When should you use a project?

Use a project when work will continue over time or several outputs will depend on the same materials. Use a standalone conversation for a self-contained request. That distinction follows OpenAI's guidance on choosing a project.

Context means the background needed to answer your request. For your shop, write down the intended customer, the product facts and the purpose of the work. Treat that short brief as your starting point.

Pick a boundary you can explain

Hypothetical example: you sell handmade home accessories and need descriptions for a new collection. Call the project “New collection product pages.” Keep supplier negotiations and recruitment outside this first exercise.

Before creating anything, complete this sentence: “I will use this project to produce ___ from ___.” If you cannot name the output and its source, narrow the task. For client work, you can adapt a starting request from these ChatGPT prompts for freelancers.

How do you add files and check the material?

Open the project and use its Sources section for uploaded files or connected sources. Start a new conversation inside that project to work with its shared files and instructions. This structure is described in the official documentation.

A regular ChatGPT project does not directly access a folder on your computer. You need to upload or connect the material you want it to use, according to OpenAI's explanation. After editing a document on your computer, check which version is available in the project.

Begin with a small source pack

1. Select the approved facts. For the shop example, prepare product names, dimensions, materials and care instructions.
2. Label the version clearly. Use a descriptive filename rather than “final final.”
3. Remove irrelevant information. Leave customer addresses out of a product-writing exercise.
4. Add the file. Check that it appears among the project's sources.
5. Ask a factual question first. Request the dimensions of one product and compare the answer with your original.

Try this request: “List the products described in the uploaded catalogue. For each product, show the stated material and dimensions. Identify the source file and mark any missing information.” Check this answer before asking for polished descriptions.

If two documents disagree, decide which one is authoritative before continuing. My recommendation is to remove obsolete material from the working set and keep a clear approved version. Do not ask the assistant to guess which supplier sheet you meant to use.

What belongs in project instructions?

Put recurring rules in project instructions: the audience, writing style and how to handle missing facts. Project instructions apply across the project's chats, as OpenAI documents. Keep the particular output you want today in the individual conversation.

For example, “Write for first-time buyers” is a continuing rule. “Draft the description for the blue bowl” is a specific request. Make that distinction when preparing your own notes.

A practical instruction template

This is a hypothetical template for the shop example. Replace the bracketed parts with your own information before using it.

We are preparing product pages for [shop name]. Our customers are [describe the buyer]. Write in clear English, using short sentences and specific product details.

Use the approved catalogue for materials, dimensions and care instructions. Do not invent certifications, delivery promises or performance claims. If a necessary detail is absent, list it under Questions for the owner.

When asked for a product description, provide a short introduction, a factual bullet list and a care section. Keep suggested wording separate from information that still needs approval.

Test the instructions with an incomplete product entry. Ask for the description, then check whether the missing material is flagged. If it is guessed instead, revise the rule and repeat the same request.

For a broader discussion of writing preferences, read the guide to ChatGPT custom instructions. In your own setup notes, distinguish preferences you want everywhere from rules belonging to this particular project.

How should you organize chats inside the project?

Create a separate conversation for each distinct output. Keep related work together at the project level and give each chat a descriptive name. OpenAI recommends this approach to keep individual conversations focused while preserving shared project context.

A hypothetical collection launch

For the online shop, try the following sequence. This is a suggested workflow, not a report of measured results from a real business.

• Catalogue questions: identify missing product details for the owner to resolve.
• Product page drafts: write descriptions using the corrected catalogue.
• Launch email: prepare an announcement based on the approved products.
• Customer questions: draft answers using confirmed care and ordering information.

At each handover, state what you approved. For example: “Use these three product names in the email. The attached descriptions are approved; the other drafts are still being reviewed.” Make your decision explicit in the next request.

For the email stage, adapt one of these email-writing prompts. A prompt is simply the request you give the assistant. Replace the sample audience and offer with the actual brief.

For the customer questions, start from these customer-service prompt examples. Have the owner review proposed replies before anyone uses them with buyers. Keep unresolved questions visible instead of smoothing them into confident copy.

Is ChatGPT Projects available on every plan?

The sources used here do not establish current availability across every plan. TechCrunch's launch report is historical coverage. Do not use it as today's complete availability table.

Check the account you intend to use before planning a larger workflow. Look for project creation, add a harmless sample file and start a conversation within the project. If a required option is missing, check the current plan information presented in your account.

For a team, write down who needs access, who supplies approved materials and who signs off the result. Verify the required access arrangements separately before moving team work into the setup. A successful personal test is only the first part of that decision.

If you are also considering another tool, use the Claude Projects guide as a starting point. Compare the same task and source material, then record how much checking each result requires.

How do you know the setup is useful?

Finish one real deliverable and record the work needed to make it usable. I suggest tracking preparation time, factual corrections and editing effort. Judge the finished result against the brief you wrote before starting.

Use a simple acceptance check

• Compare product facts with the approved source.
• Highlight any promise that the source does not support.
• Read the wording as a first-time buyer.
• Check that missing information is clearly identified.
• Have the responsible person approve the final version.

When an answer disappoints, name the problem before changing the setup. If dimensions are wrong, inspect the source and its version. If the language is vague, add an approved writing example and repeat the same task.

Keep a brief correction note: what was wrong, what you changed and whether the next answer improved. Avoid rebuilding everything after one weak response. For your first week, aim to finish one checked product page and document a process you would willingly repeat.