Save and Repeat What Works

Habit 2: Turn a Success into a System

You’ve found a prompt that works: a successful meeting summary, an effective follow-up email, or a usable data digest. The natural reflex is to move on, then repeat the same process from scratch the following week. Do the opposite: when something works, save it immediately and turn it into a reusable tool. This is the core of OpenAI Academy’s applied AI course: moving from an effective prompt to a repeatable workflow.

The reason is straightforward. You won’t remember the exact prompt that produced the result, the successive adjustments you made, or the context you provided. Without a backup, you restart the trial-and-error cycle every time. The cumulative effect adds up: for a task performed 50 times a year, a template that saves 5 minutes gives you back more than 4 hours annually on that single task alone.

Three Levels of Saving, from Simple to Robust

The first level is the raw prompt: copy the exact text that worked into a document, marking variable parts with brackets. This is the minimum, but it’s far better than nothing.

# Template: Weekly meeting summary
Here are my meeting notes: [paste notes]
Produce a structured summary with:
- Attendees
- Decisions made
- Action items (who, what, when)
- Deferred points
Professional tone, bullet points.

The second level leverages ChatGPT Projects. A project brings together conversations, files, and instructions. Create one for each recurring task, using a clear name like “Product Team Summaries.” Place your guidelines in the project instructions via the ••• menu, then Project settings: they apply to all conversations within the project and override your custom instructions. Add your reference documents, such as an internal glossary or a format example. You can also share a project with colleagues on any plan, granting either edit access or conversation-only access.

The third level involves documenting the complete workflow for complex tasks: the input data to gather before running the prompt, the template with its variables, the output points to verify, the most frequent edits, and the final recipient. On Business, Enterprise, Healthcare, and Edu plans, this workflow can become a Skill, a method that ChatGPT applies automatically when your request matches its description.

Finding the Right Template in Seconds

Over the weeks, you’ll accumulate several templates, and a disorganized library helps no one. Organize them by category: communication, meetings, analysis, creation. A shared document or a project per category is sufficient. For a task that needs to run autonomously, ChatGPT documentation advises refining the prompt in a conversation first, then scheduling the task within that same conversation.

Two pitfalls threaten this library. The first is stagnation: review each template regularly in light of the edits you consistently make. The second is rigidity: a template that’s too locked down breaks as soon as the case varies slightly, so include optional instructions. And don’t keep a working template to yourself: sharing it with your team benefits everyone.

Your Turn

Convert your first task into a template. Revisit the task you chose in the previous lesson. If you’ve already achieved a good result:

1. Copy the exact prompt that worked
2. Identify the variables, meaning the parts that change with each use
3. Replace them with bracketed placeholders
4. Create a project in ChatGPT and place the template in its instructions
5. Test it with new data

If the result doesn’t satisfy you yet, don’t freeze it: only save what works.

Official sources: Projects in ChatGPT · Skills in ChatGPT (OpenAI Help Center) · Prompting (ChatGPT Documentation) · New OpenAI Academy courses for the next era of work (OpenAI)

Key Takeaways

• When a prompt works, save it immediately: don’t rely on your memory
• Three levels: raw prompt, project with instructions, documented workflow or Skill depending on your plan
• Project instructions apply to all its conversations; projects can be shared with the team
• Organize your templates by category and update them regularly
• For a task that needs to run autonomously, schedule it once the prompt is refined