For Economist (Private Sector / Consulting)s ·
What you'll accomplish
Every new chat starts from nothing, so you spend the first few minutes re-explaining the question, the audience and the ground rules. A ChatGPT Project gives each engagement its own home for chats, standing instructions and reference files. You open it, ask for the next plan draft or a public-source question, and the context is already there.
This project holds context, not data. It carries public background, a project plan written with placeholders and a running list of open questions. The engagement's real data stays where your firm keeps it.
What you'll need
Data caution. Client data, documents produced in litigation, anything under a protective order or NDA, and non-public company data stay in your firm's approved systems, so none of them goes into this project. It holds public background, placeholders and instructions. Even the project plan should use coded placeholders, because project names and dates can identify a matter. Only if your firm and the engagement (and counsel, on litigation matters) approve ChatGPT for more should you add more. Check ChatGPT's data controls for whether chats and uploaded files are used to improve models and how long they are kept. Check, too, who can see a project if your plan offers sharing, and leave sharing off for anything you would not email to a stranger.
plan-public-version.What you should see: A plan that reads clearly but names no client, matter or person. Troubleshooting: If the plan is hard to read without real names, use consistent codes (Company A, Company B) and keep the code key in your own firm-approved file, never in the project.
What you should see: A short folder of public files and a log. Troubleshooting: Leave out anything behind a license unless your license allows you to upload it to an AI tool.
[Matter code] planning, not the real matter name.What you should see: An empty project page with a place for chats, files and instructions. Troubleshooting: OpenAI's help pages describe projects as grouping chats and holding custom instructions and uploaded files. Where each option sits can change between versions, so look for those three things on the project page, and check OpenAI's own pages for which plans include projects.
Find the project instructions area and paste the block below.
This project supports an applied economist with planning and public-source questions on one engagement. The engagement is identified only by the code [Matter code]. All names are placeholders.
Rules for every chat:
- Never ask for client data, documents produced in litigation, protected material, or non-public figures. If the user pastes any, tell them to remove it.
- Never invent a data value, citation, paper, author or series ID. Any source you suggest is a lead for the user to check on the primary source, and you must say so.
- Never produce a figure the user will rely on. The user's own spreadsheet and code compute every number.
- Never write an expert opinion, a legal conclusion, or a statement of what a court, agency or rule requires. Say "ask counsel" or "ask the project lead".
- Never choose a model or an assumption.
- Label every plan or draft "Draft for project lead review".
Uses: plan drafts, questions about the public files in this project, and a running list of open questions.
Running list: when the user types "add open question: ...", add it to a numbered list headed "Open questions". When the user types "show open questions", print the whole list. Do not remove an item unless the user says to.
What you should see: The instructions saved on the project page. Troubleshooting: The running list lives in the chat, so copy it into your own notes at the end of each session.
plan-public-version.What you should see: The files listed on the project page. Troubleshooting: If an upload fails, check the file type and size against OpenAI's file upload pages.
What you should see: A plan draft that restates your workstreams in the project's voice and ends with "Draft for project lead review". Troubleshooting: If the reply names a real organization, a statistic or a paper that is not in your files, treat it as unverified. Delete it, or look it up yourself on the primary source.
plan-public-version. Every workstream and owner should be in your plan.What you should see: A real plan changed only by things you checked. Troubleshooting: If the chat gave a source you cannot find, drop it.
show open questions.What you should see: A numbered list in your notes that matches what you asked. Troubleshooting: If the list looks shorter than you remember, check your earlier chats in the project.
Public-source question
Which of the public files in this project say anything about [topic]? Quote the relevant passage and name the file. If none do, write "not in the project files". Do not use anything from outside the files.
Add an open question
add open question: [your question, in placeholder terms]
Next-step list
Based only on the plan file, list the workstreams that have no owner or no review point. Write "none" if there are none.
Leads for a public source search
Suggest types of public source (for example, an agency release, a methodology note, an industry report) that might cover [topic]. Do not name a specific report, author, series ID or number. I will find and log the real sources myself.