How LexAlert keeps AI-drafted replies compliant with state bar rules
Every attorney who hears "AI drafts your outreach replies" has the same reaction, and it's the correct one: the risk isn't an awkward sentence, it's a bar complaint. Lawyer advertising and solicitation are governed by rules that vary by state, carry real penalties, and were written long before anyone was generating replies with a language model.
So the design question we started from wasn't "how do we make the AI write well." It was "how do we make it structurally hard for a non-compliant reply to ever reach a prospect." The answer is that the parts of the system that enforce the rules contain no AI at all.
The rules don't run on a model
An AI model is the wrong tool for a compliance rule. Models are probabilistic — given the same input twice they can give you different output, and they can't be audited line by line. A bar rule needs the opposite: deterministic, identical every time, and explainable to an investigator.
So in our pipeline the compliance stage is pure code, never a model. The language model drafts the reply. A separate, deterministic check then decides — pass or fail — whether that draft is allowed to reach your queue. The model never gets a vote on compliance, and the check produces the same verdict on the same input every single time.
The model's job is to sound human and helpful. The compliance check's job is to be boringly deterministic and auditable. Keeping those jobs in separate components is the whole trick.
What the compliance check enforces
Solicitation windows
Several jurisdictions restrict how soon after an incident a lawyer may solicit a prospective client, and the appropriate window can also depend on the channel. The check enforces the configured window for your jurisdiction and platform on every draft. A reply that would post too soon doesn't get held for human judgment — it never reaches the queue in the first place.
Prohibited phrases
Certain language is a bright line in attorney advertising. Phrases like "guaranteed," "we will win," "you are entitled to," and "no-risk" are auto-rejected — first during draft generation, then again at the compliance check, so a banned phrase has to clear two independent gates that are both designed to stop it. You can add your firm's own prohibited terms to the list.
Length and format limits
Every draft is hard-capped at 200 words. Long, salesy replies read as marketing and invite scrutiny; a tight, genuinely useful reply reads as a professional being helpful. The cap is enforced mechanically, not left to the model's discretion.
Per-platform daily caps
A real attorney participates in a handful of conversations a day. A firehose of replies looks like a bot and gets accounts flagged. Configurable per-platform daily caps keep your activity in the range of a real professional, and the check enforces them across every channel.
An attorney clicks every button
This is the line we don't move: nothing auto-posts by default. Every draft that clears the compliance check lands in a review queue where an attorney chooses to Approve, Edit, Regenerate, Decline, or Escalate. The software's job ends at "here is a compliant draft." A licensed human decides whether it goes out.
That's not a limitation we're apologizing for — it's the only model an attorney should accept. The reply posts from your account, tied to your bar profile and your reputation, so the person whose license is on the line is the person who approves the words.
If the bar ever asks
Compliance you can't prove isn't worth much. Every action in the system is written to a tamper-proof audit log: every draft generated, every edit and the diff, every approval, every decline — each stamped with the attorney, the timestamp, the platform, and the post ID.
If your malpractice carrier wants documentation, or a bar investigator asks what you posted and when, you export the record as a CSV and hand it over. The honest version of "we're compliant" is "here is the complete, immutable history of everything we did, on the record."
Rules change, states differ, and edge cases are real. We're not claiming a turnkey guarantee for every jurisdiction — we're claiming an architecture where the rules run as deterministic code, an attorney approves every reply, and everything is logged. That's the posture a regulated profession should demand.
The deeper reason this matters: the highest-intent legal leads live in public conversations on Reddit, Quora, and Avvo, and most firms avoid that channel precisely because of the compliance exposure. Solve the compliance problem properly and you unlock the best lead source in legal marketing. That's the bet LexAlert is built on.
Book a demo and we'll walk through the real posts in your practice area and state, what those leads cost, and your modeled cost per signed client.
Book a demo