← Back to all free guides
A free guide by Mariah Brunner ↗ INSTAGRAM ↗ TIKTOK

Stop Using AI
To Generate
Business Ideas

AI won’t hand you a winning business idea, no matter how clever your prompt is. Used as research, it will surface real gaps, check the demand and gut your idea before you waste six months on it. Here’s the 3-step workflow.

Every founder thread is some version of "I asked AI for the best business ideas and here’s the list." The lists are generic because you’re asking a model trained on the internet for an idea that isn’t on the internet yet. You get the average of every "best business ideas" blog post.

The smarter move is to use AI as research. Find real pain, check that people want it fixed, then stress-test your version before you build anything. It’s the same playbook in any industry.

Step 1: Find Complaints, Not Ideas

I use Perplexity for this because it’s built around searching the live web and citing every source, so you can check what it tells you. Claude with web search turned on works too. Whatever you use, make sure it’s actually searching and not answering from memory.

The reframe that matters is to ask for complaints. Pain points are where businesses live, and pain you can see, you can build for.

Find The Complaints
You are a market researcher. Your job is to find real, recurring complaints, not business ideas. The space I'm researching: [a specific industry or customer group you actually care about, e.g. "home renovation contracts for first-time buyers" or "meal prep delivery for people with food allergies"]. Search the live web, focusing on the last 12 months of Reddit threads, review sites like Trustpilot and app store reviews, forums and Facebook or community posts you can access. Then give me: 1. THE TOP 10 COMPLAINTS, ranked by how often you saw them. For each one: - The complaint in one plain sentence - 1 or 2 short quotes in people's own words, each with a link - Roughly how often it came up (for example "in 6 of the 20 threads I read") 2. THE WORDS PEOPLE USE: the exact frustration phrases that keep repeating. I'll use these later in my marketing. 3. 3 UNDERSERVED GROUPS: the specific types of people inside this space whose problems current options ignore, and why. RULES: - Every quote needs a working link. If you can't find a real quote, say so. Never write a quote yourself. - No ideas, no solutions, no summaries of "the market." I want the raw pain. - If the space is too broad to research well, tell me and suggest 3 narrower versions.

Specific beats broad every time. "Home renovation contracts for first-time buyers" is researchable. "AI businesses" is not.

Step 2: Check People Want It Fixed

A complaint isn’t enough. Plenty of things annoy people who never bother to fix them. You need to know whether people are actively searching for a fix.

If lots of people are searching and every current option frustrates them, that’s the gap. Take your strongest complaint from Step 1 and run this:

Validate The Demand
You are a skeptical market analyst. I have a pain point and I need to know if people are actively trying to fix it, not just complaining about it. The pain point: [paste the single strongest complaint from Step 1, with 2 or 3 of the quotes]. Who has it: [the group of people]. Research the live web and answer with sources: 1. ARE PEOPLE LOOKING FOR A FIX? Find "how do I fix this" style threads and questions from the last 12 months, and check whether Google Trends shows interest in related searches. Give real counts or trends only if you can see them, with links. If you can't measure it, say so. 2. WHAT ARE THEY USING NOW? The top 5 products, services or DIY workarounds people use today, with the price of each if listed and the most common complaint about each (with a link). 3. WHY ISN'T IT GOOD ENOUGH? Quotes from real users explaining what's broken about the current options. 4. IS IT GROWING? Is interest rising, falling or flat over the last 2 years? Cite what you're basing that on. 5. VERDICT: rate demand STRONG, MIXED or WEAK in one line, then give the 2 strongest pieces of evidence for your rating and the 1 biggest piece of evidence against it. RULES: - Never invent search volumes, market sizes or quotes. "I couldn't find this" is a useful answer. - Be blunt. If demand looks weak, tell me.
Step 3: Have Claude Gut Your Idea

Most founders stop after validating demand and start building. That’s the mistake. The step that saves six months is taking your actual solution to Claude and telling it to tear it apart.

AI models tend to agree with you, so you have to explicitly tell Claude not to be nice. This prompt does that. (For big decisions, my LLM Council prompt takes it further.)

Stress-Test The Idea
You are a blunt, experienced investor and operator. I'm about to spend real time and money on a business idea. Your job is to make me drop it if it deserves to be dropped, or rebuild it if it can be saved. Do not be nice. Do not soften. Do not validate me to make me feel good. MY IDEA: - What it does: [one or two sentences] - Who it's for: [the specific customer] - How it makes money: [pricing and who pays] - Why it's different: [what current options get wrong] MY RESEARCH (from Steps 1 and 2): [paste the top complaints, the current options and the demand verdict] Do this in order: 1. 10 REASONS IT FAILS, ranked most likely first. Be specific about what breaks, when and why. 2. THE CUSTOMER WHO NEVER BUYS: become the person in my target market who says no. Why exactly? What would they use instead? 3. MY 3 BIGGEST UNTESTED ASSUMPTIONS, and for each one a way to test it for $0 in the next 7 days, with what result would mean "keep going" and what would mean "stop." 4. THE BET: with $10,000 of your own money, would you fund this, yes or no? Two sentences why. 5. THE BETTER FRAME: if the idea is weak, say so plainly. If there's a real seed built on the wrong frame, tell me the better version. RULES: - Use my research. If something I claimed isn't backed by it, call it out. - No hedging and no "it depends" without saying what it depends on. - Keep it tight. Headings and short bullets.

The Loop

Pain, then demand, then stress test. If your idea survives all three and Claude can’t talk you out of it, you have something real, so go build it. If it doesn’t survive, you just saved yourself six months. Comment IDEA on the original post and I’ll send you the 3 prompts as a single Project file you can drop into Claude.