Anyone can open ChatGPT. Typing a question into a text box has never been the hard part, and it isn't getting harder - if anything, the models keep getting more forgiving of a lazy prompt. What a text box has never been able to supply on its own is legal-market knowledge, a prompt actually worth writing, the editing pass that catches what's wrong, the fact-check that catches what's fabricated, and the judgment to know what shouldn't be published at all. That whole chain is the actual service. The prompt is just where it starts.
Start with the audit → Email RussellEvery firm has access to the same models now. That's exactly why the model was never the advantage.
Type "write a blog post about OUI defense in Massachusetts" into any AI system and you'll get a blog post about OUI defense in Massachusetts - competent, generic, and functionally identical to what every other firm's marketing person typed into the same box last week. The model did what it was asked. What it wasn't asked to do, because the operator didn't know to ask, was anything specific: which county, which courts, which actual defenses come up in practice, which claims are true and which ones are the model's confident guess. A tool this powerful in inexperienced hands doesn't fail loudly. It succeeds at producing something plausible, generic, and quietly wrong in ways a non-lawyer would never catch.
Everything downstream of the prompt is where the actual value gets made or lost.
Knowing that a clerk-magistrate hearing exists, what actually happens there, and why it's an underused niche isn't something a model volunteers. It's something you have to already know before you can even ask the right question.
What it means: the prompt is downstream of the expertise, not a substitute for it. Garbage domain knowledge in, garbage content out - the model just makes it faster and better-punctuated.
These models fabricate case citations with total confidence. They generalize state law that varies county to county. Left unedited, AI-generated legal content is a bar-complaint generator with excellent grammar. The edit and the fact-check are not optional finishing touches - they're the part of the job that was always the actual job.
What it means: if nobody with legal judgment reads the output before it goes live, the firm is publishing a confident stranger's guesswork under its own name.
The hardest skill to prompt for is knowing what not to publish - the claim that's technically defensible but reads as attorney advertising overreach, the guarantee an AI will cheerfully draft because it doesn't know Rule 7.1 exists. A model has no license to lose and no bar complaint to answer for.
What it means: judgment is the one input in this whole chain that has to come from a person who's actually accountable for the result.
"I Am a Prompt Star" sounds like someone bragging about how well they can type into a text box, and that's exactly the misunderstanding this whole page is arguing against. Writing a good prompt is a real skill, and I don't want to undersell it - but it's the smallest, easiest-to-learn part of the actual work. Anyone can watch a ten-minute video and write a better prompt by lunch. Nobody can watch a ten-minute video and acquire twenty years of knowing how Massachusetts district courts actually work, or which claims a firm can defend if bar counsel ever asks.
The prompt gets you a draft. Everything that turns a draft into something worth publishing under your firm's name still requires a person who knows the business - and that hasn't changed since the models got good. It's arguably gotten more true, not less.
I was teaching computer science the day ChatGPT came out, which means I've had the same amount of time with these tools as everyone else claiming AI fluency now - the difference isn't a head start on prompting. It's the twenty-one years of legal-market knowledge that came before the prompting even started. That combination - knowing the courts and knowing the tools - is the actual service. Either one without the other produces something that looks like content and isn't quite the truth.
There's a song about this too, bubblegum dance, which is a deeply silly genre for a point this straightforward: the prompt was never the hard part.
If your firm's AI strategy right now is "someone typed some prompts into ChatGPT," you have a draft, not a strategy. The audit tells you the difference - and what it would take to close it.
For criminal defense firms using AI, or thinking about it, who want the legal-market knowledge, the editing, the fact-checking, and the judgment built in - not just a faster way to publish a generic draft.
Start with the audit → Contact Russell Matson