⏱️ The Hidden AI Features Saving People 10 Hours Per Week

Most people are wasting 80% of AI's potential. Here's how to work smarter with the tools you already have.

Everyone's using AI now. But most people are still treating it like a magic 8-ball—ask a question, get an answer, hope it's useful. Meanwhile, there's a whole layer of techniques that separate people who get mediocre results from people who are genuinely 10x more productive.

These aren't theoretical hacks. They’re practical workflows you can implement today using tools you already have access to.

Tip 1: Turn Messy Research Into Structured Data With NotebookLM’s New Data Tables

Google just added Data Tables to NotebookLM (December 23, 2025), and it’s one of those features that sounds boring but fundamentally changes how you work with information.

What changed:

NotebookLM can now scan all your uploaded sources—PDFs, meeting transcripts, research papers, web pages—and organize specific information into structured tables that export directly to Google Sheets. No more manual data entry. No more switching between 15 tabs to compare information.

The feature is powered by Gemini 3, which just got upgraded for "significant improvements to reasoning and multimodal understanding." Translation: it’s actually smart enough to pull the right information and organize it correctly.

Real-world uses:

• Competitive analysis: Upload 10 competitor websites and ask NotebookLM to create a table comparing pricing, features, target customers, and key differentiators.
• Literature review: Upload 20 research papers and generate a table of methodologies, sample sizes, key findings, and limitations—what used to take days now takes minutes.
• Meeting summaries: Feed it a month of meeting transcripts and get a structured table of all action items, owners, and deadlines.
• Curriculum planning: For educators, extract state standards, learning objectives, and assessment criteria from various documents into one clean table.

How to use it step-by-step:

1. Go to notebooklm.google.com (completely free, no credit card)
2. Create a new notebook and upload your sources (PDFs, Google Docs, Sheets, Drive URLs, .docx files, Images)
3. In the chat, ask NotebookLM to create a specific table: “Create a table comparing [X, Y, Z] across all sources”
4. Review the generated table and refine if needed
5. Export directly to Google Sheets with one click

Copy-paste prompts:

• For product research: “Create a table with columns: Product Name, Price, Key Features, Target Audience, Pros, Cons. Extract this information from all uploaded sources and organize by price from lowest to highest.”
• For academic research: “Generate a table with columns: Study Title, Year, Sample Size, Methodology, Key Findings, Limitations. Include all studies mentioned across these sources.”
• For business analysis: “Create a table showing: Company Name, Revenue Model, Market Position, Strengths, Weaknesses, Recent News. Use all the competitor analysis documents I uploaded.”
• For project management: “Extract all action items from these meeting transcripts into a table with columns: Task Description, Owner, Deadline, Status, Priority Level.”

Pro workflow tip:

Combine Data Tables with NotebookLM’s Deep Research feature (also just launched):

1. Use Deep Research to automatically find 50+ quality sources on your topic
2. NotebookLM browses the web and builds a comprehensive report
3. Then use Data Tables to extract specific data points into a structured format
4. Export to Sheets for further analysis or sharing with your team

This workflow turns “3 days of research and data entry” into “20 minutes of AI-assisted work.”

Key benefits:

• ✅ Free to use (no subscription needed)
• ✅ Exports to Google Sheets for easy sharing and further analysis
• ✅ Works with multiple file types (PDFs, Docs, Sheets, Images, web URLs)
• ✅ Gemini 3-powered means better accuracy and understanding
• ✅ Saves hours of manual data compilation

Start using NotebookLM + Data Tables guide • Deep Research tutorial

Tip 2: Master "Few-Shot Prompting" to Jump From 0% to 90% Accuracy

Here’s the truth about prompting: telling AI what you want rarely works as well as showing it examples. This technique is called “few-shot prompting” and it’s the difference between frustrating back-and-forth and getting exactly what you need on the first try.

What few-shot prompting is:

Instead of describing what you want in words, you show the AI 2–3 examples of perfect outputs. The AI learns the pattern and replicates it for new inputs.

Think of it like teaching someone to write professional emails. You could spend 10 minutes describing what “professional but friendly” means, or you could just show them 3 good examples and say “like these.”

Research shows few-shot prompting can improve accuracy from near-zero to 90%+ on complex tasks.

Basic structure:

TASK: [What you want the AI to do]

EXAMPLES:
Input: [Example 1 input]
Output: [Example 1 perfect output]

Input: [Example 2 input]
Output: [Example 2 perfect output]

Input: [Example 3 input]
Output: [Example 3 perfect output]

NOW DO THIS:
Input: [Your actual input]
Output:
  

Real examples you can use today:

For writing product descriptions:

TASK: Write compelling product descriptions for e-commerce

EXAMPLES:

Input: Wireless headphones, $79, noise-canceling, 30hr battery Output:
“Block out the world and lose yourself in crystal-clear sound. These wireless headphones deliver professional-grade noise canceling and an industry-leading 30-hour battery life—perfect for long flights, deep work sessions, or marathon gaming weekends. At $79, it’s studio-grade audio without the studio price.”

Input: Yoga mat, $45, eco-friendly, extra thick Output:
“Your practice deserves better than a thin, slippery mat. This eco-friendly yoga mat provides the cushioning your joints need and the grip your poses demand. Made from sustainable materials that are kind to the planet and comfortable for you, $45 for a mat that’ll last for years.”

NOW DO THIS: Input: Standing desk, $299, electric, memory settings Output:

For data extraction:

TASK: Extract key information from customer feedback

EXAMPLES:

Input: “The app crashes every time I try to upload photos. Really frustrating! Love the interface though.”
Output: Issue: “app crashes during photo upload”, sentiment: “negative”, feature_mentioned: “Interface”, praise: “loves the interface”

Input: “Been using this for 3 months. Game changer for my workflow. Would love to see dark mode added.”
Output: Issue: “none”, sentiment: “positive”, feature_mentioned: “none”, feature_requests: “dark mode”

NOW DO THIS: Input: “Customer support was amazing. Helped me recover my account in 5 minutes. The app itself is confusing to navigate but getting better.” Output:

For email responses:

TASK: Write professional but friendly email responses

Input: “When will my order ship?”
Output: “Hi! Your order is scheduled to ship tomorrow (Dec 2nd) and you’ll receive tracking info within 24 hours. Let me know if you need anything else!”

Input: “I want a refund. This product doesn’t work as advertised.”
Output: “I’m sorry to hear the product didn’t meet your expectations. I’ve initiated a full refund to your original payment method—you should see it in 3–5 business days. Would you mind sharing what didn’t work as expected? We’d love to improve.”

NOW DO THIS: Input: “Your pricing page is confusing. What’s the difference between Pro and Enterprise?” Output:

Advanced technique: Add negative examples

Show the AI both what to do AND what NOT to do:

GOOD EXAMPLE: “Thanks for reaching out! I’ll have those numbers to you by Friday.”

BAD EXAMPLE (Don’t write like this): “Per your request dated 12/15/2025, please be advised that the requested numerical data will be furnished to your attention no later than the close of business on Friday, December 20th, 2025.”

NOW WRITE: [Your task]

Why this works so well:

• Shows, don’t tell: Concrete examples are clearer than abstract descriptions
• Captures nuance: The AI picks up on subtle patterns in tone, structure, and style
• Consistent outputs: Once you have good examples, results become predictable
• Works across all AI tools: This technique works in ChatGPT, Claude, Gemini, and others

Pro tips:

• Use 2–5 examples (more isn’t always better)
• Make examples diverse enough to show the range you want
• Include clear inputs + clean outputs
• Save your best few-shot prompts as templates for reuse

Learn more about few-shot prompting • OpenAI prompt best practices

Tip 3: Build AI Workflows That Actually Scale (Stop Treating AI Like a Chatbot)

Most people use AI for one-off tasks instead of building repeatable workflows. The difference is massive—it's like using a calculator for one math problem vs. building a spreadsheet that calculates everything automatically.

The mindset shift:

Chatbot approach: Open ChatGPT → Ask question → Copy answer → Repeat tomorrow

Workflow approach: Design multi-step process → Save as template → Reuse for every similar task → Continuously improve

Real workflow example:

Content creation in 30 minutes (used to take 4 hours):

1. Use NotebookLM Deep Research to gather 50+ sources on your topic
2. Ask Claude to create an outline based on the research
3. Use ChatGPT to write first draft (Memory knows your brand voice)
4. Run through Gemini to add specific examples and data
5. Use Claude again to polish and fact-check
6. Export to your CMS

Save this as a template. Next article follows the same steps.

How to build your own workflow:

Step 1:Identify repetitive tasks What do you do multiple times per week that takes 30+ minutes and follows a similar pattern?

Step 2:Break it into AI-friendly steps Research → Synthesize → Create → Review → Finalize (each step = clear inputs/outputs)

Step 3:Choose the right tool

• ChatGPT: General tasks, shopping research, conversational work
• Claude: Long documents, code review, detailed analysis
• Gemini: Web research, Google tool integration
• NotebookLM: Document synthesis, data extraction

Step 4:Document it Create a simple checklist:

☐ Upload materials to [Tool]
☐ Run [Prompt Template A]
☐ Copy output to [Tool B]
☐ Run [Prompt Template B]
☐ Export and quality check
  

Step 5:Improve over time Track what works, refine prompts, remove steps that don’t add value.

Quick wins to implement this week:

• Use AI Memory: Set up your role and preferences once—AI remembers for all future tasks
• Create prompt library: Save your best prompts organized by task type (writing, analysis, research)
• Implement review steps: Use one AI to generate, another to critique (catches errors and improves quality)
• Batch similar tasks: Research 5 topics at once, generate 10 descriptions together, process all notes on Friday

Example: Research → Present workflow (45 min vs. 2 days)

1. NotebookLM Deep Research creates comprehensive research
2. Upload to Claude for executive summary + insights
3. ChatGPT generates 5 presentation title options
4. Feed to Canva AI to create slide deck
5. Review with Gemini for consistency

Measure your success:

Before: How long does this task take?
After: New time per task

If you're not saving at least 50% of your time, refine the workflow.

Start small:

Pick ONE repetitive task this week. Build a simple workflow for it. Document what works. Then expand to other tasks.

The goal isn’t to automate everything—it’s to automate the repetitive parts so you can focus on work that requires human judgment and creativity.

NotebookLM guide • Claude prompting • ChatGPT tips

The Bottom Line

Using AI efficiently isn't about knowing every feature—it’s about building systems that compound over time:

1. Data Tables in NotebookLM turn hours of manual research into minutes of structured analysis
2. Few-shot prompting gets you from random results to 90%+ accuracy by showing instead of telling
3. Workflow design transforms one-off tasks into repeatable systems that save hours every week

The difference between people thriving with AI and people frustrated by it? The thriving ones stopped treating AI like a magic chatbot and started building actual systems.

This week’s action plan:

• Try NotebookLM Data Tables on your next research task
• Convert one repetitive prompt into a few-shot prompt with examples
• Map out one workflow you do weekly and identify which AI can handle each step

Additional resources:

• NotebookLM complete guide
• The Prompt Report – Most comprehensive prompt engineering study
• Anthropic’s prompt library – Ready-to-use prompts
• OpenAI's best practices – Official techniques
• Lenny’s Podcast on prompt engineering – Deep dive with experts

Just before you go

Hope you enjoyed this edition of Brain Bytes. Got feedback, suggestions, or cool AI tools to add?

You can always reply directly to this email — I read everything you send.

See you later this week, — Oliver