Lesson 3Beginner

Same Model, With and Without Legal Sources

See what changes when GPT-5.6 answers directly versus when the same assistant can use legal source data.

You will know why a good-looking legal answer still needs a source path, and what source-backed research makes easier to check.

From safe use to source-backed use

The last lesson focused on what you can safely put into an AI assistant. This lesson focuses on what comes back out.

Even a new, highly capable model can give an answer that looks organized, confident, and useful. That is helpful, but it can also hide the work you still need to do: checking whether the cases exist, whether they say what the answer says, whether newer cases matter, and whether the sources fit your jurisdiction and posture.

So before you connect any legal research tool, it helps to see the difference between a polished answer and an answer with a source path you can inspect.

The sample question

The same public research question was asked two ways: first to GPT-5.6 in ChatGPT by itself, then to the same model with Descrybe Legal Engine connected.

The point is not that the model-only answer is useless. It is not. The point is that a legal answer can look polished and helpful while still leaving important verification work hidden from view.

Sample research question

I need cases about whether a public school can discipline a student for off-campus social media speech. Give me a short research starting point.

Two real answer excerpts

Both examples below came from GPT-5.6 in ChatGPT and the same public legal question. The excerpts are shortened so you can compare the research behavior without reading a full memo.

Model only

A useful, polished starting point

The answer starts with Mahanoy Area School District v. B.L., 594 U.S. 180 (2021), and explains that Mahanoy does not create an absolute bar against disciplining off-campus speech.

It identifies circumstances where school authority may remain stronger, including threats, bullying or harassment, school activity rules, and school security. It then lists cases worth pulling next: Tinker, Kowalski, Bell, D.J.M., and Doninger.

This is helpful orientation.
It gives likely leading cases and a usable research string.
But you still need to confirm the cases, source text, current treatment, circuit fit, and whether newer post-Mahanoy cases are missing.
Because the answer looks good, the missing work may not jump out at you.

Same model + Descrybe

Similar reasoning, with an inspection path

The answer again starts with Mahanoy, but it includes a Descrybe case link and builds a source-backed set of authorities.

It includes Kowalski, Layshock, Doninger, Chen, and C.G. v. Siegfried, including newer post-Mahanoy cases from the Ninth and Tenth Circuits. It also warns that Doninger predates Mahanoy and should be read cautiously.

The answer ends with a practical framework and notes that the federal First Amendment analysis is for public schools. Private-school authority may depend more heavily on contracts, handbooks, and state law.

The source links give you somewhere to inspect the cases instead of only trusting the summary.
The newer post-Mahanoy cases make the result more targeted.
It still needs human review, but the checking path is clearer.

Mahanoy Kowalski Layshock Doninger Chen C.G.

Subtle differences to notice

The model-only answer is not obviously bad. It is organized, plausible, and useful for orientation.
The source-backed answer does not replace legal judgment. It gives you source links and a clearer place to start checking.
The source-backed answer adds newer post-Mahanoy cases and warns that an older case should be read cautiously.
The source-backed answer names a scope limit: this is a federal First Amendment starting point for public schools, while private-school authority may turn on different sources.
The biggest risk with the model-only answer is not messy writing. It is a polished answer that may omit something important without announcing the omission.

The habit

Use the assistant directly when you want orientation, brainstorming, issue spotting, or a first draft of a research plan.

When you need legal authority you may rely on, move from a polished answer to source-backed research. The goal is not to make the assistant sound smarter. The goal is to make the answer easier to inspect.