• The AI Brief
  • Posts
  • AI Prompting 101: Getting Better Results from AI Tools

AI Prompting 101: Getting Better Results from AI Tools

Plus: latest reveals from Google I/O, OpenAI’s new hardware play, and why prompting is quickly becoming a must-have skill.

Author

Thaddeus Demeke
May 27, 2025

💼 In Today’s 5-min AI Brief

  • Your Everyday AI Toolkit: How to use Google’s NotebookLM as a hands-on prompting sandbox for your own documents and ideas

  • Prompt of the Week: A meta-prompt to help you learn prompting by doing, customized to the tasks you care about

  • AI News You Can Use: Google’s annual 2025 I/O event highlights, OpenAI buys Jony Ive’s LoveFrom, & Why prompting is quickly becoming a must-have skill in the job market (with stats to back it up)

  • What I’m Learning: Check out how AI is reshaping work, from Satya Nadella’s vision of agent-managers to real-world experiments shared by startup CEOs

If you've ever typed something into ChatGPT or Gemini and thought, “That’s not what I meant…”, you’re not alone. There’s the saying in tech, garbage in = garbage out; and this is also true for using your AI chatbots. 

In this issue of The AI Brief, I’ll break down the CARE framework for prompting, so you get more useful results from AI.

🛠️ Your Everyday AI Toolkit

Prompting in NotebookLM

If you liked last week’s deep dive on Gemini’s Deep Research, you’ll want to check out NotebookLM. It lets you ask AI questions about your own files - Docs, PDFs, Slides - so answers stay grounded in your content.

Notebook LM Interface

Try It Out

Go to notebooklm.google.com, create a new project, and follow these steps: 

  1. Upload at least 3 resources (docs, PDFs, slides - whatever you’ve got).

  2. Try the following prompts within NotebookLM:

Based on these resources, what are the top three takeaways if I were a [insert role] presenting to [insert audience]?

Create a study guide from these chapters if I were a [insert role] preparing for a [insert context].

Compare the viewpoints across these sources and format them into a [insert expectation].

Write Your Prompts with CARE

Here’s a simple framework that you can leverage to improve your results with AI:

C 

Context

Background for the task. 

E.g., “I’m preparing for our Q3 planning cycle at a fintech company. We’ve just finalized an internal roadmap that outlines key features, infrastructure upgrades, and customer feedback themes.”

A 

Action

What to do and for whom.

E.g., “Summarize the Q3 priorities in a way that makes sense for senior leadership reviewing strategic direction.”

R 

Result

Tone, role, and success criteria

E.g., “Act as a product manager and write an executive-level summary that’s high-level, outcome-focused, and avoids technical jargon.”

E 

Expectation

Optional:  A specific, good-output reference for the model to mimic

Q2 Strategic Priorities – Product Summary

In Q2, we focused on activation, reliability, and customer feedback integration.

• User Activation: Launched the new onboarding experience, which improved Day 1 activation by 14%—slightly under the 15% target but trending up.

• Platform Reliability: Completed backend migration to support increased usage, reducing average load times by 27%.

• Customer Feedback Loop: Rolled out usability improvements across our enterprise dashboard, directly addressing top support tickets.

These priorities laid the groundwork for deeper engagement and system scalability heading into Q3.

Put it together, and your prompt might look like:

“I’m preparing for a Q3 planning cycle at a fintech company (context). Act as a product manager (result), and based on the attached roadmap, write an executive-level summary for senior leadership explaining the Q3 priorities (action). Use the Q2 summary format as a reference (expectation).”

Prompting is in High-Demand

Here are three stats you should know:

💭 Prompt of the Week

You are a thinking parter designed to help me improve how I work with AI. I’m just starting to learn prompting. This week, I want to:

1. [insert text here] (e.g. write better summaries)

2. [insert text here] (e.g. brainstorm content ideas)

3. [insert text here] (e.g. draft professional emails)

Based on this list, walk me through three prompt variations I could try for each task. Help me understand why they work and how to refine them. Then, suggest how to practice this week.

This prompt helps you learn prompting by doing. It’s meta - using AI to improve how you use AI. With consistent practice, you’ll get faster, more accurate results, and be better equipped to use AI in meaningful, productive ways.

Try this prompt in your favorite AI tool - ChatGPT, Claude, Gemini, or even NotebookLM. You might be surprised how each one responds a little differently.

💡 AI News You Can Use

Last Week in AI

  • Everyone’s talking about Google I/O 2025. They released a NotebookLM project for anyone to explore. Here are our top 3 updates from the 2-hour event for you:

    • For everyone – Gemini AI Ultra introduces the ability to run deep reasoning over your own files - with full Google Drive integration coming soon. It’s part of the new Gemini Ultra plan at $250/month.

    • For creative work – Google dropped a new creative suite called Flow: (1) Veo 3 for videos with sound, Imagen 4 for images, and Lyria 2 for music. A Reddit user showed just how disruptive Veo 3 is - he used to make $500K pharma ads, but Veo 3 achieves the same thing for just $500… in a single day.

    • For dev workflows – Jules is a new, free coding agent & Stitch turns plain prompts into Figma or HTML. Jules has gotten great positive feedback online.

  • Microsoft’s annual event Build 2025 also happened - here’s the keynote in 14 mins.

    • M365 Copilot is live. Copilot Tuning is a big element of this because it allows you to build custom agents on your company data and workflows..

    • Azure AI Foundry. This platform lets you build custom agents & workflows.

    • Microsoft wants to lead the 'agentic web', where people and AI work together in every role. They’re also, like Meta, leaning heavily into open-source.

  • OpenAI’s future hardware plans are revealed with the $6.4 billion dollar acquisition of Jony Ive’s (designer of iPhone & Macbook) design firm, LoveFrom.

    • First of all, this might be the most cinematic acquisition announcement I’ve seen.

    • OpenAI might be going after Apple. In the announcement, they say the way we interact with our everyday physical tech is very outdated. They also mention they have prototypes of their product that’s in the works, and both agreed that it’s the most exciting piece of technology they’ve ever seen, and it’s planned to release in 2026.

  • Claude released new models, Opus 4 & Sonnet 4, and they’re calling them hybrid models. They both have two modes: (1) fast responses and (2) extended thinking for deeper reasoning. They claim Opus 4 is the world’s best model for coding & the internet agrees so far. Users have noted the high costs though.

📖 What I’m Learning

If you’re interested in what I’ve been digging into… check out:

  1. Satya Nadella (Microsoft CEO) discusses the future of knowledge work - Satya agrees that every knowledge worker will need to become an agent manager instead of just a knowledge worker.

  2. CEOs Discuss their AI Experiments - CEOs of Levels (CGM startup) & Tiny (holding company) exchange interesting perspectives on how AI will transform everyday jobs.

Final Thoughts

Prompting is the foundation for how we’ll collaborate with AI going forward.

Right now, these models aren’t perfect. But they’re getting sharper fast. And the people who’ve spent time learning how to think, communicate, and delegate with AI will be miles ahead when these systems can execute nearly any task at a human level, or better.

Think of prompting like compound interest: small reps now lead to huge leverage later.

Thaddeus 

Get in touch at [email protected]!

Reply

or to participate.