Start Lesson
Here is what most people do with AI: need something, open ChatGPT, write a prompt from scratch, use the output, close the tab. Next week, same task, start over. The prompt that produced a great client email last month? Gone. Buried in chat history.
Write an email to the client about the delay.
Twelve minutes later, they have a usable email. But they spent the same twelve minutes last month on the same type of email. And they will spend it again next month.
Now here is what happens with a system:
[Open prompt library → email-client-escalation-v3]
Act as a senior project manager at a design agency.
Write a follow-up email to a client who has not responded
to [NUMBER] messages about [APPROVAL NEEDED].
Context: [SITUATION AND STAKES]
The email should:
1. State the impact of the delay in one sentence
2. Reference one specific positive result from the last phase
3. Offer two next steps: [DEADLINE] or [ALTERNATIVE]
Tone: professional but firm. Under 120 words.
Do not use: "I hope this email finds you well,"
"unfortunately," or "please do not hesitate."
Ninety seconds: fill in the brackets, run the prompt, get a usable draft. The template already has the right role, structure, constraints, and exclusions because you solved this problem once and saved the solution.
This is the difference between someone who is "good at AI" and someone who is productive with AI. It is a system, not a skill.
After this lesson, you will be able to: convert your best prompts into reusable templates, organize them in a prompt library, and set up system prompts that make every interaction faster.
When you write a prompt that produces great output, do not just save it. Convert it into a reusable template. Here is the prompt that does that conversion for you:
I have a prompt that worked well. Convert it into a reusable
template by:
1. Replacing specific details with clearly named [VARIABLES]
in all caps
2. Adding a "Variables to fill in" section at the top that
lists each variable with a one-line description
3. Keeping all constraints, structure, and role instructions
intact
Here is my original prompt:
---
[Paste your working prompt here]
---
Output the template in a code block I can copy directly.
Expected output: A clean template with labeled variables, a fill-in guide at the top, and all the structural elements that made the original prompt work. This takes a one-time success and makes it infinitely reusable.
Works with Claude, GPT-4, and Gemini. Save the output directly to your prompt library.
If your AI tool supports custom instructions or system prompts (ChatGPT, Claude, Gemini all do), this is the highest-leverage prompting work you can do. A system prompt sets persistent behavior -- it is like the AI's standing job description.
Build me a system prompt for [AI tool] that sets these
defaults for all my conversations:
Role: I am a [your role] at [company type] working in
[industry].
Default audience: [who I usually write for].
Tone: [your standard tone -- direct/warm/technical/casual].
Format preferences: [bullets vs. paragraphs, length defaults].
Standing constraints: [things to always avoid -- specific
phrases, jargon, behaviors].
Keep it under 200 words. Every instruction should be
concrete and actionable -- no vague guidelines like
"be helpful."
Expected output: A concise system prompt you can paste into your AI tool's custom instructions. Once set, every prompt you write benefits from these defaults automatically. You stop repeating your role, tone, and constraints in every single prompt.
Set this up once. It takes 15 minutes and saves that time back every single day.
Here is the structure for organizing your library. You do not need a fancy tool -- a shared doc, a Notion page, or even a folder of text files works.
Help me set up a prompt library structure. I work as a
[role] and my most common AI tasks are:
1. [Task type 1, e.g., client emails]
2. [Task type 2, e.g., content writing]
3. [Task type 3, e.g., competitive research]
4. [Task type 4, e.g., meeting prep]
For each task type, create:
- A category name using the format: task-type (e.g.,
"email-client")
- A naming convention for prompts: task-type-specific-use-v#
(e.g., "email-client-escalation-v1")
- One starter template prompt with [VARIABLES] that I can
use immediately
Format the output as a structured document I can copy into
my library tool.
Expected output: A ready-to-use library skeleton with four categories, naming conventions, and one working template per category. You start with four prompts and grow from there.
Naming convention in practice:
email-client-follow-up-v2 -- v2 added the exclusion for "I hope this email finds you well"research-competitor-analysis-v1 -- first version, tested with 3 competitor setscontent-linkedin-post-v3 -- v3 switched to AIDA structure from Lesson 4Save new versions with a one-line note: "v2: added constraint to avoid jargon." Update when the model changes, your voice evolves, or the team keeps making the same manual edit.
Everything you learned in this course stacks. Here is a single prompt that uses every technique:
[SYSTEM PROMPT - set once]
You are a senior marketing strategist at a B2B SaaS company.
Default to concise, direct communication. Never use buzzwords.
[USER PROMPT - from your library: research-market-entry-v2]
Act as a market analyst with 10 years in [INDUSTRY].
Research whether [COMPANY] should enter [MARKET SEGMENT].
Before giving your recommendation, work through these steps:
1. What are the 3 key factors in this decision?
2. For each factor, what does the evidence suggest?
Rate confidence: High / Medium / Low.
3. What is the strongest argument AGAINST entering?
4. Recommendation: enter, wait, or skip -- with reasoning.
Limit to 400 words. Format as numbered steps followed by
a final recommendation paragraph.
That single prompt uses CGC (Lesson 1), role prompting (Lesson 2), chain-of-thought (Lesson 3), research framing (Lesson 5), and reusable variables (this lesson). The techniques compound.
Do these three things before you close this lesson:
Pick your best prompt from this course -- the one that produced the most useful output. Run Template 1 (Prompt-to-Template Converter) to turn it into a reusable template with variables.
Set up your system prompt. Use Template 2 to build custom instructions for whichever AI tool you use most. Paste it in. Every future prompt benefits immediately.
Start your library. Use Template 3 or simply create a doc with four categories that match your actual work. Save your converted template from step 1 as the first entry.
Check your progress: If you can open your library, grab a template, fill in the variables, and get a usable output in under 2 minutes, your system is working.
You have finished Prompt Engineering That Works. You built: the CGC framework, role prompting, chain-of-thought reasoning, domain templates for writing and research, and a system that makes every prompt reusable. These techniques stack -- and the more you combine them, the faster you get results.
The next course, AI Strategy for Your Business, takes this further: where AI fits in your operations, how to calculate the ROI, and how to build an adoption plan for your team.