DoNotPay called its service the “world’s first robot lawyer.” In website advertising reproduced in the Federal Trade Commission’s September 25, 2024 complaint, the company promised to “generate perfectly valid legal documents in no time” and “replace the $200-billion-dollar legal industry with artificial intelligence.” Those were claims about performing legal work, not merely helping someone write a letter.
The FTC’s account was more specific than a charge that artificial intelligence makes mistakes. The agency alleged that DoNotPay had not tested whether its service’s output matched a human lawyer’s work. The missing evidence concerned the comparison the company itself had advertised.
What did the company promise?
The FTC’s complaint placed the robot-lawyer label beside concrete claims about legal tasks. Those included allowing consumers to “sue for assault without a lawyer” and producing valid legal documents. Read together, the representations offered a substitute for professional work, not simply access to a document generator.
That distinction matters. A service can produce a legal document without demonstrating that the document accomplishes its intended legal purpose. Producing text and establishing professional equivalence are different claims. The second requires evidence about the quality of the work, not just the existence of an output.
The FTC’s Policy Statement on Deception directs the agency to examine representations in context, including their likely effect on reasonable consumers. An advertiser does not get to isolate a modest description of one feature from a larger promise of professional replacement. Here, the advertised comparison was with a lawyer.
What testing was missing?
According to the September 2024 complaint, DoNotPay had not conducted testing to determine whether its AI chatbot’s output was equal to the level of a human lawyer. The FTC also alleged that the company had not hired or retained attorneys to test its law-related services.
These are separate deficiencies. The first concerns comparative evidence: had anyone established that the product performed at the advertised professional level? The second concerns qualified evaluation: had lawyers tested the services being sold as a replacement for their work?
Neither question is answered by showing that a chatbot responds to a prompt. Nor is it answered by showing that the software generates a document with the requested heading. Those demonstrations establish functions. They do not establish the comparison DoNotPay advertised.
The complaint’s allegation was therefore not that DoNotPay lost a disclosed, head-to-head trial against lawyers. It was that the company made the replacement claims without conducting the testing needed to assess that comparison. There was no lawyer-equivalence score in the FTC’s account to inspect or debate.
When did the evidence have to exist?
Before the advertising, not after the regulator asked for it.
The FTC’s Policy Statement Regarding Advertising Substantiation says advertisers must possess a reasonable basis for objective claims before disseminating them. The required support depends on the claim, the product, the consequences of error, and the evidence experts would reasonably expect.
That policy does not impose one universal testing protocol on every product carrying an AI label. It does reject the proposition that a company may advertise an objective capability first and establish its basis later. Evidence developed afterward does not erase the initial failure to possess substantiation.
Section 5 of the Federal Trade Commission Act supplies the underlying prohibition on unfair or deceptive acts or practices. The legal issue was not whether DoNotPay used fashionable technology. It was whether the company had a reasonable basis for the capabilities it sold.
The FTC had also addressed the comparison problem directly before this case. Its February 27, 2023 business guidance, “Keep your AI claims in check,” warned companies against advertising that an AI product outperforms a non-AI alternative without adequate proof. Calling the product AI did not reduce the evidentiary burden.
Did DoNotPay admit the allegations?
No. The distinction between an agency complaint and an established fact belongs at the center of this case.
The September 2024 consent materials provided that DoNotPay neither admitted nor denied the complaint’s allegations, apart from the jurisdictional matters specified in the agreement. Accepting restrictions through a settlement is not the same as admitting the alleged testing failures.
The FTC’s own enforcement overview distinguishes settlement through a consent order from litigation of a contested administrative complaint. A complaint states the agency’s charges. A contested proceeding provides a route for deciding those charges on a developed record. The materials examined here are the September 2024 complaint and proposed settlement, not a judicial determination of the product’s capabilities.
The procedure also matters. Under 16 C.F.R. § 2.34, an accepted proposed consent agreement is placed on the public record for comment before the Commission decides whether to make the order final. The September announcement described a proposed resolution, not the result of a trial.
What would the settlement require?
The September 2024 proposed order required DoNotPay to pay $193,000. It also required a notice to consumers who subscribed between 2021 and 2023, warning them about limitations of the service’s law-related features.
The central advertising restriction addressed the actual comparison. DoNotPay would be prohibited from making claims that its service could substitute for a professional service without evidence supporting those claims. The restriction was broader than retiring the phrase “robot lawyer.” Changing the label would not supply proof for the same underlying promise.
These provisions served different purposes. The payment imposed a monetary obligation. The notice addressed people who had already subscribed. The substantiation requirement governed future representations about replacing professional work.
None of those provisions supplied a validation study. The $193,000 was a settlement obligation, not a measured error rate or a calculation showing how often the service failed. The proposed order established conditions for advertising. It did not certify the product as a lawyer’s equivalent.
What does the record establish?
The record supports a narrow but consequential judgment. DoNotPay made a professional-replacement claim. The FTC identified an absence of comparative testing and attorney evaluation as the basis for challenging that claim. The company agreed to a proposed remedy that would require evidentiary support for professional-substitution advertising.
The public record does not establish that DoNotPay had evidence its AI could replace a human lawyer. It contains the replacement promise and a proposed settlement requiring substantiation, but no disclosed demonstration of lawyer-equivalent performance. The replacement claim outran the evidence disclosed for it.
