Context, Context, Context: The Three Golden Rules Every Lawyer Needs to Know

If there is one principle that separates a mediocre AI output from an excellent one, it is context. Not the tool you chose, not the subscription you purchased, and not the word count of your prompt. The context you provide to the model is what determines the quality of the answer you get back.

This was the subject of my talk to some two hundred in-house counsel at a conference celebrating three years of the AI tools that have transformed legal practice, held at the offices of Arnon, Tadmor, Levy. Here are the principles I presented.

Imagine you are consulting an expert. But this is no ordinary expert - this is a Renaissance professional who is simultaneously a lawyer, an accountant, a doctor, and an astronaut. Here is the catch: to give you the effective advice you need, he has to put on a special pair of glasses - a different pair for each profession.

If you ask him which painkiller to take for a headache, it probably does not matter much which glasses he puts on - or whether he puts any on at all. But if you consult him on a complex matter that demands real expertise, it makes an enormous difference.

This is precisely why giving the models we work with as much context as possible about their task matters so much. When we provide that context, they put on the right glasses and give us the most effective answer available.

As human beings, we have five senses that allow us to absorb context without anyone spelling it out for us. When we walk into a boardroom and sense that the atmosphere is extremely tense, we immediately understand that we need to shift into crisis mode. AI models do not work that way - the context of the task has to be communicated to them explicitly.

Of the Five Golden Rules I developed for prompt engineering for lawyers, three deal directly with context:

1. Define a persona. This is how we tell the model who we want it to become - exactly as in the Renaissance professional example above.

2. Provide maximum information and context about the task. This is where the broader background comes in: Are we in a negotiation where we hold little leverage? Is there a time constraint? Have we already exchanged drafts with the other side?

3. Improve the output through critique and examples. Ask the model to take on an even more senior persona than the first and use it to critique the initial output. The model enters a high-performance mode and consistently delivers a higher-quality, more effective result.

My thanks to the two hundred in-house counsel who came to celebrate, and to Arnon, Tadmor, Levy for the exceptional production.

The most common mistake I see with AI tools is the expectation that the model will somehow guess what we need. It cannot. But when you supply it with full context, it does not need to guess. Context is not a technical detail inside a prompt - it is the difference between a support tool and a genuine thinking partner.