AI systems are ingesting synthetic content at scale, the outputs are getting worse – and law firms should be worried. When AI Overviews or ChatGPT pull from content that was itself AI-generated and that content contains errors, those errors don’t just persist, they amplify. Each retrieval cycle compounds the problem.
Law firms are caught in this loop whether they chose to be or not. If your firm published AI-generated content with errors, that content is now part of the training and retrieval pool. When an AI system surfaces it to a prospective client, a judge, or opposing counsel, your firm’s name is in the citation. There is no algorithm to blame.
That is the ouroboros. The snake is already eating its tail, and most firms don’t know they’re holding it.
How Does a Law Firm’s Content End Up Feeding the Loop?
The loop works like this. A law firm needs content to rank. They hire an agency that uses AI to produce it at scale without lawyer review and errors slip through. The content gets published. AI search tools index it and cite it in their answers. Those answers become source material for the next round of AI-generated content.
Each pass introduces more error and less original thought, but the content keeps getting cited because it exists and it ranks.
SEO researcher Lily Ray named this process the AI Slop Loop in an April 2026 piece for Search Engine Journal. The mechanism she describes is the same one this article is about: AI systems trained on AI content, citing AI content, distributing AI content to users who treat it as fact.
The BBC ran a version of this test that is hard to dismiss. Journalist Thomas Germain published a fake article naming himself the best tech journalist at eating hot dogs. Within 24 hours, Google’s AI Overviews and the Gemini app were surfacing his fabrication as if it were fact. ChatGPT did the same. The point was not that AI is gullible. The point was that AI search tools do not distinguish between real sources and invented ones. They cite what exists.
For law firms, that is the problem in one sentence. Your content exists. It will be cited. The only question is whether what gets cited is accurate.
Why Can’t Law Firms Absorb This the Way Other Industries Can?
Your firm’s content is being cited by AI search tools. That sounds like a win. For most firms right now, it is not.
The citation is only as good as the content being cited. A lot of law firm content is currently being produced by AI tools, cited by AI search tools, and landing in front of potential clients who treat it as legal advice. Nobody in that chain has a law degree. That is not a disclosure. That is a gap.
A potential client does not read a law firm website the way they read a general interest article. They read it the way they would read advice from a lawyer (even with your boilerplate disclaimer at the end). When a recipe blog gets the timing wrong, you end up with dry chicken. When a law firm blog gets the statute of limitations wrong, someone misses their filing window. That is not a marketing failure – it’s a professional responsibility problem.
To put it more plainly, the AI tool that wrote the post does not have a bar number. The firm that published it does.
The fiction of “agency insulation” makes it worse. A lot of firms believe that because a marketing agency wrote the content, the liability stays with the agency. It does not. The content runs under the firm’s name, on the firm’s website, with the firm’s phone number in the footer. The agency’s name is nowhere on that page.
What Does Getting This Wrong Actually Look Like?
These are not edge cases. They are the predictable output of a process with no qualified reviewer in the loop.
The sanctions record makes the pattern concrete. In Mata v. Avianca, a federal judge sanctioned attorneys who submitted briefs citing cases that did not exist. ChatGPT had hallucinated them. Nobody checked. That was 2023. As this goes live, two lawyers were just disciplined in Mississippi for submitting briefs with AI hallucinations. The judge barred the two attorneys from appearing in the Northern District of Mississippi for two years and issued fines. Between those two points in time, a researcher at HEC Paris who tracks AI sanctions counts more than 1,200 cases worldwide, approximately 800 from U.S. courts. Penalties have crossed $100,000 in some cases.
All of those cases involve court filings, not law firm websites. That is worth noting, because it means the enforcement arc is still early. The applicable rules already cover website content. The cases have not caught up yet.
What Happens When No One Is Reviewing the Work
These situations are hypothetical. The underlying errors (wrong statutes, hallucinated cases, jurisdiction mismatches) are documented failure modes of AI content production.
The Statute of Limitations Error
A personal injury firm publishes a blog post about claims in their state. An AI content tool writes the post and gets the statute of limitations wrong by six months. A potential client reads it, calculates her deadline from the wrong number, and waits. By the time she calls, her window has closed.
The Hallucinated Precedent
A litigation firm publishes a post citing a case that supports an argument the firm makes in practice. The case does not exist. An AI tool hallucinated it. Opposing counsel finds it. The firm spends more time explaining that citation than it spent producing the post.
The Jurisdiction Mismatch
A firm publishes content written generically by a vendor serving clients in multiple states. The post describes a legal standard from the wrong state. A client relies on it, gets expectations the law never promised, and ends up unhappy with the outcome.
What Does Getting This Right Actually Require?
The answer is not to stop using AI in content production. AI has a place in the content creation process, and vendors who do not use it are doing their clients a disservice.
There is also a competitive argument for getting this right. The content attributes that make AI systems more likely to surface your firm – original analysis, first-hand experience, genuine expertise signals – are exactly what unreviewed AI content cannot provide.
A post written by an AI tool and published without legal review looks like every other post written by an AI tool and published without legal review.
AI citation systems are increasingly rewarding information gain: content that adds something new to the conversation rather than restating what already exists. Lawyer-reviewed content that reflects real practice experience, jurisdiction-specific knowledge, and defensible legal positions is not just safer to publish. It is more likely to be cited.
The question is not whether AI is in the process. The question is whether a pre-publication review by someone with a law degree is in the process too. Those are different questions with very different answers.
What Your Duty of Supervision Actually Requires
Model Rule 5.3 requires lawyers to make reasonable efforts to ensure that non-lawyer assistance is compatible with the professional obligations of the firm. That rule was written with paralegals and staff in mind. It applies with equal force to the AI tools and marketing vendors producing content under your name.
The duty has three practical implications for law firm content.
First, supervision cannot be retroactive. If your process is to review content after a client complains or after opposing counsel finds a bad citation, you are not supervising the work, you are responding to its consequences. Reasonable efforts means a qualified reviewer in the loop before publication, not after.
Second, delegation does not transfer responsibility. The agency relationship does not move your bar obligations to your vendor. When an AI tool hallucinates a statute of limitations and your firm’s name is on the post, the analysis under Rule 5.3 starts with you. What did you do to ensure the work was accurate? “We hired an agency” is not an answer to that question.
Third, the standard for marketing content is not editorial quality, it is accuracy. Rule 7.1 prohibits false or misleading communications about a lawyer’s services. That rule applies directly to your website, your blog, and anything else published under your firm’s name. It does not contain an exception for content produced by a vendor, and it does not require a client to be harmed before it is implicated. ABA Formal Opinion 512 reinforces the broader principle that accuracy obligations apply regardless of how content was generated, but Rule 7.1 is the provision that puts your marketing content squarely within the Rules.
The practical question is straightforward: before your firm’s content is published, is there a licensed attorney reading it for legal accuracy? Not for tone. Not for SEO. For accuracy. If you cannot answer that question with a name and a bar number, your supervision of that content does not meet the standard the Rules require.
Is There a Lawyer Evaluating Your Content Before It Goes Live?
Law firms that want to publish content responsibly in the AI search era need one thing most marketing agencies cannot provide: a licensed attorney in the content review loop. If your current vendor cannot name the JD who reads your posts before they go live, you are publishing on the honor system. Reach out if you want to discuss what a different standard looks like.
