An adviser drafts a market commentary with a chatbot on a Friday afternoon, likes it, and posts it. Nothing about that copy passed a review, and the firm's records show no version of it before publication. The content itself may be fine. The process is the exposure.

The rules did not change because drafting got faster. What changed is how easy it became to publish something nobody checked.

What is actually safe to hand to a machine?

Work that is not a claim.

First drafts of educational content. Reformatting a long piece into shorter ones. Summarizing a public document you have already read. Rewriting a paragraph you wrote at the reading level your clients use. Coming up with subject lines.

In each of these, a person still decides what is true, what is included, and what goes out. The machine is doing production work, not making a representation to the public.

Where does the wire sit?

At anything that reads as a claim about performance, outcomes, or your firm's standing.

These systems are fluent, and fluent copy drifts toward the promotional. Ask for a paragraph about your process and you may get "consistently strong results" back, because that is how marketing copy generally sounds. If that sentence reaches your site, the fact that a machine produced it is not a defense.

Two other places to watch. Anything resembling a testimonial or endorsement, which sits under rules with specific disclosure requirements. And any statistic the model supplies. Do not publish a number these systems produce unless you have found it yourself at the source. In my experience that is the single most common way trouble enters, because the number always looks reasonable.

What does the review step look like in practice?

One person, named, reads everything before it publishes, and looks for three things: claims that promise or imply results, numbers without a source you have verified, and anything that could read as a client endorsement.

Keep the record. Whatever your firm already does to retain marketing materials should cover this too, including the version that went out and evidence of who approved it. If your process depends on remembering, it will fail on a busy week.

The specifics of what your firm must retain, and how the SEC Marketing Rule applies to a given piece, are matters for your compliance officer. Bring them in before you build the workflow rather than after something is live. What follows here is a description of how firms structure the work, not legal or compliance advice.

Do you need to disclose that AI was involved?

There is no general requirement that every AI-assisted sentence carry a label, and your obligations depend on your firm and your regulator, so ask your counsel rather than a blog.

Setting the requirement aside, there is a judgment question underneath it. If a client learned how a piece was produced, would they feel misled? For educational content that you wrote the substance of and reviewed, almost certainly not. For anything presented as your personal view or your individual analysis, a machine-written draft published unedited is a different matter, and the honest move is to write that one yourself.

Some firms keep a short note on the site describing how they use these tools. It answers the question calmly for anyone who wonders, which is better than leaving them to guess. We take the same posture in our own work, described under Clarity AI and in the competencies on the AI fluency page.

Used with a review step and a record, these tools remove real drudgery from a small firm's marketing. Used without one, they let you publish faster than you can check. The difference is one named person and one habit.

If you want to see how your current site reads before you add anything to it, The Second Look is free and takes about four minutes.