Legal AI; when the answer must be grounded in statute

General chatbots like ChatGPT produce fluent text on any topic; but their knowledge comes from training memory, not statute text. Real legal AI is different: it is connected to a live legal corpus, searches it for every question, and builds the answer from the articles themselves.

For an Iranian user the difference is practical: Iranian statutes are amended constantly, and the right answer only comes from the current text of those statutes.

The phenomenon called hallucination takes a dangerous form in law: the model confidently cites an article that either does not exist or says something else. The reader has no way to tell — the text looks fluent and assured.

In the real world, relying on a fabricated citation means a lost deadline, a defective petition, or professional embarrassment. So the first criterion for choosing a legal AI must be: how are the citations guaranteed?

In Dadyar, a citation is not a claim; it is a process. Every answer is built from a live search across thousands of statutes, tens of thousands of articles, unification rulings and advisory opinions. Then every citation in the text is automatically matched against the real article before display — an unverified citation is never shown.

Two more guarantees: the research trail is visible (you see which statutes were searched), and when no clear legal basis exists, Dadyar honestly says no legal basis was found instead of inventing one. Honesty is part of the architecture.