How to research a new business idea using ChatGPT, Claude, and Vora IQ

Level

Foundational

Reading time

11 minutes

Updated

November 3, 2026

Sources

9 cited

Useful for

All business types

What AI can and can't do for your research

Researching a business idea comes down to a handful of questions. Who has the problem? How do they handle it today? What do they pay for the alternatives? Would they switch to you, and pay for it?

AI assistants help you ask those questions faster, prepare for the conversations that answer them, and organize what you learn. They are not a reliable source for the answers.

The U.S. Small Business Administration describes two kinds of market research. Existing sources save time but may not be specific to your audience. Asking customers yourself gives a more nuanced picture but costs more time and money.^[1] AI mostly speeds up the first kind and helps you prepare for the second. It can't do the second for you.

The biggest risk is a confident mistake. NIST's guidance on generative AI calls this confabulation: a system confidently presents false content, often called a hallucination. The output can even include made-up citations or reasoning that make a wrong answer look justified.^[2] Anthropic says Claude can show quotes that look authoritative but aren't grounded in fact, and that you shouldn't treat it as a single source of truth.^[3] Assume the same of every AI tool, including the ones in this guide.

Where each tool fits

Based on what each company publicly documents:

Either general assistant can handle every step below. Where we suggest one, it's because a documented feature makes it a convenient fit, not because we've measured it as better.

Vora IQ owns The Founder Almanac. We include it where it fits the work, describe it only by what its own site says it does, and point out its limits as we do for the others.

The scenario we'll follow

The research workflow, step by step

1. 01

   Write down what you need to learn

   Before you open any tool, write the idea in one sentence, name the customer as specifically as you can, and list your biggest assumptions. Then let an assistant turn that into a research plan. Tell it not to answer the questions, so you don't anchor on a guess.

   I'm researching a business idea before spending money on it.

   Idea: [one sentence]

   Customer: [who, as specifically as you can]

   My assumptions: [your three biggest assumptions]

   Turn this into a research plan. List the five questions I most need answered. For each one, tell me:

   1. Whether existing data can answer it, or only conversations with customers can
   2. Where I'd find a trustworthy answer
   3. What answer would make me change or drop the idea

   Don't answer the questions yourself, and don't give me statistics.

   Tasha's list includes: how many suite renters are within reach, how they handle bookkeeping today, what they pay for it, and whether they'd pay monthly to stop doing it themselves.

2. 02

   Scan the market, then check every number

   A market scan is where AI tools save the most time and do the most damage. Without web search, an assistant answers from what it learned in training, which may be out of date. Anthropic notes that Claude might not have the most up-to-date information in some subject areas.^[3] Anthropic documents web search in Claude for current information.^[4] If you use another tool, check whether it's searching the web or answering from memory.

   Ask for a source, a link, and a date for every number. Then open the sources. Anthropic advises reviewing Claude's cited sources, because the original pages may hold context its summary left out.^[3] For hard numbers, go to the original publisher. The SBA lists free federal sources, including the U.S. Census Bureau and the Bureau of Labor Statistics.^[1]

   Search the web and give me an overview of the market for [product or service] for [customer] in [location].

   For every number or factual claim, include the source, a link, and the publication date. If you can't find a source, write "no source found" instead of estimating.

   Keep what the sources say separate from your own interpretation.

   End with the three facts I should verify first because my plan depends on them.

   Vora IQ also describes a research teammate, Insight, for researching your market, customers, and competitors.^[5] The same rule applies to any tool: open the sources.

   When Tasha's scan returns a figure for how many stylists rent suites, she doesn't copy it into her plan. She opens the source. If she can't trace a number to its original publisher, she writes "unknown." That's a useful answer: it tells her what to ask in interviews.

3. 03

   Map customers and competitors, including the indirect ones

   Your competition is whatever customers do today, not only businesses like yours. The SBA recommends identifying competitors by product or service and by market segment, and including indirect competitors.^[1] AI tools are good at producing a long first list. Treat every item as an assumption to check.

   I'm exploring [idea] for [customer].

   List every way this customer might handle [problem] today:

   • Direct competitors that sell something similar
   • Indirect alternatives, such as software, spreadsheets, a friend or relative, or a different kind of professional
   • Doing nothing

   For each one, describe what it probably costs the customer in money and time, and where it probably falls short. Label every item as an assumption.

   Then write one interview question per alternative that asks about past behavior, not opinions.

   Tasha's list includes a general local bookkeeper, accounting software, the tax preparer stylists see once a year, a spouse who handles it, and a drawer of receipts. Each one becomes a question for her interviews.

4. 04

   Prepare for customer interviews

   Conversations with customers are where the real evidence comes from. The National Science Foundation's I-Corps program uses this kind of customer discovery to help teams quickly assess their market potential.^[6] For a full method, read how to run customer discovery interviews and use the customer interview guide.

   AI is useful here as a critic. Write your questions first, then ask it to find the weak ones.

   Here's my draft interview guide for [customer] about [problem]:

   [paste your questions]

   Flag any question that:

   • Is leading, or hints at the answer I want
   • Asks about the future ("Would you...") instead of what they've actually done
   • Asks about my idea instead of their problem

   Rewrite each flagged question. Then suggest two follow-up questions for when someone says they've already tried to solve this.

   Tasha schedules twelve 20-minute conversations with suite renters, recruited through her old salon contacts and two suite buildings.

5. 05

   Synthesize your notes, then check the synthesis against them

   Sorting a dozen sets of notes into patterns is slow by hand and fast with AI. Claude's projects let you upload files to a project's knowledge base and set instructions that apply to every chat in it.^[7] That makes them a natural place to keep interview notes.

   Remove names and identifying details before you upload anything. Then ask for counts and exact quotes, and check them. An assistant can produce a quote that sounds right and isn't in your notes.^[3]

   The files in this project are notes from [number] customer interviews about [problem]. Names have been removed; each file has an interview ID.

   1. Group what people said into recurring problems. For each, give the number of interviews where it came up and their IDs.
   2. Under each problem, add one or two short quotes copied exactly from the notes, with the interview ID.
   3. List anything that contradicts my assumption that [assumption].
   4. List the questions the interviews didn't answer.

   Don't generalize beyond these interviews. If something came up only once, say so.

   Suppose the synthesis says that 8 of 12 stylists sort receipts only once a year, that 3 already pay someone for bookkeeping, and that "most" want a monthly summary. Tasha checks every count against her notes. The first two hold up. The third doesn't: only 4 of 12 mentioned a summary at all. "Most" was the assistant's word, not her customers'.

6. 06

   Turn findings into a validation plan

   Research ends with a test, not more research. The Lean Startup approach treats validated learning as the measure of progress and tests each part of an idea with experiments.^[8] Write each risky assumption as a statement with a pass mark you set in advance.

   Tasha writes three:

   • At least 5 of the 12 stylists she interviewed, or people they refer, pay for a first month at $79.
   • A one-page site promoted in three suite buildings gets at least 25 sign-ups in three weeks.
   • She can serve one client in under two hours a month, timed across her first three clients.

   This is where Vora IQ fits. Its site says you describe your idea and get an Idea Report Card that scores it on demand, competition, money, and risk and tells you what to work on first. The report then becomes a step-by-step roadmap, and Vora IQ picks three tasks from it each morning. It also says it can build a landing page with a sign-up form and visitor tracking to test demand.^[5] Vora IQ says its evaluation gives reasoning, not just a score, so ask for the reasoning and check it against what you've learned.^[9]

   [One sentence describing your idea, for example: A flat monthly bookkeeping service for hair stylists who rent salon suites.]

   What I've learned so far: [three to five findings from your interviews, with counts]

   What I haven't confirmed: [your open questions]

   Score this idea, explain the reasoning behind each part of the score, and tell me what to test first with real customers.

   Vora IQ's own description is a fair one: the scoring is a first pass, and the next step is to test the idea with real people.^[5] A good score doesn't replace Tasha's three tests. It helps her decide which one to run first.

When combining tools makes sense

You don't need all three tools. For an early, simple idea, one general assistant can handle the whole workflow, and switching tools costs time and context. Combining them makes sense in three situations:

• As a cross-check. Paste one assistant's market summary into another and ask it to flag every claim without a source. This catches some errors cheaply, but it isn't verification. Two AI tools agreeing isn't evidence. Only the original source is.
• When you have a lot of material. A workspace that keeps files and instructions together, such as a Claude project, saves you from pasting the same context again and again.
• When research turns into execution. Once you have findings to act on, a structured system can keep the plan, the tasks, and what you've learned in one place. Vora IQ says it gives you a prompt to paste into the ChatGPT or Claude conversation where you shaped the idea. That assistant writes a short brief, which you bring into Vora IQ to build the report card and roadmap.^[5]

Your next step

Do step 1 today. Write your idea, your customer, and your three biggest assumptions, and turn them into a research plan. It takes 20 minutes, and every later step depends on it.

Then keep going:

• Follow the Validate Your Business Idea Founder Path, which this tutorial is part of, to turn your research into a written decision.
• Learn the full method in how to validate a business idea.
• Get more from each assistant with our guides to Claude and ChatGPT.
• When the tests pass, work out how to price your first product and how to explain why you're the right choice.