Consumer AI promised cheaper legal work. Instead, it is quietly generating more of it.
By Elliot Steinmetz, Partner, Abrams Fensterman, LLP *
Not long ago, a client forwarded me a twelve-page “analysis” of a commercial purchase agreement, accompanied by a cheerful note: “Ran this through AI to save us some time.” The analysis flagged the absence of a “mandatory force majeure carve-out for title defects” (no such thing exists), proposed deleting the survey objection deadline because it was “unfavorable to buyer” (our client was the buyer, and the deadline protected it), and recommended a governing-law clause selecting a state in which neither party, nor the property, had ever set foot. It also suggested, with the serene confidence of a first-year associate who has never been yelled at, that we “consider” a 1031 exchange on a transaction involving a nonprofit. The client did not save time. The client bought roughly four hours of my time, spent explaining why the robot was wrong.
This is the quiet scandal of the AI era in legal practice. The industry spent two years promising that artificial intelligence would compress legal fees. It may yet. But a new and largely unpriced force is pushing in the opposite direction: clients armed with consumer-grade chatbots, generating confident nonsense at scale, and handing it to their lawyers to clean up. Unless something changes, legal bills are about to go up, and clients will have only themselves, and their favorite chat window, to thank.
Start with the obvious. General-purpose AI tools are extraordinary at producing text that sounds like law. They are far less reliable at producing law. They do not know the deal, the market, the jurisdiction, the negotiating history, or the fact that the seller’s counsel conceded the environmental indemnity three drafts ago in exchange for a shorter due diligence period. They do not know that the “missing” estoppel requirement was intentionally dropped because the anchor tenant refused to sign one. What they produce, therefore, is a kind of legal fan fiction: plausible in tone, untethered in substance, and delivered without a hint of self-doubt.
Real estate transactions are particularly fertile ground for this mischief. A title commitment run through a chatbot comes back with a breathless list of “critical risks,” including routine utility easements, a long-expired mortgage that was released in 1987, and a standard exception for rights of parties in possession. A lease abstract returns a CAM reconciliation provision that the tool has helpfully “improved” into something no landlord would sign and no court would read the way the client assumes. A tenant redlines a lease by pasting in AI-generated assignment language that would, if accepted, prohibit the tenant’s own planned corporate restructuring. Each of these must be read, understood, reverse-engineered, and answered. Each answer must be diplomatic, because the client is proud of its homework. That is not five minutes of work. That is billable time, and it lands on the client’s invoice.
Then there is the client who arrives “fully prepared.” This is the person who opens the kickoff call by announcing that they have “already done the research” and simply need the lawyer to “confirm a few things.” The research, it emerges, includes a case that does not exist, a statute from the wrong state, and a firm conviction that a liquidated damages clause can be drafted to recover “any and all” losses without being an unenforceable penalty. Litigators know the species well: the client certain they are sitting on a slam-dunk, million-dollar verdict, until counsel reads the chatbot’s analysis and discovers that the controlling authority says the opposite, assuming it exists at all. Seasoned practitioners now spend an increasing portion of their days in a peculiar new genre of advocacy: not persuading opposing counsel, but gently persuading their own clients that the machine was hallucinating. It is exhausting, it is delicate, and it is expensive.
The cruelest twist is that law firms have actually done what clients asked of them. Many firms have invested heavily in purpose-built legal AI platforms, tools trained and configured for legal work, deployed inside secure environments, connected to the deal documents, and supervised by lawyers who know what they are looking at. Used properly, these platforms do precisely what clients want. They accelerate diligence review, surface inconsistencies across lease portfolios, generate first drafts that reflect the firm’s precedent, and eliminate the redundant hours that have long inflated transactional bills. The efficiency is real.
But efficiency requires a clean starting point. When a client has already pre-processed its emails, diligence materials, and draft documents through a consumer chatbot, sometimes rewriting the documents themselves and sending the altered versions to the other side, the lawyer’s first job is no longer legal work. It is forensic archaeology. Which version of the rent roll is authentic? Did the seller actually agree to that representation, or did the client’s AI “summarize” an email into a concession that was never made? Was that redline the counterparty’s, or the chatbot’s? Every hour spent untangling the confusion is an hour the firm’s own AI was supposed to save. The gains are not just offset; they are erased, and then some.
And the irony does not stop at cost. Clients adopted these tools to reduce legal spend. What some are actually purchasing is legal risk. A purchase agreement term “simplified” by a chatbot and sent directly to the seller can become a binding concession. Confidential deal information pasted into a consumer platform raises questions about confidentiality and, potentially, privilege that would never have arisen had the client simply picked up the phone. An AI-drafted side letter can contradict the main agreement in ways that will be litigated long after anyone remembers who typed the prompt. The cheapest legal advice remains the advice that does not need to be undone.
None of this is an argument against AI. It is an argument against confusing fluency with judgment. The value of a lawyer was never the ability to generate words about a contract; it was the ability to know which words matter, why, and to whom. Consumer AI has made the words free. It has done nothing to make the judgment cheaper, and by burying that judgment under mountains of confident noise, it has made the judgment more expensive to exercise.
So here is a modest proposal for the cost-conscious client. By all means, use AI; your lawyers certainly are. But let them use the tools built for the job, on the documents as they actually exist, before the chatbot gets its hands on them. Ask your questions before you ask the machine. Otherwise, the next time you open your legal bill and wonder why it went up, you may want to scroll back through your chat history. The answer will be there, written in perfect, confident, entirely billable prose.
* The author acknowledges that this column was drafted with the assistance of an AI platform, a purpose-built legal one, naturally, and that a human lawyer then reviewed every word. That review was not billed to anyone.
