You don’t need better prompts. You need a better conversation.
A practical guide for smart people who still don’t get much out of AI — and the VOICE framework to change that.
Last month, I watched a brilliant product manager — someone who'd shipped features used by millions — spend 20 minutes trying to get ChatGPT to write a simple user story. She kept tweaking her prompt, adding more specifics, getting increasingly frustrated. Finally, she gave up and wrote it herself in 5 minutes.
"I thought AI was supposed to save time," she muttered.
This scene plays out in offices everywhere. Smart people, accomplished professionals, founders who've raised millions — all getting mediocre results from tools that supposedly revolutionize knowledge work. And I couldn't help but wonder: what if we're all using it wrong?
Research on How AI Makes Us Smarter and Dumber
Here's what sent me down this rabbit hole: A 2025 study from Microsoft Research and Carnegie Mellon found that knowledge workers with high confidence in AI showed less critical thinking when evaluating AI outputs compared to those who trusted their own skills more. Let that sink in. The more we trust AI, the lazier our brains become.
The researchers have a term for it: "cognitive offloading." It's what happens when you delegate mental tasks to external tools, and it's been linked to a 68.9% increase in what they politely call "laziness" and a 27.7% decrease in decision-making abilities.
But here's the plot twist: the same body of research shows AI can actually enhance cognition when used differently. A study from Surrey's Centre for Translation Studies found that specific types of human-AI interaction improved working memory and task-switching abilities.
The difference? How you engage with it.
My Own Journey
I discovered this the hard way. As Welltory's founder, I face problems that span disciplines I never studied. Tuesday it's behavioral psychology for user retention. Wednesday it's regulatory compliance for health data. Thursday it's investor relations. Friday it's team dynamics.
I used to approach AI like everyone else — looking for quick answers. "Write me a pitch deck outline." "Summarize this research." "Generate ideas for improving day-7 retention."
The results were... fine. Generic. The kind of stuff you'd get from a mediocre consultant who'd googled your industry five minutes before the meeting.
The Mental Shift That Changes Everything
Most people approach AI wrong because they misunderstand what it is. Here's what finally clicked for me:
AI is a bridge between your not-knowing and the world's knowledge
When we pretend we know everything, we cut off our inner voice — that part that wonders, doubts, and seeks. AI lets you give voice to that uncertainty and connect it with vast pools of information you'd never access otherwise.
It's not weaker or stronger — it's different
Yes, AI knows less about your specific context and goals. But it knows vastly more about the world, can process information without fatigue, and has no emotional investment in being right. These aren't weaknesses to work around — they're complementary strengths to leverage.
Talk to it like a thinking human, not a search engine
You don’t have to know special syntax or perfect prompts. Say "hmm, that doesn't sound right" or "wait, I meant something else" or "let me think out loud here." Human interaction works.
It's a dialogue facilitator, not an answer machine
The magic isn't in the answers AI gives you. It's in how explaining your thinking to AI forces you to structure it. Like therapy, the act of articulation is where insights emerge.
Without the research phase, AI makes you dumber
If you just ask for quick answers, you're outsourcing your thinking. But if you use AI to explore, investigate, and synthesize, you're expanding your own capacity. The difference is whether you're having a conversation or placing an order.
You need intention, not clarity
You don't have to know what you want. You just need to care about finding out. That intention — that sense of "something here matters" — is enough to begin.
When you make this mental shift, AI transforms from a shortcut tool into what I call a "cognitive interface" — a way to have faster, deeper dialogue between what you don't know and what the world knows. That's where the real power lies.
The Science of Thinking Together
This approach has a scientific basis. Researchers call it "metacognitive prompting" — using AI to enhance rather than replace your thinking process. Instead of cognitive offloading (bad), you get cognitive augmentation (good).
The key difference:
Cognitive offloading: "AI, write me a strategy document"
Cognitive augmentation: "I'm thinking about our strategy. Here's my current view... what am I not considering?"
A fascinating study on AI and scientific research found that papers using AI as a thinking partner (not just a writing tool) had significantly higher citation rates. Why? Because the researchers used AI to challenge assumptions and explore angles they wouldn't have considered alone.
The VOICE Method I Use
After months of experimentation, I developed a framework that consistently produces those "aha" moments. I call it VOICE — not just because everything nowadays is a framework, but because it literally gives voice to thoughts you didn't know you had.
V — Vibe (The Honest Starting Point)
Forget perfect prompts. Start with what's actually in your head:
"Something feels off about our pricing model"
"I'm stuck on this team restructuring"
"I have a hunch about our market position but can't articulate it"
Research shows that when we try to formulate perfect questions, we actually limit our thinking. Starting with uncertainty opens more neural pathways.
O — Orientation (Setting the Exploration Mode)
Don't tell AI what to think. Tell it how to think with you:
"Help me explore this from different angles"
"Let's think through this like a behavioral economist would"
"I need to understand what I'm not seeing"
This triggers what researchers call "perspective-taking," which enhances creative problem-solving.
Our goal here is to get a problem statement or a goal statement or a challenge description. Make AI do this for you and correct it until it's ok. Use its ability to pretend a VC, experienced CPO, Marty Cagan, or anybody you need as a coach now.
A problem well stated is a problem half solved.
ChatGPT is the best now for this stage, because it works well with voice and conversation.
I — Investigation (The Part Most People Skip)
Here's where I part ways with every AI guide I've read. After initial exploration, I ask: "What research should I look at to understand this better? Give me three specific research prompts for Perplexity."
Then I copy them to Perplexity. Usually, I dig into:
Scientific studies on the psychology/behavior involved
Competitor approaches and case studies
Expert opinions and contrarian views
Why? Because AI trained on internet data will give you internet-average answers. But when you feed it specific, high-quality research, it can help you synthesize insights nobody else has.
Perplexity is the best now for this. Use deep research, work in one thread, follow relevant additional questions, and don't hesitate to get more interesting angles. Bookmark this thread when you are done.
The most important part here is to read it. Take your time to read the most interesting and relevant cited sources. That's the most important part when you are extending your reality and worldview.
If it's difficult or some sources are too big, use ChatGPT to read and summarize it for you, extracting the most important and relevant thoughts from URLs, PDFs, and other sources. You can even read sources in different languages - easier than ever before.
At the end, export the whole thread from Perplexity to markdown - that is your source of proofs and insights with URLs. Post it back to ChatGPT to start working on solution for your problem.
C — Clarify and Critique (The Uncomfortable Mirror)
Ask ChatGPT to find a solution. Or use Claude to do this (because it takes more context). You can paste your problem statement from ChatGPT and research from Perplexity. You will get something much better than a mediocre AI answer. But we are not finished yet. Now ask AI to tear apart his and your thinking:
"What's wrong with this approach?"
"What would someone who disagrees say?"
"Where are the logical flaws?"
Studies show we're terrible at self-criticism due to confirmation bias. But AI has no ego. It can show you your blind spots without the emotional baggage.
Don't stop at the first insight. Push further:
"What systemic issue does this reveal?"
"How does this connect to our other challenges?"
"What would have to be true for the opposite to work?"
This is where breakthroughs happen — in the space between your initial question and where you end up.
E — Extraction (Beyond the Obvious)
This is where AI shines. Have it rewrite your solution for different audiences:
"Now as a technical spec"
"As a business case with ROI"
"As a user story"
"As a sales one-pager"
We all have blind spots. I think like a founder, my CTO thinks in systems, my designer in user flows. But AI can instantly switch lenses, research shows considering multiple stakeholder perspectives improves solution quality by 40%.
Example: For our onboarding fix, AI generated four versions. The technical spec revealed dependencies I'd missed. The user story showed emotional friction points. The business case forced me to quantify impact. Same solution, four lenses — each revealing what I couldn't see alone.
I'm Living Through My Most Intellectually Intense Period — and Growing Faster Than Ever
Take this very article you're reading. I didn't sit down to write it. I spoke my thoughts into ChatGPT during my morning walk, rambling about how smart people misuse AI. Then I asked it to find relevant research on cognitive offloading and enhancement. Then I refined the draft based on what the studies actually revealed.
The same method works for everything. Product decisions? I share the chat with my team — they see not just the conclusion but the entire thinking journey. Strategic planning? Export to Confluence. The raw dialogue often beats any polished document.
Sometimes the research rabbit hole goes so deep that I never publish what I started. (That's why my blog has gaps — not writer's block, but getting so fascinated by what I discover that the original article becomes irrelevant.) Last month I started writing about investor psychology and ended up deep in neuroscience papers about threatened egotism. Never published it, but the insights transformed how I approach board meetings.
Here's the shift: I've never before encountered such a volume of important, new-to-me information as I do now. Before, my growth was limited by how many books I could read in a year. Now it's only limited by my capacity to absorb.
The bottleneck isn't finding information anymore. It's having the courage to follow where it leads.
The Cognitive Gym Effect
Here's what the research convinced me of: using AI this way is like going to a cognitive gym. You're not outsourcing thinking — you're resistance training for your brain.
Studies on expertise show that masters in any field got there through deliberate practice with feedback. VOICE creates a feedback loop on steroids. You can explore ideas, get challenged, research, synthesize, and iterate faster than ever before.
But — and this is crucial — only if you do the work. The moment you treat AI as a shortcut rather than a thinking partner, you're in cognitive offloading territory. Your brain gets flabby. Your insights get generic.
Most people won't use AI this way. It's harder than asking for quick answers. It requires admitting you don't know things. It means sitting with uncertainty instead of grabbing the first plausible response.
But for those willing to embrace the discomfort? You get something remarkable: a tireless thinking partner who knows more than you, has no ego, never gets tired, and can help you see around corners.
The research is clear: AI will make you smarter or dumber depending on how you use it. The choice is yours.
Have a Challenge AI Might Actually Help With?
Pick something that's been nagging at you. Something complex. Something you've been avoiding because you don't know where to start.
Open ChatGPT and say: "I'm struggling with [thing]. I don't even know what my real question is. Help me figure out what I should be thinking about."
Then follow the VOICE framework. Give yourself 1-2 hours. Don't aim for answers — aim for better questions. In my experience, the best insights come not from AI, but from humans brave enough to think alongside it.
Further Reading on Cognitive Enhancement with AI:
VOICE Framework Checklist
How to use AI for thinking, not just answering
Before You Start
Block 1-2 hours of uninterrupted time
Have accounts ready: ChatGPT (best for voice), Perplexity (for research), Claude (for synthesis)
Pick a real problem you're facing — something complex, not a simple task
Accept that you don't need a perfect question to begin
V — Vibe: Start with Your Real Thoughts
What to say:
"I have this feeling that [describe the situation]..."
"Something's been bothering me about [topic]..."
"I can't quite articulate it, but [your hunch]..."
"Help me think through what I might be missing with [challenge]..."
Example starts:
"I feel like our user onboarding is broken but I can't say why"
"My team seems disengaged but all the metrics look fine"
"Our pricing doesn't feel right but I've analyzed it to death"
Tools:
ChatGPT with voice input (mobile or desktop) works best for this stage
Success marker:
You've rambled for 5-10 minutes and AI has reflected back what it heard
O — Orientation: Set the Thinking Mode
Ask AI to help you explore:
"Let's think through this like [specific expert] would"
"Help me see this from different angles"
"What questions should I be asking myself?"
"Act as a [role] and help me understand what I'm not seeing"
Useful roles to invoke:
Behavioral economist
Your target user
A skeptical board member
Someone from a different culture
Your competitor's CEO
Key phrases:
"Help me formulate a clear problem statement"
"What would need to be true for this to be the right approach?"
"What assumptions am I making?"
Success marker:
You have a clear problem statement that feels more precise than when you started
I — Investigation: Go Deep with Research
Ask ChatGPT:
"Based on our discussion, what are the 3 most important things I should research? Give me specific prompts for Perplexity, one paragraph each."
In Perplexity:
Use "Deep Research" mode
Work in one thread to build context
Follow the "Related" questions that intrigue you
Click through to actual sources (this is crucial!)
Look for:
Scientific studies on the human behavior involved
Case studies from other companies/industries
Contrarian viewpoints and critics
Expert opinions and frameworks
If sources are complex:
Copy URLs to ChatGPT and ask: "Summarize the key insights relevant to my problem"
When done:
Export the Perplexity thread to Markdown
Save/bookmark the thread for later reference
Success marker:
You've discovered at least 3 things you didn't know before that change how you see the problem
C — Clarify and Critique: Challenge Everything
Paste your research back to ChatGPT/Claude with:
"Here's my problem [paste]. Here's what I learned [paste research]. Help me develop a solution."
Then critique ruthlessly:
"What's wrong with this approach?"
"What would someone who completely disagrees say?"
"What are we not considering?"
"Where are the logical flaws?"
"What would have to be true for this to fail?"
Push deeper:
"What systemic issue does this reveal?"
"How does this connect to our other challenges?"
"What if we did the exact opposite?"
"What would [specific competitor] do?"
Success marker:
You've identified at least 2 major flaws or blindspots in your initial thinking
E — Extract: Generate Multiple Perspectives
Ask AI to rewrite your solution as:
A technical implementation spec
A business case with ROI projections
A user story from the customer's perspective
An email to your team explaining the change
A one-page brief for the board
A FAQ for sales/support teams
For each version, ask:
"What new issues does this perspective reveal?"
Final step options:
Share the entire chat thread with your team
Create a fresh chat and regenerate a clean version
Export to Confluence/Notion/Docs
Create action items from each perspective
Success marker:
Each perspective reveals something you hadn't considered
Quality Checks
Ask yourself:
Did I learn something that surprised me?
Do I understand the problem differently than when I started?
Could I explain this to someone who knows nothing about it?
Do I have concrete next steps, not just ideas?
Am I addressing root causes, not just symptoms?
Common Pitfalls to Avoid
Starting with a solution in mind → Stay in question mode longer
Skipping the research phase → This is where breakthrough insights come from
Accepting the first answer → Always push for "what else?"
Being too polite with AI → Interrupt, redirect, challenge
Sanitizing for the final output → Raw thinking often has more value
Time Allocation (2 hours total)
Vibe: 20 minutes (rambling is productive!)
Orientation: 20 minutes (getting the right question matters)
Investigation: 45 minutes (this is where the magic happens)
Clarify & Critique: 20 minutes (be ruthless)
Extract: 15 minutes (let AI do the heavy lifting)
Remember
You're not using AI to skip thinking. You're using it to think harder, deeper, and from more angles than you could alone. The goal isn't efficiency — it's insight.
The moment this feels like a shortcut, you're doing it wrong.
Hope it helps!