Every tool I have built on this site does one of three things: drafts a document from structured input, extracts information from a long text, or summarizes something that would take too long to read manually.
None of them write legal arguments. That is not an accident.
There is a version of legal AI hype that says the technology can replace attorney judgment. It cannot. And attorneys who use it as if it can will produce worse work than attorneys who do not use AI at all. The line between what AI does well and what it gets wrong is specific enough to be useful.
What AI is good at
Drafting from structured input. If you give a model a client name, matter type, fee arrangement, and scope of representation, it will produce a coherent engagement letter. The structure is consistent and the language is professional. What it cannot do is know whether those terms reflect what you actually agreed to with the client.
Extracting from long documents. A 47-page deposition transcript contains key testimony, admissions, and contradictions. A model can find most of them faster than a paralegal reading front to back. What it cannot do is tell you which admission changes the case and which one is noise.
Summarizing and organizing. Take a set of medical records and ask a model to produce a timeline of treatment. It will do that accurately and quickly. What it cannot do is recognize that the gap between the injury date and the first treatment visit undermines the causation argument.
The pattern is consistent: AI handles the mechanical work of reading, organizing, and drafting. The judgment about what matters and why belongs to the attorney.
What AI gets wrong
Legal arguments
An AI model will write a legal argument if you ask it to. It will cite cases, use proper structure, and sound authoritative. The problem is that it will also cite cases that do not exist, misstate holdings, and miss the controlling authority in your jurisdiction.
Example: You ask the model to draft a motion to dismiss arguing lack of personal jurisdiction. It produces a clean four-page brief citing three cases. Two of the cases are real. One is fabricated. The holding on the second real case is misstated. The brief reads well enough that you do not catch the errors before filing.
This is the hallucination problem in its most dangerous form. For factual documents like demand letters, a wrong date is a credibility problem. For legal arguments, a fabricated citation is a professional responsibility problem.
Strategic advice
A model does not know your client, your judge, your opposing counsel, or the settlement history in your jurisdiction. It cannot weigh those factors. It will give you something that sounds like strategy but is really a generic answer to a generic question.
Example: You have a PI case with clear liability but soft tissue injuries and a gap in treatment. You ask the model whether to demand policy limits or open low. It tells you to "consider the strength of liability, the nature of injuries, and the insurer's settlement history." That is not advice. That is a list of things to consider, which you already knew.
Judgment calls on ambiguous facts
Legal work is full of situations where the facts support more than one interpretation and you have to choose. AI will present both sides with apparent balance and leave you exactly where you started.
Example: Your client says the other driver ran a red light. There are no witnesses. The police report is ambiguous. You ask the model to assess liability. It says liability is "disputed" and lists arguments for both sides. That is accurate. It is also useless. The judgment about whether to take the case, how to value it, and how to position the demand is yours.
The useful version of AI in legal work
The tools that actually save time are the ones that handle the low-judgment work so attorneys can spend more time on the high-judgment work.
Running an intake call and producing a structured case summary is low-judgment work. Reading a deposition and flagging the admissions is low-judgment work. Drafting a fee agreement from a form is low-judgment work.
Deciding whether to settle, how to argue causation, or whether to take a case to trial — that is not. No tool on this site touches those decisions and that is intentional.
The attorneys who get the most out of AI are the ones who use it to clear the administrative work so they can focus on the parts of the practice that actually require a lawyer.