# How do you turn the book's prompt templates into a reusable project or GPT?

Date: 2025-03-03 · Author: Robert Scales · https://guidetoartificialintelligence.com/en/learn/turn-books-prompt-templates-into-reusable-project/

**Answer:** Split each template into the part that never changes (the role, your business context and the output format) and the part that changes every time (the task). Put the fixed part into a ChatGPT Project's instructions and files, or into a GPT's instructions and knowledge, and from then on you only type the task.

*A Beginner's Guide to ChatGPT* teaches a method that still holds up: set ChatGPT's role, give context about your situation, state the task, say what format you want, refine through follow-ups, and verify the result. The book shows it in prompts for many professions, each written in three versions: simple, improved and refined. The refined version gives the best answers, but it is also the longest to type, and much of it is the same every time you use it.

ChatGPT now has two places to store that repeated part. A **Project**, added in December 2024 for Plus, Team and Pro users, is a workspace that groups chats and files and has its own custom instructions, which every chat inside the project shares. A **GPT**, introduced in November 2023, is a custom version of ChatGPT with its own instructions, extra knowledge in uploaded files and chosen capabilities such as web search or data analysis, which you can keep private or share.

## Step 1: Separate what stays from what changes

Take the book's marketer example. Its refined prompt asks ChatGPT to analyze customer data for an e-commerce store selling eco-friendly products, find three customer segments and their characteristics, and build a campaign for each with channels, messages and key performance indicators (KPIs, the numbers you track to judge success). The book's preparation chapter adds that a marketing manager should first explain the strategy, audience, brand voice and KPIs.

Sort those pieces:

| Stays the same | Changes each time |
|---|---|
| Role: experienced digital marketing manager | Which campaign or season |
| Business: eco-friendly online store, its products and values | Which data file to analyze |
| Audience, brand voice, channels used | Budget and deadline |
| Output format: segments, then a plan per segment with channels, message, KPIs | Any special angle or offer |
| Rules: flag assumptions, don't invent statistics | |

The left column becomes the instructions. The right column is what you will type in each chat.

## Step 2: Build it as a Project

Create a project, name it after the job ("Store marketing"), and paste instructions like these into its custom instructions. The store details are illustrative; replace them with yours.

```prompt
You are an experienced digital marketing manager working for our online store, which sells eco-friendly home products to customers in Canada.

Context: our brand voice is warm, practical and never preachy. Our main channels are Instagram, a monthly email newsletter and our blog. Our usual KPIs are email click rate, Instagram saves and online sales per campaign.

When I ask for a campaign:
1. If I attach data, identify up to three customer segments and describe each in two lines.
2. For each segment, propose: channel, main message, one example post or email subject line, and the KPI to track.
3. End with a short list of the assumptions you made.

Rules: use only figures from my files or from me. If a number is missing, say so instead of estimating. Keep each plan under 250 words.
```

Then upload the reference files that stay the same: a one-page brand guide, your product list, last year's campaign results. Each new chat in the project starts with all of this in view, so your message can be as short as "Spring campaign, budget $800, focus on reusable kitchen wraps, data attached."

## Step 3: Or build it as a GPT

A GPT suits the same template when you want to share it with colleagues or use it outside a single workspace. You build one without code, by describing what you want in a conversation with OpenAI's builder and then giving it instructions, knowledge and capabilities. Paste the same instruction text, upload the same reference files as knowledge, and turn on only the capabilities the job needs, for example data analysis for spreadsheets. Give it a clear name and a one-line description so colleagues know what it is for.

The main difference is scope. A project also keeps your chat history and working files together, which suits ongoing work. A GPT is a reusable tool you open fresh each time and can hand to someone else.

> **Try it:** Pick the book prompt closest to your own work. Write the "stays the same" column for your situation, paste it into a new project's instructions, then run two different tasks in two chats and compare the answers with what the original refined prompt gave you.

## Keep the book's last two steps

Storing the instructions does not remove iteration or verification. Your first few uses will show gaps: a format that is too long, a channel you forgot, a rule the model ignores. Fix those in the instructions rather than in each chat, so every later conversation benefits. And as the book warns, the model can still produce wrong figures or claims with confidence, so check any number before it reaches a customer.

## Which templates are worth turning into tools?

The best candidates share three traits. You use them often, at least every week or two. Most of the prompt stays the same between uses. And the result has a recognizable shape, such as a campaign plan, a lesson outline or a monthly report, that you can describe once as a format.

Good examples from the book's professions include the marketer's campaign planner, an educator's lesson-plan builder, a researcher's summary of sources in a fixed citation style, and a salesperson's outreach email in the company's voice. Each depends on background that rarely changes: your brand, your curriculum, your field, your product.

Poor candidates are one-off questions, tasks whose context changes completely each time, and anything where the stored files would hold sensitive personal or client data you should not upload. For those, a well-written prompt saved in a notes file is enough. Start with one tool, use it for a few weeks, refine its instructions, and only then build the next.

## Sources

- A Beginner's Guide to ChatGPT: https://www.amazon.com/dp/B0BZRRPXZW
- ChatGPT — Release Notes (entry: Projects in ChatGPT, December 13, 2024): https://help.openai.com/en/articles/6825453-chatgpt-release-notes
- Introducing GPTs: https://openai.com/index/introducing-gpts/