This page lists three turns, the messages you send to the AI one after another in the same chat, and then a check of the answer. Keep it open next to your chat. Each turn shows the next thing to paste, with the values you entered already filled in. Nothing is sent from this page, and your conversation stays in the AI tool you are using.
The development itself. Quote the operative language from the order or notice if you have it.
The real-world consequences, with the numbers you are willing to stand behind.
What you are doing, by when, and the single decision you need from them.
Name, role, and anything about their situation that changes how this lands.
Sets the vocabulary and how much procedure you can assume.
How this gets delivered.
Watch for these throughout
Candid assessments become exhibits. A verdict range in an email can surface in a coverage dispute, a fee fight, or a malpractice claim. Put in writing only numbers you would defend under oath.
Keep settlement positions and authority requests out of the same message as the news where you can. They travel differently, and the ask contaminates the disclosure.
Mark the communication privileged and check the recipient list. A business stakeholder copied on legal advice can break privilege over the whole thread.
Do not speculate in writing about appellate outcomes, insurance coverage, or how the matter was handled before you had it.
The model will smooth the news if you let it. Read the first sentence by itself and ask whether a client who read only that sentence would understand what happened.