How “Context, Task, and Format” Prompting Changes the Game - Making ChatGPT-5 Work for Law.

A Plain-Spoken Guide to Reliable Legal AI for Lawyers and DUI Defense

Every lawyer wants to know the secret to getting accurate, safe, and useful results from AI. With ChatGPT-5 now available, it’s more tempting than ever for lawyers to trust the tool with real work, from drafting motions and reviewing evidence, to brainstorming cross-examination and sharpening trial narratives.

But with AI, if your prompt is as oblique as “just write it up” you get about the same quality output as you might expect from from your new junior associate. Incomplete instructions lead to incomplete output, and wasted time fixing what should have been right from the start.

For all the power of ChatGPT-5, it still needs clear, well-structured instructions. Over the last year, I’ve tested dozens of approaches, and the pattern that works, time and again, boils down to three labeled sections: <context>, <task>, and <format>.

While it may look like it, this isn’t computer code. It’s not XML or anything close. What is is though is something more important, a method to clarify your own thinking while simultaneously making your intentions “easier” for the AI to “understand.” Let me show you how it works, why it works, and how it can save you hours and protect your clients.

---

Why Labeling Matters - And Why This Isn’t Coding

When you see angle brackets like <context>, <task>, or <format>, you might think you’re looking at some programming language. It’s not. You could just as easily write “CONTEXT:” or “FACTS:” in all caps. The purpose is about forcing yourself to separate the three things every good prompt needs:

• The facts and background the AI needs to know (context)

• The job you want done (task)

• The structure or style you want the answer in (format)

This mirrors how you might train an associate. You wouldn’t send them off with “go file a motion.” You’d brief them, about the facts of the case, the jurisdiction, the Judge’s predilections, opposing counsel’s strategy. With that context, you establish the task - now find the law, and write xyz for this purpose. The labeled prompt just does all that for AI, while clarifying your needs and your intended use.

---

The Power of <context>: Telling the AI What Really Matters

Context is everything the model needs to “get” your situation. If you leave it out, the AI fills the gaps - with guesses, generalizations, and sometimes total fabrications. This lack of context is often a big contributor to what leads to hallucinated output.

Example from DUI Defense:

<context> On April 1, 2025, in Oakland Township, Michigan, police responded to a single-vehicle crash. The driver, “Marshal Mathers,” denied drinking, claimed he was distracted by his phone, and showed no signs of intoxication other than confusion at the scene. No open containers or odor of alcohol. The warrant affidavit left out several key facts from the crash report and focused only on the driver’s confusion and stumbling. <context>

Don’t be afraid to get detailed. The clearer your context, the less room for hallucination. In doing so however, a word of caution - be sure to use initials, a fictitious name, and few enough details that it would be impossible for the AI to “know” who you’re talking about.

You may want to be cautious even about location (Oakland Township) or date, unless these too are fictitious and added to aid in the confidentially. But in this case it’s take a step further because, as you may know, Slim Shady once lived in Oakland Township Michigan, so it’s actually a useful rouse.

---

The Power of <task>: Be Clear About What You Want

The “task” is the ask. You’re not just telling the AI to write something, you’re telling it what to do, what question to answer, what argument to focus on.