Hi, there!
Welcome to the 5th edition of Work in Beta.
In this edition, we urge you to build a system of records for your most used prompts and build your own Prompt Library. We have done this and see the merits in it and give you a step-by-step journey of building a similar one.
We are also hosting our first ever AI WORKSHOP for non-technical professionals later this month on 21st February and would love to have you join. The details are below in this email.
So, let’s dive in!
THE ‘HOW TO’ PLAYBOOK
Don’t lose your Best Prompts - Build your own Prompt Library instead
Image Credits: Nano Banana Pro / Work in Beta
Have you ever created an incredible prompt for a task you do repeatedly on ChatGPT? You used it all day. Multiple rounds of work. Great results every time.
Then a week later, you need to do the same work again. You search and search, but you can't find that chat thread. You can't find the prompt. So you write a new one from scratch. But this time? It doesn't work as well.
We've lost count of how many times this has happened to us. The LLM tools don't work well for retrieval. We know, it is FRUSTRATING, to say the least. We're used to file and folder systems. The emails have labels. But ChatGPT, Gemini, and Claude have hundreds of chat threads scattered across with no decent way to search.
Good prompts are the most valuable repeat assets for AI workflows. Yet, we treat them as disposable scratch paper.
We all need a systematic Prompt Library. Especially for prompts we use in repeat work that save us hours.
The Mistakes We See Everyone Make
The term "Prompt Library" is misleading. Social media is flooded with prompt libraries - top 100 prompts for sales, marketing, consulting, so on and so forth. A lot of us think, getting these prompts is going to save us time and give a quality output. It doesn't work out that way and here's why.
Mistake 1: Collecting prompts instead of indexing capabilities
Saving every clever prompt you find online is like building a museum. Instead, you need a toolkit. You need prompts that work for the tasks YOU do repeatedly.
Pasting prompt text into a doc without any meaningful indexing means when you need a prompt, you will run a random search of keywords. Along with that you'll have to re-read all the prompts that match the keywork you want, every single time.
Mistake 2: Organizing by AI tool instead of task
Creating "ChatGPT prompts" vs. "Claude prompts". The prompts are not different across tools. Good prompts work across, equally well.
Mistake 3: Treating prompts as final once saved
Saving prompts once and never updating them even when you discover better wording. So everytime you use this prompt, you keep editing inside your LLM tool of choice. But, you don't invest the time to update it ever in your prompt library.
Mistake 4: Saving prompts with hardcoded specifics
Saving "Analyze Acme Corp's website..." instead of "Analyze [COMPANY]'s website...". This again creates a reusability problem. Everytime you need this prompt, you will edit it for hardcoded stuff.
Mistake 5: Building a library you never open
You create a beautiful system that lives in a tab you never click. Basically, you have made it so hard to use that you never return to it. :)
The Prompt Card System
After testing different systems and making all the mistakes we spoke of over the last several months, we finally found a system that works. We are going to show you how to build it via prompt cards that are organised for searchability and reusability. Here is how the system works:
Part I - Task-Based Buckets (Organize by What You're Trying to Do)
Don't organize prompts by topic like files. Organize them by the actions they help you do.
Use 5 task buckets:
Write - emails, posts, summaries, reports
Analyze - research, comparisons, breakdowns
Transform - notes to action items, long to short, messy to clear
Think - brainstorm, decide, plan, solve
Extract - pull key points, data, insights from sources
Why this works: When you need a prompt, you think "I need to analyze something" or "I need to write something." You think in actions. In verbs. This helps you recall prompts faster.
The bucket structure matches how your brain searches for tools.
Part II - The Prompt Card Format
Save each prompt in a "card" with 4 elements:
1. Trigger - The task this handles, in your words (e.g., "competitor deep-dive" or "polite follow-up")
2. The Prompt - With [VARIABLES] clearly marked in brackets. These are the parts that change each time.
3. Variables Guide - What goes in each bracket. So you don't have to re-read the whole prompt.
4. Expected Output - Format and length you'll get. So you know if it fits your need.
Part III - Prompt Versioning (Keep It Lean)
Unlike documents, prompts improve through continuous iteration and use.
You run it. You tweak the wording. You get better results. You update the saved version.
Our rules:
Keep ONE current version per card (not v1, v2, v3 cluttering your library)
When you improve a prompt, replace the old version immediately
Delete prompts you haven't used in 2 weeks - they're not part of your real workflow
The goal is a lean, high-quality set you actually use, not a museum
(Unless you are building products, where versioning is a big topic, it is not something you need for solely personal use.)
Example
Here's what a complete Prompt Card looks like:
BUCKET: Analyze
TRIGGER: Competitor Analysis
PROMPT:
You are a market research analyst. Analyze [COMPANY]'s website and positioning.
Focus on: (1) What problem they solve, (2) Who they target,
(3) How they differentiate, (4) Pricing approach if visible.
Format as 4-section brief, max 300 words total.
VARIABLES:
- [COMPANY] = company name or URL
OUTPUT: ~300 word brief, 4 sections
Why this structure works:
Bucket & Trigger = how you'll search for it
Variables marked = what to swap without re-reading
Variables guide = no guessing on format
Expected output = know if it fits before running
After months of work, we have 15 Prompt Cards. We use 12 of them more than once weekly. The other 3 we use at least once every two weeks. That's it.
Every card earns its place by being useful repeatedly.
How to Build Your Library This Week
Day 1: Mine Your AI Chat History
Scroll through the last 2 weeks of AI conversations on ChatGPT, Gemini, Claude, etc.
Find 5 prompts you wrote more than once. Or wished you had saved.
Look for repeated tasks, not one-off work. The email report you draft weekly. The analysis you run for client proposals. The summary format you use after every meeting.
Copy the raw text into a new Google Doc titled "My Prompt Cards."
Day 2: Create Your First 5 Prompt Cards
For each prompt, add:
Trigger (how you'll search for it - "competitor analysis", "follow-up email")
Bucket (Write, Analyze, Transform, Think, or Extract)
Mark [VARIABLES] in the prompt text
Add a one-line Variables guide
Note expected output format
Imperfect cards you use beat perfect cards you don't.
Another example from our library:
BUCKET: Write
TRIGGER: Follow-Up Email
PROMPT:
Write a follow-up email to [PERSON] about [TOPIC].
Tone: [TONE - options: friendly, formal, direct].
Reference our previous conversation [CONTEXT].
Main ask: [SPECIFIC REQUEST].
Keep under 100 words.
VARIABLES:
- [PERSON] = recipient name
- [TOPIC] = what you're following up about
- [TONE] = friendly/formal/direct
- [CONTEXT] = when you last connected and what was the conversation about
- [SPECIFIC REQUEST] = what you're asking for
OUTPUT: ~75-100 word email
Day 3-5: Use the Cards
Before writing a new prompt, check your library first.
When you write a good new prompt, create a card immediately.
Day 6: Refine What You Used
Which cards did you run this week? What would make them faster to use?
Update based on real usage, not theory.
Day 7: Identify Your Next 5-10
What repeated tasks still have no card? Add these to a "Cards to Build" list.
Week 2 Goal: 12-15 active Prompt Cards, organized by bucket, that you check before every AI conversation.
Your Prompt Library After 2 Weeks
Here's what you'll have:
12-15 Prompt Cards organized into 5 buckets (Write, Analyze, Transform, Think, Extract).
Each card has: Trigger, marked [VARIABLES], Variables guide, Expected output.
You check it before every AI conversation. You add to it when you write something good. You update cards when you improve them.
You stop starting from scratch on the work that matters most.
Takes 15 minutes to set up. Saves 15 minutes per day forever.
That's the ROI.
LEARN WITH US
Workshop on Using ChatGPT Reliably at Work with PD & Sonali (for small group)
If you are a non-technical professional and want to use ChatGPT at work without worry and in complete control, we are running a 2-hour small-group session.
What you will learn:
what tasks to use AI for (and what not to)
hot to make AI outputs more reliable and consistent
one universal template for shareable, work-ready outputs
how to set up ChatGPT so it makes your style and needs
If you wish to join the session, fill this form and either of us will reach out to you with details: https://forms.gle/D6TGEVoFBUqKPsP49
THE OUTRO
That is all for today.
Thanks for taking 10 minutes to read through this newsletter. Please do let us know what did you think of it especially on how useful was it for you.
What did you think of today's newsletter?
You can also reply to this newsletter email if you want to share something specific with us. Have a great week ahead and see you again next week!
-Sonali & PD