. Containment should precede root-cause analysis: suspend access, preserve records, notify responsible personnel, and assess affected users. A postmortem should then identify whether the failure came from source data, model generation, ranking, contract design, human review, or an external attack.

Most directory searches are free, and referral fees are often paid by the receiving provider, by the consumer, or by both. Some legal marketplaces also charge subscription, lead, or consultation fees, but there is no reliable single U.S. price. As a broad market estimate rather than a legal rule, individual AI legal research tools commonly run from about $20 to $200 per month for a paid account, while enterprise deployments are priced by contract and may cost far more. A referral broker should publish a clear quote before accepting payment; a consumer should reject a fee that remains undisclosed until after an engagement begins. Cost comparisons are incomplete unless they include data processing, provider compensation, screening, insurance, and any required human review.

Overall Legal Assessment

The best defensible position is that an AI referral service can be liable when it performs more than neutral information retrieval, but liability is not automatic merely because an AI-generated recommendation proved wrong. The court would likely examine the service’s promises, degree of control, screening process, compensation, disclosures, professional status, and the user’s reliance. A plain directory, a paid curated broker, and a law-firm referral network can therefore face different duties even when they display nearly identical AI recommendations.

A broker that verifies current credentials, checks conflicts, discloses fees, uses source-grounded information, requires human review, and preserves an audit trail has a stronger factual defense than one that labels unsupported output “AI vetted.” These controls cannot eliminate liability, and they may be legally ineffective where the operator intentionally misleads users. For the consumer, the practical response is to verify the person rather than debate whether the word “AI” caused the error, preserve evidence, and act quickly when money, confidential information, or a legal deadline is at stake.