Your Prompt Generator: One Front Door to Consistent AI Outputs

What this episode is about

Episode 5 of Zero to AI is about creating one front door for better AI prompting. Instead of bouncing between tools, chats and half-remembered prompt styles, the episode introduces a simple prompt manager system.

The idea is practical: you describe what you want once, the manager routes the task to the right expert persona, checks the brief against basic quality guardrails and gives you a tool-ready prompt with run notes.

This matters because prompt quality often breaks down through inconsistency. Brand tone drifts. Outputs vary. People use different tools in different ways. A prompt generator gives you a reusable operating system for better, more repeatable AI work.

Why build an expert prompt system?

If you have ever tried to get consistent results from different AI tools, you will know how messy it can become. A prompt that works well in one tool may fall flat in another. A brand voice that feels strong in one output can disappear in the next.

A prompt manager helps solve that problem by turning prompting into a structured process. The user describes the outcome, chooses an output type or tool, and the manager produces a stronger prompt with format, tone, constraints, accessibility and run notes built in.

For businesses already exploring generative AI, workflow automation or AI agents, this is a useful pattern because it reduces power-user dependency and makes good prompting easier to share.

What you gain

A prompt manager gives you speed, consistency and reuse. You spend less time rewriting prompts from scratch and more time shipping useful outputs. It also helps teams because new people can follow the same route instead of guessing which chat, tool or prompt format to use.

The system can also support better governance. Brand wording, NZ spelling, accessibility reminders, citation nudges, privacy checks and disclaimers can be built into the normal flow rather than treated as afterthoughts.

Over time, successful prompts become reusable building blocks. They can be saved as templates, adapted for new channels and handed to another person without losing the context behind the task.

Key ideas from the episode

The basic flow

The prompt generator follows the same pattern every time: describe the outcome, choose how to proceed, route the request, draft and check the prompt, then return a final prompt package.

Describe the outcome

The most important part of the system is the outcome brief. You do not need to know the perfect tool or the best prompt structure before you begin. You need to describe the destination clearly.

Describe the outcome you want.

Include:
- Goal and format
- Audience and channel
- Success criteria
- Constraints, such as brand, time, file type or word count
- Inputs, such as links, data, templates or examples
- Failure fallback, such as: if anything is missing, create a placeholder and tell me what you need

This gives the manager enough context to route the request, ask smart follow-ups and return a better final prompt.

Example: content brief

A simple content brief can become a stronger prompt when it includes the audience, channel, tone and constraints upfront.

Create a 180 to 220 word LinkedIn post announcing our Spring Layering guide for NZ outdoor hikers.

Audience: mid-30s to 50s weekend hikers.
Channel: LinkedIn company page.
Success: conversational tone, NZ spelling, one CTA to download, no hashtags in the body, add three hashtags at the end.
Constraints: warm and practical brand voice, avoid jargon.
Inputs: link to guide, three product SKUs to mention.
Failure fallback: if SKUs are missing, write the post without them and add a [NEED SKUS] note.

Example: automation brief

An automation brief works the same way. It should name what runs, when it runs, where the data comes from, what success looks like and what happens if the data is missing.

Build a small workflow that posts a four-line daily sales summary to Slack #leadership at 8:30am NZ time.

Audience: executive team.
Success: shows yesterday’s revenue, top product, variance versus seven-day average and one actionable note.
Constraints: Google Workspace first; prefer n8n or Apps Script; under two hours to ship.
Inputs: Google Sheet called DailySales with date, product and revenue.
Failure fallback: if the sheet is unavailable, send a short data unavailable message and alert me.

Set up your Prompt Generator project

The episode walks through a simple ChatGPT Projects setup. Create a new project called Prompt Generator, then create three folders: Manager, Personas and Templates.

The Manager is the single starting point. Personas are the specialist chats. Templates store brand guardrails, reusable snippets and examples. The key habit is simple: always start in the Prompt Manager, not in the persona chats.

Prompt Manager starter card

Create a new chat in the Manager folder called Prompt Manager (Router), pin it, then paste this start card near the top of the chat.

START HERE: How to use the Prompt Generator

Always start in: Prompt Manager (Router).
Do not post requests in persona chats.

Flow:
1) Open Prompt Manager (Router)
2) Choose Output Type or Specific Tool
3) Paste your Outcome: goal, audience, channel, success, constraints, inputs and fallback
4) Manager asks for anything missing, routes, runs QA and returns:
- FINAL PROMPT
- RUN NOTES
- FOLLOW-UPS

Golden rule: if you are about to paste a new request and you are not in Prompt Manager (Router), stop and go there first.

Prompt Manager role script

Paste this into the Prompt Manager chat as the operating instruction for the router.

You are the Prompt Manager (Router).

Your job:
1) Prompt the user step by step to describe the outcome in plain English:
   goal, audience, channel, success criteria, constraints, inputs and failure fallback.

2) Ask how to proceed:
   - Output Type: Content, Image, Video, Research, Design, Automation
   - Specific Tool: Canva, Figma, Flux, Descript, CapCut, Runway, Kaiber, Uizard, Consensus, Elicit, scienceOS, n8n, Zapier, Make, Google Apps Script

3) Route to the best persona for that choice. Do not invent tools.

4) Draft a tool-ready prompt. Then run QA:
   size and format, brand tone, NZ spelling, accessibility, citations/legal, CTA.
   For automation, also check retries, errors and rollback.

5) Return:
   - FINAL PROMPT
   - RUN NOTES
   - FOLLOW-UPS

Guardrails:
- Prefer free or built-in tools when possible.
- Be concise and practical.
- Avoid jargon.
- Do not reveal chain-of-thought; summarise decisions only.

Persona examples

Personas are specialist chats you can add as needed. Start small. Add only the roles you actually use, then grow the bench as the system becomes useful.

n8n Flow Engineer persona

This is a useful first automation persona if you want the system to produce no-code workflow specs.

You are the n8n Flow Engineer.

Output a beginner-friendly mini-spec:
- Trigger
- Nodes with configuration hints
- Auth and credentials map
- Transform steps
- Error paths and retries
- Alerts by Slack or email
- Rollback plan
- Three-step test plan

Return JSON-ish hints that a beginner can follow in the n8n UI.

Google Apps Script Dev persona

This persona is useful when your workflow lives inside Google Apps Script, Sheets, Docs or Gmail.

You are the Google Apps Script Dev.

Produce minimal, commented code for Sheets, Docs and Gmail.

Include:
- What the script does
- Where to paste it
- Required permissions or scopes
- Trigger setup
- A tiny test
- A rollback note

Assume the user is not a coder. Explain each step in plain English.

Brand guardrails

In the Templates folder, create a note called Brand Guardrails. This should hold the core rules the manager and brand-sensitive personas need to remember.

Brand Guardrails

Voice:
- Clear, practical and human
- New Zealand spelling
- Avoid hype and jargon
- Explain things step by step

Design:
- Brand colours
- Fonts
- Layout rules
- Image or no-image rules
- Export formats

Quality:
- Accessibility reminders
- Citation or source expectations
- Legal or privacy notes
- Must-include and must-avoid phrases

How to ask for the best output

Always start in the Prompt Manager. Choose either output type or specific tool, then paste your outcome. Keep it simple, but include enough detail for the manager to route the task well.

I choose: [Content / Image / Video / Research / Design / Automation]

Outcome:
Goal and format:
Audience and channel:
Success criteria:
Constraints:
Inputs:
Failure fallback:

Please ask for anything missing, route this to the best persona, run QA and return a final prompt package.

Run your first ten-minute test

The best first test is small. Do not try to build the perfect internal AI operating system on day one. Pick something useful you will actually run today.

For example, ask the manager to create a LinkedIn carousel prompt, a short newsletter brief, a Canva design prompt or a small automation spec for a daily Slack summary. Then copy the final prompt into the tool it mentions and see what happens.

Save the best result into Templates as an example. That is how the prompt generator becomes an internal playbook over time.

Where to go next

This page is designed to stand alone as a foundation episode. You can listen, read, reflect and build a simple Prompt Generator without moving into a more advanced learning experience. If the idea resonates, return to the Season 1 archive and keep exploring the foundation journey.

You can also visit the Zero to AI blog for related reflections, or use the Start Here page to understand the practical learning approach behind Zero to AI.