Projects: organise recurring work

0:00 Projects in ChatGPT

If you work on the same client, product or report week after week, you should not be re explaining it every time. ChatGPT Projects give each workstream its own home with shared files, instructions and memory. In this lecture you will learn how to set one up properly, when to use a Project instead of a custom GPT, and how to keep it accurate.

0:28 Why Projects matter

Why do Projects matter? Because without them, your work on a client or initiative is scattered across dozens of unrelated chats, and every new chat starts from zero. A Project is one home for that work, with the files, the rules and the history in one place. And with project only memory, what ChatGPT learns about that client stays with that client. Think of it as a labelled filing drawer instead of papers spread across your desk.

1:01 What a Project is

A Project groups related chats with shared files and project instructions. Projects are available on free and paid plans, with limits that vary. Each Project can use default memory or project only memory. With project only memory, what ChatGPT learns inside the Project stays there, and chats can reference each other without dragging in unrelated conversations. On business plans you can share Projects with colleagues.

1:29 Project, GPT or chat?

Choose the right container. A one off question belongs in a normal chat. Your ongoing work on a client or initiative, with evolving files and chats, belongs in a Project. A reusable assistant with fixed instructions that colleagues can open and use is a custom GPT, covered in module five. And repeatable procedures or bundled workflows belong in plugins and skills, which we cover in module six.

1:58 Setup step 1 and 2

Setting up well takes five steps. One Project per client or initiative, because mixing clients mixes rules and facts. Upload current, clearly named files, like twenty twenty six September price list approved, and remove superseded versions. If both old and new price lists sit in the Project, ChatGPT has to choose between conflicting sources, and that is where wrong prices come from.

2:25 Setup step 3

Step three, focused instructions. For a premium dates brand selling in the UAE and UK, that might be: use product names and prices exactly as in the approved product sheet, British English for the UK, English and Arabic for the UAE, disclosure labels on paid partnerships, never invent prices, reviews, awards or statistics, write data needed instead, and end every deliverable with a checks needed list.

2:54 Setup steps 4 and 5

Step four, choose project only memory for client work. Step five, test with a trap question the files cannot answer, such as the price of a product that is not listed. A good Project says it is not in the files. If it invents a number, tighten the instructions and check for conflicting documents.

3:17 Working inside a Project

Inside the Project, keep one chat per deliverable. Ask ChatGPT to use earlier Project chats, for example, using the decisions from the kickoff chat, draft the brief. Save important approved outputs back into the Project as files so future chats use them. And refresh files monthly, because Project knowledge is a snapshot, not a live database.

3:41 Simple example

Here is a simple, personal example. You are renovating a kitchen. Create a Project and upload three builders' quotes, your measurements and a budget sheet. In one chat, ask ChatGPT to compare the quotes line by line and flag anything missing, like one quote leaving out waste removal. In another chat, ask for a task timeline based on the chosen quote. Both chats draw on the same files, and neither needs you to re upload anything. The business version works exactly the same way, just with clients instead of cupboards.

4:20 Worked example: a Jeddah fragrance launch

A marketing lead at a fragrance brand in Jeddah creates a Project called winter launch. Files include the approved product sheet, the brand voice guide, a dated summary of the regulator's disclosure guidance, and last year's results. Instructions set bilingual rules and banned claims. Chats cover messaging, the influencer brief, the email sequence and the FAQ. When pricing changes mid campaign, she replaces the product sheet and notes the change date in the instructions, and every later chat uses the new price. Midway through the campaign, a new team member joins. Instead of a long handover call, the lead shares the Project. The newcomer reads the instructions, sees the current product sheet and the decisions in the kickoff chat, and drafts her first Eid post that afternoon with the right prices and the right disclosure label. The Project became the handover document without anyone writing one.

5:23 Monthly maintenance

Give every important Project fifteen minutes a month. Remove superseded files and add the newest approved ones with dated names. Add one instruction for each recurring error you logged, and delete rules you no longer need. Archive finished chats and rename active ones clearly. Re run the trap question. And on shared Projects, confirm that only the right people have access. Agree a naming convention too, such as client, workstream and year, so anyone can find the right Project in seconds.

5:58 Try this now

Try this now. Create a Project for one real workstream you return to every week. Paste in focused instructions based on the template, with exact names, disclosure rules and a never invent rule. Upload three current, clearly named files, and choose project only memory if it is client work. Run two real tasks. Then ask a trap question the files cannot answer. If ChatGPT says it is not in the files, you are ready. If it invents something, fix the instructions and look for conflicting files.

6:35 Watch me do it, part 1

Let me set up the dates brand Project. I create a new Project and name it using our convention, Al Waha Dates, social, twenty twenty six. I paste the instructions. Use product names and prices exactly as in the approved product sheet, British English for the UK, English and Arabic for the UAE, disclosure labels on paid partnerships, never invent prices, reviews or awards, write data needed, and end with a checks needed list. I upload three files with dated names, the product sheet, the disclosure rules and the brand voice guide. Then I choose project only memory, because this is client work.

7:20 Watch me do it, part 2

First task, in a chat called Eid campaign captions, five bilingual captions for the premium gift box. The prices match the sheet, and the paid partnership label is included. Now the trap. What is the price of the two kilogram family box? That product is not listed, and ChatGPT says it is not in the project files and writes data needed. Next, a second chat. Using the decisions from the kickoff chat, draft the influencer brief. It pulls the budget range and the three content pillars we agreed earlier, without me pasting anything. The Project is working as a single home for this client.

8:05 Recap and next step

Recap. Use a Project for every recurring workstream, with clean, current files and focused instructions. Choose project only memory for client work, run a trap question before you trust it, and refresh monthly. On business plans, share carefully and name an owner. Your next step: set up one Project for real recurring work, run two tasks and a trap question, and refine the instructions.