From Matt Gullett at Between Silicon and Soul - Part 2 of the Skills Half-Life Series

In Part 1, we talked about why great researchers struggle with new tools—it's not about capability, it's about applying old mental models to new realities. Now let's get practical: how do you actually make AI work for research instead of just getting frustrated with it?

The secret isn't better prompts. It's breaking the work into steps the machine can handle.

Why Most AI Attempts Fail

Let me show you what I mean with a real example that happens in research teams every week.

The Prompt That Doesn't Work

"Analyze these 2,000 open-ends on our new flavor. Give me themes with definitions, sentiment by theme, three quotes per theme, and a 400-word exec summary."

What you usually get:

Generic buckets like "taste," "price," "packaging" with fuzzy overlap
Quotes that all sound the same because the model didn't diversify phrasing
Sentiment that confuses intensity with volume
A summary that reads like marketing copy instead of research insights

Why it failed: You asked the model to do six jobs at once—theme extraction, definition writing, sentiment analysis, quote selection, synthesis, and executive communication—with no steps, formats, or quality checks.

It's like telling a junior analyst to "go finish the study" with no brief, no examples, and no review process.

Most disappointment with AI comes from asking for results instead of working a process. The solution is the Five-Card Method.

The Five-Card Method: Your New AI Framework

Think of this as creating a work order that any machine (or junior analyst) could follow. You're making the invisible work of research visible and structured.

Card 1: Inputs

Set up your data, rules, and context clearly.

Example:

"You are helping with verbatim analysis for a consumer product study. 

Files: 2,000 open-ended responses about a new flavor launch
Rules: 
• Don't invent counts or percentages
• Keep slang and colloquialisms; expand abbreviations
• Never include personally identifiable information
• Focus on authentic consumer language

I'll provide 3 annotated examples of the theme definitions I prefer.
[Insert your 3 examples here]

Acknowledge these inputs and wait for Step 2."

Why this works: You're establishing context, boundaries, and examples before asking for any output. The AI knows what game it's playing.

Card 2: Steps

Break the job into 6-8 smaller, sequential moves.

Example:

"We'll complete this analysis in sequential steps:

1. Extract approximately 20 distinct themes from the verbatims
2. Write a 1-sentence definition for each theme (describe the concept, not just label it)
3. Cluster related themes into 5-7 major groups and name each group
4. Score sentiment for each theme: Strongly Positive, Somewhat Positive, Neutral, Somewhat Negative, Strongly Negative
5. Score intensity for each theme: High, Medium, Low (based on language strength, not volume)
6. Select 3 representative quotes per theme, ensuring varied phrasing and perspective
7. Draft a 400-word narrative summary for stakeholders
8. Identify 3-5 core insights that weren't obvious before analysis

Confirm this approach before I provide the data."

Why this works: Each step has one job. The AI can focus, and you can quality-check each stage.

Card 3: Format

Show exactly how you want the output structured using formats you already work with.

Example:

"Present your analysis in this structured format that I can copy directly into our reports:

**THEME 1: [Theme Name]**
ID: T01
Definition: [One sentence describing the consumer concept]
Sentiment: [Strongly Positive/Somewhat Positive/Neutral/Somewhat Negative/Strongly Negative]
Intensity: [High/Medium/Low]
Representative quotes:
• "[Quote 1 - varied phrasing]"
• "[Quote 2 - different perspective]"  
• "[Quote 3 - distinct language style]"

**THEME 2: [Theme Name]**
[Repeat format for each theme...]

---
**THEME GROUPS:**
• **Taste Experience:** T01, T02, T03
• **Value Perception:** T04, T05, T06
[Continue for all groups...]

---
**NARRATIVE SUMMARY:**
[400-word stakeholder summary goes here]

---
**CORE INSIGHTS:**
1. [Insight connecting themes to business implications]
2. [Unexpected finding that wasn't obvious before analysis]
3. [Actionable recommendation based on patterns]

This format lets me copy-paste sections directly into PowerPoint or Word documents."

Why this works: The structured text format is immediately recognizable and works with any word processor or presentation tool. No technical knowledge required—just copy, paste, and format as needed.

Card 4: Quality Checks

Define what "good" looks like before you start.

Example:

"Quality standards for this analysis:

• Theme definitions must describe consumer concepts, not just restate labels
• Include both positive and negative themes if present in data
• Quote examples must vary in phrasing—avoid repetitive language
• If data is insufficient for confident analysis, state limitations clearly
• Sentiment scoring should reflect emotional tone, not just positive/negative words
• Core insights should connect themes to business implications

If any step doesn't meet these standards, flag it for revision."

Why this works: You're establishing success criteria upfront, not trying to fix problems after they're baked in.

Card 5: Next Action

Make sure the output connects to your broader workflow.

Example:

"After analysis completion:
• Export JSON to our dashboard template
• Flag any themes that need human validation
• Prepare three follow-up questions for deeper exploration
• Identify verbatims that warrant individual client attention

Confirm you understand the complete workflow before beginning analysis."

Why this works: The AI knows this isn't the end—it's part of a larger research process.

Real Results: Before and After

Traditional one-shot prompt result:

20 minutes of back-and-forth trying to get usable output
Generic themes that could apply to any product
Inconsistent formatting requiring manual cleanup
Quotes that don't actually represent the themes well

Five-card method result:

Clean, structured analysis in first attempt
Themes that reflect actual consumer language and concepts
Text format that copies directly into PowerPoint or Word
Quality consistent enough that human review focuses on insights, not cleanup

Quick Start: Your First Five-Card Prompt

Pick a simple task you do regularly—maybe cleaning up survey data or categorizing verbatims. Try the five-card method:

Inputs: Define the data, rules, and examples
Steps: Break it into 4-6 smaller tasks
Format: Show the exact output structure you want
Quality: List 3-4 standards for good work
Next: Explain how this fits your broader workflow

Start small. Master the method on low-stakes work before applying it to client deliverables.

The meta-lesson: The five-card method isn't just about AI—it's about making any complex process teachable, reviewable, and repeatable. Whether you're training a junior researcher, documenting a workflow, or prompting an AI, the same principles apply.

Coming next in Part 3: A practical 90-day plan for refreshing your team's skills without losing your sanity, plus how different generations can learn from each other in this new environment.