Every week a new quick guide appears on how to use ChatGPT for lawyers. Most are marketing shortcuts dressed up as pedagogy. Directing them at legal professionals starting from zero, without giving them the critical foundations, is not just a disservice: it is a failure toward the profession and toward adult learning.
Before committing time or budget to AI training for lawyers, it is worth learning to read critically what is on offer. These are the questions that separate a rigorous guide from material that will generate more risk than it mitigates.
The problem is not the tool: it is what is being taught about it
The classic argument in these guides never changes: “use this prompt, save hours, be more efficient.” It sounds reasonable. But they almost always cultivate a false sense of confidence in the model’s “capabilities,” implying it can replace complex legal analysis or automate sensitive tasks like contract drafting or case analysis.
That is a fundamental mistake. ChatGPT generates text based on statistical patterns. It does not understand legal contexts, does not rank sources by authority, does not distinguish an obiter dictum from a ratio decidendi, and its knowledge is time-bounded, potentially stale for applicable positive law. It also does not “know” when it does not know.
When a guide sells the opposite for the sake of pedagogical simplicity, the predictable result is a lawyer who signs off on a filing riddled with fabricated case law or arguments that would not survive serious scrutiny.
Which techniques serious training must teach
The second gap is technical. Almost no guide mentions the practices that genuinely reduce the risks of using AI in law:
- Rigorous prompt construction (context, role, explicit constraints, verifiable output formats).
- Methods to systematically verify generated information before incorporating it into a filing.
- Task decomposition strategies, which avoid asking the model for complex conclusions in a single query.
- A basic understanding of model behavior: hallucination, confirmation bias, contextual drift.
Telling ChatGPT “act as the best lawyer in the country” guarantees nothing. It is a semantically empty instruction. The system remains the same, with the same limitations, producing plausible text. What changes, at best, is the stylistic register of the response.
Mentioning limitations is not enough
Some guides do include a closing paragraph on “limitations.” But they treat it almost as a legal disclaimer, without the seriousness the subject demands and without offering any concrete mitigation strategies. The reader retains everything else: the copy-paste prompts, the productivity promises, the sense of having “integrated AI” into their practice.
In the legal domain, where a single error can cost a case, a disciplinary sanction or a malpractice claim, that imbalance between enthusiasm and warning is dangerous.
What to demand from training before paying for it
Rather than strengthening lawyers’ practice, this type of content can lead them to commit serious errors under the guise of modernization. What the field needs are more rigorous training resources for introducing AI into legal practice: programs that combine technical understanding of the model, legal judgment and systematic verification.
That is precisely the logic behind the training and workshops I offer to legal teams, and the reason I always start with the questions no quick guide ever asks: what task are we automating, what risk does the client tolerate, and how do we verify what the model produces before signing off?