Workshop: Build Your Own Prompt Workflows

You have the techniques, the templates, and the library. One step remains: assembling them into workflows. A workflow is a sequence of interconnected prompts that covers an entire process, from initial research to final deliverable. This is what separates ad hoc usage—one prompt, one result, starting from scratch the next day—from a true production system you can run every week without thinking.

The Logic of the Chain

In a workflow, each step produces an output that becomes the input for the next. Think of a recipe: each action transforms the ingredients and prepares the next step, and the order is non-negotiable. Concretely, where you once had an isolated prompt, you now have a chain: research, then structuring, then drafting, then optimization, all the way to the deliverable. The gain isn’t just time saved; it’s quality, because each step works on material that’s already been refined and validated instead of starting from scratch.

The Sequential Workflow

This is the most common form and the easiest to set up. Each link depends strictly on the previous one, giving you a natural validation point between steps—you can stop, correct, and restart. The article creation workflow below is a typical example: the angle is decided before the outline, the outline before the draft, and optimization comes last, once the text exists and there’s something to optimize.

STEP 1 — Research and Angle:
You are a content strategist. For the topic "[TOPIC]" aimed at
[PERSONA], propose 3 original angles. For each angle, indicate:
- The working title
- The targeted search intent
- What differentiates it from existing articles

STEP 2 — Detailed Outline (after choosing the angle):
The chosen angle is: [CHOSEN ANGLE]
Create a detailed outline with:
- SEO-optimized H1
- H2/H3 structure with key content for each section
- Estimated word count per section
- Data points or examples to include

STEP 3 — Drafting (section by section):
Draft the section "[H2 TITLE]" following the outline.
Tone: [DEFINED TONE]. Length: [NUMBER] words.
Include the following example: [EXAMPLE]
Ensure smooth transitions with the previous section.

STEP 4 — Final Optimization:
Review the complete article and:
1. Check tone and style consistency
2. Add missing secondary keywords: [LIST]
3. Optimize the introduction to hook readers in 2 lines
4. Verify that each section provides unique value
5. Write the meta description (155 characters)

The Parallel Workflow

When tasks are independent, there’s no reason to chain them: you open multiple simultaneous conversations, each with its own role, and only converge them at the end. Brought together in a single project, these conversations share reference files and project instructions, which limits drift. A product launch fits this model particularly well, because messaging, content, and the sales kit don’t need each other to move forward—as long as you verify afterward that they tell the same story.

BRANCH A (conversation 1) — Messaging:
You are a product marketing manager.
Create the messaging framework for [PRODUCT]:
- Core value proposition
- 3 key messages per persona
- Elevator pitch (30 seconds)

BRANCH B (conversation 2) — Content:
You are a content manager.
Plan the launch content:
- Announcement blog post
- 5 LinkedIn posts (1 per day)
- Announcement email to existing customers
- Product FAQ (10 questions)

BRANCH C (conversation 3) — Sales:
You are a sales director.
Prepare the sales kit:
- Battle card vs. competitors
- Common objections and responses
- Demo script (15 minutes)
- Prospecting email

CONSOLIDATION:
Here are the results from the 3 branches. Check message
consistency across all documents and flag any discrepancies.

The consolidation step isn’t an administrative formality: it catches the divergences that parallel work inevitably produces—the value proposition from Branch A that no longer quite matches the demo script from Branch C.

The Iterative Workflow

Here, the same task is tackled from multiple angles, then the best elements are merged. This method suits deliverables where perspective determines success: a pitch, a scope note, a sales argument. Each iteration illuminates a facet the others leave in the shadows, and the final synthesis recovers the best of what each produced.

ITERATION 1:
You are a storyteller. Write a 2-minute pitch for [PRODUCT]
using the "problem - solution - proof" structure.

ITERATION 2:
You are a VC investor. Write the same pitch but emphasize
the market and the business model.

ITERATION 3:
You are a potential customer. Write the same pitch from the perspective
of concrete benefits for the user.

SYNTHESIS:
Here are 3 versions of the pitch. Combine the best elements from each
version to create the final pitch. Take the storytelling from V1,
the business numbers from V2, and the customer benefits from V3.

Designing Your Own

Start by choosing the right candidate: a process you repeat that costs you time. Three questions are enough to uncover it—what task takes me more than thirty minutes every week, what process always follows the same steps, where do I lose the most time reformatting or iterating? Then map the process as you actually practice it today, not as you think you practice it: for each step, note what it consumes and what it produces.

My current process for [TASK]:
1. [STEP] -> produces [DELIVERABLE]
2. [STEP] -> produces [DELIVERABLE]
3. [STEP] -> produces [DELIVERABLE]
...

All that remains is to write a dedicated prompt for each step, drawing on the course techniques—role, constraints, format, few-shot as needed. Then run the complete workflow on a real case and note every friction as it arises: only this live run reveals prompts that are too vague and connections that don’t click. Correct, then document the final version.

Running a Workflow in ChatGPT

A workflow doesn’t have to remain a series of copy-paste actions. ChatGPT Work is designed for tasks that chain steps, draw on multiple sources, and produce deliverables: you can entrust an entire chain to it, provided you request a plan, review intermediate results, and keep final approval for any sending or publishing. For a recurring workflow, ChatGPT documentation recommends first refining the prompt in a normal conversation, then, once the result is reliable, scheduling a task in that conversation; a standalone scheduled task works better when each execution must open a new conversation.

A Complete Workflow: The Sales Proposal

Here is a ready-to-use chain that you can adapt to your offer. Notice how each step explicitly feeds on the previous one: the profile fuels the needs analysis, which fuels the proposal, which fuels the sending email. No link restarts from scratch.

STEP 1 — Client Research:
You are a B2B sales rep. Analyze the profile of [COMPANY]:
- Industry and size
- Probable business challenges
- Keywords from their communication (website, LinkedIn)
Summarize in 10 lines what we need to know to personalize
our approach.

STEP 2 — Needs Identification:
Based on this profile, identify:
- The 3 problems our [PRODUCT] solves for them
- Quantifiable benefits (time saved, costs reduced, etc.)
- Probable objections

STEP 3 — Proposal Drafting:
Write a structured sales proposal:
1. Cover page (personalized title)
2. Context and understanding of challenges (show you know them)
3. Proposed solution (only relevant features)
4. Expected benefits (quantified)
5. Deployment methodology
6. Investment and terms
7. Next steps

STEP 4 — Sending Email:
Write the email accompanying the proposal.
Short (80 words), personalized, with a reminder of the identified need
and a meeting proposal to discuss the proposal.

Your Turn

Your first workflow. Put this into practice with a process you know by heart, so your attention focuses on the mechanics of the chain, not the content.

EXERCISE: Create a workflow for [CHOOSE ONE TASK]:
a) Prepare a client meeting from A to Z
b) Produce a complete weekly report
c) Onboard a new team member
d) Launch a marketing campaign

Step 1: List the 4 to 6 steps of your process
Step 2: For each step, identify the input and output
Step 3: Write the prompt for the first step
Step 4: Test it and refine
Step 5: Move to the next step, injecting the result
          from the previous step as context

Once the chain is complete, test it three times with different cases. This is the only way to know if it holds up beyond the specific case used to build it—a workflow calibrated for a single client often breaks on the second.

Document to Avoid Starting Over

A workflow you haven’t documented will be forgotten within three weeks, and you’ll rebuild it worse than before. Five minutes of documentation is enough to make it reusable by you and transferable to a colleague who wasn’t present for its creation.

WORKFLOW NAME: [name]
OBJECTIVE: [what it produces]
ESTIMATED TIME: [execution duration]
NUMBER OF STEPS: [X]

Step 1: [title] — prompt in [location]
  Input: [what is needed]
  Output: [what is obtained]

Step 2: [title] — prompt in [location]
  Input: [result from step 1]
  Output: [what is obtained]
...

NOTES: [tips, pitfalls to avoid, possible variations]

Two drifts then threaten your workflows. The first is bloat: beyond half a dozen steps, it becomes hard to execute the chain to the end, and you fall back on improvisation halfway through; you must then simplify or split into sub-workflows. The second is more insidious—poorly linked steps, where the prompt doesn’t explicitly reference the previous result. Never count on the model to make the connection on its own: write it, at every link.

Official sources: Prompting (ChatGPT documentation) · Projects in ChatGPT · ChatGPT release notes (OpenAI Help Center)

Key Takeaways

• A prompt workflow chains multiple prompts to cover a complete process
• Three forms: sequential, parallel, and iterative
• Design it in five stages: identify, map, design, test, optimize
• Document each workflow with its steps, inputs, and outputs
• To execute, rely on Projects, ChatGPT Work, and scheduled tasks, while keeping control over sending

Test Your Knowledge

Before the final workshop, a check on the course techniques.

1. What are the four elements of a good prompt according to ChatGPT documentation?

Answer: Goal, context, output, and constraints. For a simple request, one sentence suffices; for an important task, include the elements that matter, describing the expected result rather than every step.

2. When is few-shot the right technique?

Answer: When the expected result is easier to show than to describe: precise format, subtle tone, edge cases. Try without an example first, then add two or three examples that follow the instructions exactly.

3. Should you still write "think step by step"?

Answer: No. ChatGPT models reason internally, and OpenAI considers this phrase unnecessary, even harmful. For a difficult problem, raise the reasoning level, provide all data and your success criteria, and ask for a structured answer you can verify.

4. What is meta-prompting?

Answer: Having ChatGPT write or improve a prompt itself: you describe the task and observed failures, it proposes an optimized version—an accelerator to validate by testing.

5. Name three of the most common prompting errors.

Answer: Asking for multiple things at once without structure, omitting the output format, and not iterating—accepting the first response instead of refining through successive feedback.

The preceding workshop turns these techniques into personal workflows—this is where the course becomes a reflex.