Beginner’s Guide to ChatGPT Work (ChatGPT Projects 101)

Create the source folder on your computer before opening ChatGPT. Put a two-digit number at the beginning of its name, such as 01-meeting-notes or 12-client-research.

The folder gives the project a stable home in your file system. The numeric prefix keeps related projects ordered and easy to scan, while selecting the folder yourself prevents ChatGPT-created work from disappearing into a location you did not choose.

Open the desktop ChatGPT app, then go to New chat > Work > Choose project > New project. Name the project and select the numbered folder as its source folder. Connect the folder even if it is empty; this gives the project a known home before ChatGPT creates anything.

*The Meeting Notes project is connected to its local demo-materials source folder before any processing begins.*

Give ChatGPT a small set of representative files and ask it to inspect them before it changes the folder:

Inspect the starter files in this project and propose a folder structure for processing meeting notes.

The workflow should:
- accept new files through one input folder
- rename and organize successfully processed notes
- extract decisions, action items, owners, deadlines, conflicts, and open questions
- produce a meeting index and weekly action plan

Show me the proposed structure before creating or moving anything.

Treat the proposal as a first draft. In the recorded example, ChatGPT suggested separate input, originals, normalized, and archive folders. That created more file juggling than the job needed.

Push back when the structure is too elaborate:

This feels duplicative. Simplify the project to three working folders:

- input for new files
- notes for the permanent source library
- outputs for finished indexes and action plans

Keep in mind that automations will maintain and process this project.

This creates a simple mailbox pattern. New work enters through input, trusted source files live in notes, and regenerated deliverables go to outputs. You can reuse the same pattern for screenshots, research, customer feedback, or any assignment with repeatable inputs and outputs.

Once you approve the structure, ask ChatGPT to create it and process the starter files one time:

Scaffold the approved folders, then process the starter meeting notes once.

For this first run:
- rename and move valid sources from input to notes
- build the meeting index and weekly action plan in outputs
- cite the source filename for every confirmed decision and action
- do not overwrite source files or invent missing owners, dates, priorities, or decisions

Stop after the first run and show me what changed before creating any schedule.

Review the result before automating it. Confirm that each processed file moved into notes, the meeting index includes every source, and the weekly action plan does not invent owners or deadlines. Open the generated files so you can compare important decisions and actions with the underlying notes.

Once the manual run works, turn the verified process into reusable routines:

Turn the verified process into two reusable workflows:

1. Process new files in input, rename and move valid sources to notes, then rebuild every output.
2. Refresh the meeting index and weekly action plan from all files currently in notes.

Create scheduled automations for the timing I approve, then turn both workflows into skills I can run manually. Make the scheduled and manual versions use the same rules.

The meeting-notes demo uses one intake skill and one refresh skill. The intake skill processes new material, while the refresh skill rereads the permanent notes library and rebuilds the rollups. Keeping the rules in reusable skills means you do not have to explain the process again every time.

Use schedules only when the timing is useful. A weekly refresh may be enough for a team action plan; an inbox that changes throughout the day may need a more frequent run. If Cloud scheduling is available in your account, the blank task form lets you choose where and how often a task runs before you add its instructions.

*This blank task form shows one Cloud/Daily schedule option; add a title and instructions before saving it.*

Pro tip: Let a new automation run once, then point future runs to the existing project chat when that option is available. Creating a fresh chat for every run can quickly fill the project with duplicate threads.

Use the lightest capable option for each part of the workflow. Routine intake, renaming, and scheduled refreshes usually do not need the most demanding model. A lighter model can make recurring maintenance more efficient.

Switch to Codex when the project needs heavier file edits, scripts, or debugging. The recorded workflow uses this distinction to keep basic maintenance simple while reserving the more capable coding workflow for jobs that actually need it.

Specific model names and controls can change. Focus on the job instead: use a routine option for predictable maintenance, and move to a deeper coding or reasoning mode when the work becomes complex.

Pro tip: Run automations locally when they need to change files on your computer. Cloud schedules are better for recurring research or tests that do not depend on those desktop files.

Before you share a result, check it against the source material. Open the generated files directly and verify the important decisions, actions, owners, and dates against the notes that produced them.

If the work includes web research, ask ChatGPT to open its cited sources in the built-in browser:

Open the cited sources in the browser so I can review them before sharing the result.

*The cited research source is opened beside the ChatGPT conversation for a quick evidence check.*

This gives you a fast way to verify that a source exists and supports the claim. It does not replace reading the source or applying your own judgment.

When the job is finished, archive completed chats and keep the reusable project, skills, and source folder. The next run should start with new material in input, not with another round of setup.