For Corporate Attorneys ·
What you'll accomplish
Monday morning, your inbox holds a vendor NDA, a question about who can sign a renewal, a customer asking for a data security addendum, and one line from HR that deserves your attention before lunch. You will set up a ChatGPT Project that sorts each pasted request into Self-serve, Quick answer, Contract review or Escalate, gives a one-line reason, and suggests the first question to ask the requester. You make the final call on every category. The assistant gets you to a sorted queue faster.
This guide uses a Project.
What you'll need
A caution before you paste. Requests from the business often contain employee names, customer details, and sometimes the start of an HR complaint. Check your account's data settings first: whether chats are used for training, how long they are kept, and who at your company can see them. A free consumer plan may train on chats unless a setting is off. Replace names with roles ("Sales Lead", "Employee A") and delete personal data, account numbers and deal values before pasting. Do not paste anything privileged or market-sensitive unless your company has cleared the tool for it. Whether AI use affects privilege is unsettled, so follow your company's policy and your state bar's guidance.
OpenAI's help pages could not be read from our research machine, so the menu wording below is deliberately general. Look for the feature by name and expect small differences.
The assistant sorts the way you tell it to. Write your own tests first.
What you should see: A half-page of definitions in your words. Troubleshooting: If you cannot define a category in two sentences, the assistant will not sort it consistently either. Simplify first.
What you should see: Two dated files with headings. Troubleshooting: If a file is a scanned image, retype or export it as text first.
What you should see: A project page with a place for chats, files and instructions. Per OpenAI's Academy page, a Project keeps chats, files, instructions and context in one place, and the project instructions apply to every chat in the project. Troubleshooting: If you cannot find Projects, your plan or workspace may not include it. Ask your ChatGPT administrator.
You help the legal department of [Company] triage incoming requests. The lawyer decides the final category.
For each request I paste, return:
Category: one of Self-serve, Quick answer, Contract review, Escalate
Reason: one line
First question to ask the requester: one question
Source: the FAQ question or signing authority row you relied on, or "none"
Definitions:
Self-serve: the answer is in the FAQ or a template.
Quick answer: a short reply from the lawyer is enough.
Contract review: a document needs a real review.
Escalate: [paste your list of Escalate triggers].
Rules:
- If the request touches any Escalate trigger, choose Escalate even if the rest looks routine.
- If you are unsure between two categories, choose the more cautious one and say why.
- For signing authority questions, quote the row from the sheet. If the sheet does not cover it, say "the sheet does not cover this" and choose Escalate or Quick answer.
- Never state what the law requires. Never cite cases or statutes.
- Requests arrive with names replaced by roles. Do not ask for names.
What you should see: Both files listed in the project and the instructions saved. Troubleshooting: If the assistant ignores the format in a chat, check that the chat was started inside the project and not outside it.
What you should see: Matching categories on obvious cases and a cautious category on borderline ones. Troubleshooting: If it under-escalates, add the missed pattern to the Escalate list and run the test again.
Batch sort:
Triage each of the numbered requests below. Return a table with request number, category, reason and first question. Then list the Escalate items first.
Draft the reply for a Quick answer:
This was categorised as Quick answer. Draft a three-sentence reply to the requester in plain language. Do not state what the law requires. Flag anything I must confirm before sending.
Draft the self-serve pointer:
This is a Self-serve request. Write a short reply that points to the FAQ answer and tells the requester when to come back to Legal.
Find FAQ gaps:
From the requests in this chat, list the ones categorised Quick answer that came up more than once. These are candidates for a new FAQ entry.