Can a law firm use AI to write its marketing content?

Roughly half the states have now issued their own guidance on lawyers’ use of generative AI. None of it says do not use the tools.

The real question is the one behind it. If AI drafts the content and it invents a case result or overstates a credential, whose problem is that? It is the firm’s, always, regardless of who or what wrote the draft. Which is exactly why the process matters more than the tool does.

This is a marketing operations article, not legal advice. Advertising rules vary by state and generative AI guidance is being reissued constantly. Confirm anything here against your own bar’s current rules before you rely on it.

What the guidance actually requires

California, Florida and New York have gone furthest with detailed ethical direction on generative AI, and what runs through all of it is a verification requirement rather than a prohibition. A competent human has to check the output before it goes out.

ABA Formal Opinion 512 adds a consultation duty on top: lawyers must reasonably consult with clients about the means used to accomplish their objectives. That matters more for client work than for marketing copy, but it sets the tone for how these tools are expected to be handled.

The failures that have produced discipline were not “a lawyer used AI.” They were “a lawyer published something AI produced without checking it.” A California attorney was recommended for professional probation after briefs were filed containing AI-generated fake quotations. That is a verification failure, and it would have been just as sanctionable if a paralegal had invented the quotes.

The two things that actually create exposure

Fabricated specifics. Case results, verdict amounts, settlement figures, credentials, awards, years of experience, bar admissions. These are exactly the details a language model will produce fluently and confidently when it does not know them. They are also exactly the details a regulator will ask you to substantiate.

Unsubstantiated claims. Claims of expertise, client results, testimonials, awards, and review authenticity all require underlying documentation if a regulator or the FTC challenges them. “Best personal injury lawyer in the state” is a problem whether a human or a model wrote it. The tool did not create that risk, it just made it faster to produce.

Notice that neither of these is unique to AI. AI raises the volume, which raises the odds that something unverified slips through. The control is not avoiding the tool. It is making the verification step structural rather than optional.

The process that works

Five steps. The important part is that the gate is a step, not a habit.

1. Facts come from the firm, never from the model. Case results, attorney bios, bar admissions, awards, and any number go into the draft from a firm-supplied source document. The model writes around verified facts; it is never the source of one. This single rule removes most of the risk.

2. No superlatives or outcome promises in the template. “Best,” “top-rated,” “we win,” and any implied guarantee stay out at the drafting stage rather than getting caught at review. It is easier to never generate them.

3. Attorney review of the finished piece, not the outline. Approving a topic is not approving the copy. The review has to happen on the words that will publish, by someone at the firm who can be accountable for them, and it has to be a required step in the workflow rather than a courtesy email.

4. Substantiation stored with the piece. If it claims a result, the file supporting that result is attached and retrievable. If the claim is challenged eighteen months later, you want the documentation next to the content, not in somebody’s inbox.

5. A work log. What published, when, and who approved it. This is unglamorous and it is what makes a compliance question answerable in ten minutes instead of a week.

What this costs you in speed

Less than firms expect, because the review is on finished pieces rather than scattered across a process. A batch of practice-area pages reviewed together takes an attorney an hour or two, and the drafting that fed it took a fraction of the time it used to.

What it does cost is the ability to auto-publish. Nothing goes live untouched. In a regulated advertising category that is not a limitation to engineer around, it is the point.

The question to ask a vendor

“Walk me through what happens between a draft existing and it being live on my site.”

If the answer includes a named person at the firm reviewing the finished copy, and a place where substantiation is kept, they have thought about this. If the answer is that their writers are experienced and careful, they have not built a control, they have expressed a hope. Experienced and careful is how the fake-quote cases happened.