LLMs are helpful but won’t magically solve your problems unless you know how to use them well. You must be specific, strategic, and not just accept the first answer. Here are some ways to make these AI models work for you instead of giving you generic fluff.

1. Find hidden assumptions

Every idea has blind spots. LLMs help surface them.

✅ “What am I missing here?”

✅ “What assumptions am I making that could be wrong?”

✅ “What’s a macro event that could totally reverse the outcomes of this test?”

You can even push it further:

❌ “List assumptions we’re making about the user, their context, and their motivations.”

❌ “Now go deeper—list 20 more.”

❌ “Out of these, what are the top 3 most critical ones, and how do we test them?”

Perfect for reality-checking your strategy.

2. Give the LLM a job title

LLMs love roleplay ;). The more specific the role, the better the response.

✅ “Act as the Head of Growth at Duolingo. Design an experiment to measure the impact of daily notifications.”

✅ “Pretend you’re a CTO reviewing this PRD—list out critical but constructive feedback.”

PRD: Product Requirements Document

It’s like asking a friend for advice vs. asking your friend who’s a CFO for financial advice. The difference is huge.

3. Diagnose problems like you know what you’re doing

LLMs are great for debugging ideas, not just code.

✅ “We launched a feature and adoption is low. List the top 5 likely root causes and how we can validate each.”

✅ “Users drop off at step 3 of onboarding. What are possible reasons?”

Great for avoiding “we have no idea why this isn’t working” meetings.

I definitely know what I’m doing

4. Stress-test your ideas

LLMs are great at playing devil’s advocate.

✅ “Tell me 5 reasons this feature won’t work as intended.”

✅ “Tell me 5 unintended consequences of this feature.”

✅ “Why shouldn’t we do this?”

This is where LLMs are really helpful. They give you insights that you might’ve never thought of.

5. Get smarter about meetings

Before a meeting: ✅ “I’m meeting the Head of X to discuss Y. What should I ask her?”

After a meeting: ✅ “Summarize my notes into action items.”

Bonus: Use something like granola.ai to record, transcribe and generate notes for you.

Another big big helpful use-case for LLMs.

6. Ask for the right kind of output

Tell it what you want first, then let it figure out how to deliver.

✅ “I want to clearly communicate context, outcomes, and success metrics to my dev team. Can you generate a detailed prompt I can use with you to write this PRD?”

Then you use that prompt output to start the conversation again! Recursive LLM

7. Use it to iterate quickly

Let it generate variations for copy, ads, emails, and UI text so you can choose the best one.

✅ “Rewrite this push notification to be clearer and more engaging.”

✅ “Give me 5 variations of this email subject line.”

Good for when you need ideas, not just answers.

8. Frame decisions clearly

Complex trade-offs? Let it help structure your thinking.

✅ “I must choose between launching Feature A or Feature B this quarter. Outline clear pros and cons for each.”

✅ “Give me a decision-making framework for whether to build internally vs. buy from a vendor.”

Whenever you’re undecided about something. Try to leverage an LLM.

9. Validate messaging before you ship

Use AI to test different angles before spending time and money on real user tests.

✅ “Give me 3 different messaging angles to position our premium subscription for freelancers.”

✅ “Which objections might users have to this messaging?”

10. Extract insights from messy data

Got raw app store reviews or usage metrics? Let the LLM do the heavy lifting.

✅ “I pasted 1,000 reviews—find the top 3 most requested features and complaints.”

✅ “Here’s some usage data. What’s the conversion rate? What’s an insight I might be missing?”

Data in, insights out. Noice.

In essence

LLMs work best when:

  1. You tell them who they are.
  2. You tell them what outcome you want.
  3. You use them to generate ideas and iterate, not just answer questions.
  4. You push them to surface hidden assumptions and unexpected insights.

They’re great for brainstorming, summarizing, and stress-testing your thinking. But it’s still up to you to make the final call.