Getting More from AI Tools: 4 Lessons from 18 Months of Daily Use
👋 Welcome to New Vintage, a weekly taste of tech for wine professionals to level up with no-code and AI tools.
OVER THE PAST 18 months, I've spent countless hours exploring how AI tools can help with day-to-day work. Through all this experimentation, I've discovered patterns and approaches that consistently lead to better results. I'm sharing these insights to help you get more value from tools like ChatGPT and Claude, whether you're just starting out or looking to level up your usage.
Quick Context
When we talk about AI tools in this guide, we're specifically referring to chat interfaces like ChatGPT and Claude that can help with writing, analysis, and decision-making. These tools use Large Language Models (LLMs) that have been trained on vast amounts of data to engage in helpful dialogue. Think of them as collaborative partners who can help think through problems or draft content. While we won't cover them today, just know that "Generative AI" also includes other forms of output like audio, images, and video.
Here are the four approaches I've found most valuable in my daily work with these tools.
1. Embrace the Back and Forth
The Concept
If you search for "Prompt Engineering", you'll find lots of guides on how to craft a really long, thoughtful prompt. This absolutely has merit, but I find that for tasks that aren't well-defined, it's better to keep it simple. Instead of trying to get perfect output in a single prompt, treat your interaction like a flowing conversation where each response informs your next question. This approach is particularly powerful for content creation and analysis tasks.
What Not to Do
Write a comprehensive wine club newsletter that includes our latest releases, tasting notes for each wine, food pairing suggestions, winemaking process details, and upcoming events. Make it engaging and personal, with proper formatting and examples throughout.
Better Approach:
-
Start with a lot of Ideation
What are 25 unique angles we could use to present our latest Pinot Noir release to wine club members? -
Develop the Outline
Based on angle #13 (the vineyard's microclimate story), create an outline that connects weather patterns to flavor development. -
Section by Section Development
Let's expand the section about how the coastal fog influenced this vintage. Include specific details about timing and temperature. -
Review and Refinement
Review this section for technical accuracy. What questions might a curious wine club member have that we haven't addressed?
Key Applications
- Developing wine club communications
- Creating educational content for staff training
- Drafting social media calendars
- Structuring customer feedback analysis
Pro Tip
Save successful conversation flows as templates. For example, keep a "Blog Post Development" template with your preferred sequence of prompts. This helps you replicate successful outcomes and save time on future similar projects.
2. Paint the Picture
The Concept
Instead of giving abstract instructions, frame your request through specific scenarios that provide rich context for the AI to work with. Just like our colleagues benefit from context, you can get more relevant responses from AI when it understands more about your situation.
What Not to Do
Give me customer service best practices for handling wine club complaints.
Better Approach
I'm the tasting room manager at a small winery in Sonoma. A wine club member just reported that two bottles from their latest shipment were corked. They've been members for 3 years and this is their first complaint. Our standard policy is to replace corked bottles, but they're also asking for shipping compensation since this is their second delivery attempt. How should I handle this to maintain their loyalty?
Key Applications
- Handling customer service scenarios
- Developing staff training materials
- Creating marketing strategies
- Planning events and experiences
- Analyzing market opportunities
Pro Tip
For one-off scenarios, use this template to ensure you're providing all relevant context:
I am [role/position] working on [specific task/project].
My goal is to [desired outcome].
My main challenge is [obstacle/problem].
My audience/users are [description].
Key constraints are [limitations/boundaries].
Your task is to [insert task].
Your task steps are [task steps].
Structure your response like: [example of output, if you have one].
Even better: Use platform-specific features to set persistent context that carries across all your conversations:
- ChatGPT Custom Instructions: Set up your business context, role, typical audience, and preferred output style in your custom instructions. This information will automatically inform all your conversations without needing to repeat it in each prompt.
I manage DTC sales at a boutique winery in Sonoma Target audience: Wine club members and tasting room visitors Brand voice: Educational, approachable, focused on craft How I'd like you to respond: - Use accurate wine terminology - Consider both DTC and tasting room contexts - Keep luxury hospitality standards in mind
You can access "custom instructions" in the user menu:
- Claude Project Instructions: Use Claude's project feature to set up context for specific long-term projects or recurring tasks. This ensures consistent context while allowing you to maintain different settings for different projects.
Project: Wine Club Communications Context: Boutique winery, focus on small-lot wines Audience: Engaged wine club members, varying expertise levels Brand values: Authenticity, education, connection to place Key constraints: Small team, seasonal communication needs
This approach saves time and ensures consistency across conversations while allowing you to maintain different contexts for different projects or aspects of your work.
3. AI as Your Assistant
The Concept
AI excels at taking messy, unstructured content (meeting notes, brain dumps, random ideas) and transforming it into well-organized, coherent content. This approach leverages AI's ability to identify patterns, fill gaps in logic, and create structure while preserving your original thinking.
What Not to Do
[Starting from scratch]
Write me a plan for our harvest party event.
Better Approach
Start with your raw thoughts, no matter how messy:
Here are my rough notes from our harvest planning meeting:
- thinking late Sept but weather dependent
- want to show off new sorting table
- maybe do small group tours this year instead of one big group
- food? last year's caterer was too expensive
- club members asking about bringing friends
- need more parking than last year
- maybe partner with neighboring winery?
- definitely need better signage
- photographers?
Can you help organize this into a structured event plan? What key areas are we missing?
Key Applications
- Creating a coherent plan from a brainstorming session transcript
- Organizing scattered email threads about an upcoming event into an action plan
- Consolidating customer feedback from various channels (social, email, tasting room) into clear themes
- Turning a rambling voice memo about a new wine club idea into a structured proposal
- Saving time on figuring out "next steps"
Pro Tip
Record important meetings, brainstorms, or your stream of consciousness and upload these transcripts to ChatGPT or Claude to extract the key points. At the time of writing this, ChatGPT's advanced voice mode is a killer feature and you should use it on the mobile app.
4. Deepen Your Thinking, Faster
The Concept
While organizing thoughts is valuable, AI can be even more powerful as an analytical partner - helping you explore different angles, challenge assumptions, and make better decisions. Instead of just asking for answers, use AI to deepen your thinking and catch potential issues early.
What Not to Do
What's the best wine club structure for our winery?
Better Approach
Use structured frameworks to explore decisions thoroughly:
-
Perspective Analysis Template
Analyze [situation/decision] from these perspectives: 1. [Stakeholder 1] 2. [Stakeholder 2] 3. [Stakeholder 3] For each perspective, consider: - Primary concerns - Potential objections - Success criteria -
Assumption Testing Template
I'm planning to [action/decision]. My key assumptions are: 1. [Assumption 1] 2. [Assumption 2] 3. [Assumption 3] Please: 1. Identify potential flaws in these assumptions 2. Suggest tests to validate each 3. Propose alternative scenarios to consider -
Decision Framework Template
I'm deciding between: Option A: [description] Option B: [description] My criteria are: [list criteria] Please help me: 1. Evaluate pros and cons 2. Identify hidden tradeoffs 3. Consider long-term implications 4. Suggest hybrid solutions
Why This Works
- Forces structured thinking about complex decisions
- Helps identify blind spots and assumptions
- Provides a framework for thorough analysis
- Creates documentation of your decision-making process
Pro Tip
I find it helpful to evaluate decisions through the lens of specific customer personas. For example:
Before we analyze this wine club structure, here's our core member persona:
Michael, Wine Enthusiast
- 45-year-old finance professional
- Collects wine but still learning
- Values exclusive access and education
- Visits wine country 2-3 times per year
- Price sensitive but will invest in special bottles
- Often shares wine experiences on social media
Please analyze our proposal from Michael's perspective:
1. What would immediately appeal to him?
2. What concerns might he have?
3. How does this align with his wine journey?
4. What's missing that he would expect?
This approach transforms generic feedback into insights specifically tailored to your target audience. You can create similar templates for different personas, brand values, or even competitor perspectives to get a comprehensive view of your decisions. Feel free to use these as documents that are uploaded into a chat.
Bonus points if you have these as documents that can be uploaded to the knowledge base of a Custom GPT, assistant, or Claude Project.
Putting It All Together
The key to getting more from AI chat experiences is approaching these tools as collaborative partners rather than magic answer machines. Each of these tips builds on the others:
- Embrace the back and forth - for complex tasks, start simple and build depth through conversation
- Paint the picture - if you want AI to give you more relevant answers, give it relevant information
- AI as your assistant - remember that AI is really good at taking unstructured walls of text or disjointed bullet points and organizing it for you
- Deepen your thinking, faster - LLMs have been trained on the internet, tell it which perspective or information to access, so you can get more specific feedback
As you get comfortable with these basics, you'll develop an intuition for when and how to combine these techniques for even better results.
Additional reading
Beginner
- Generative AI for Everyone (Coursera)
Intermediate
- Study up on other techniques like Retrieval Augmented Generation (RAG) to further improve your output
- Incorporate API calls to different models within a no-code workflow to iterate quickly before building an application
- If you've never used Make.com, here's a 5 minute guide to get started
Advanced
- Evaluate whether fine-tuning makes sense for your use case
- Get smart on agents, these are the future
- Read Situational Awareness, by Leopold Aschenbrenner
Cheers,
Stephen
Comments ()