How to compress a month of GTM research into three hours

Use AI to find a market's hidden assumptions, pressure-test your idea, and prepare for better customer conversations

“Research the market” sounds like a sensible prompt. It is also vague enough to produce exactly the kind of answer you could have written yourself: a market summary, a list of familiar competitors, a few trends, and a neat SWOT table which says almost nothing.

The problem is not that Claude, ChatGPT, or another capable model cannot help with go-to-market research. The problem is that you have asked it to reason before giving it enough evidence to reason from.

A more useful founder workflow is to treat AI like a tireless research partner. First build a small but information-dense picture of the market. Then use the model to compare sources, notice patterns, expose assumptions, and attack your first conclusion. Done well, a few hours of work can replace weeks of unfocused browsing.

Start with a market dossier, not a prompt

Before asking for conclusions, collect the material that contains the market's real language and incentives. A strong starting dossier might include:

• 30 or more fully indexed competitor websites
• 10 or more earnings call transcripts from large incumbents
• 100 customer reviews across several competitors
• 10 complaint-heavy Reddit or community threads, if the market has them

Each source answers a different question. Competitor websites show how the market sells. Earnings calls show what established companies worry about when speaking to investors. Reviews reveal what customers praise after buying. Community complaints show the frustration that never makes it into a polished testimonial.

Do not only collect pages from the obvious market leader. Include expensive and cheap products, specialists and generalists, beloved tools and badly reviewed ones. The disagreements between them are often more useful than the consensus.

Ask for what customers do not say out loud

Once the material is indexed and available in the model's context, the first useful question is not “What are the main trends?” It is:

What does every successful player in this market understand that customers never say out loud?

This asks the model to look for revealed behaviour rather than repeat stated preferences. Customers may say they want more features while repeatedly choosing the product that feels safest. They may complain about price but remain with an incumbent because migration is worse. Every competitor may sell speed while quietly investing in onboarding and reassurance.

These are not automatically truths. They are patterns worth investigating. Ask the model to cite the evidence behind each one, distinguish strong signals from guesses, and show which sources disagree. That makes the output useful for your next decision instead of merely persuasive.

Find the assumptions holding the market together

The second question turns consensus into possible opportunity:

What are the assumptions this entire market is built on, and what would have to be true for each one to be wrong?

Mature markets accumulate invisible rules: customers need a dashboard, implementation requires a consultant, buyers want an annual contract, the product must serve the whole team, or the best acquisition channel is paid search. Some of these rules are real constraints. Others are habits inherited from an earlier version of the market.

This exercise gives you three useful maps: what everyone agrees on, where incumbents may be blind, and which customer-relevant ideas are being ignored. That is a far better foundation for positioning than choosing three generic adjectives for your landing page.

Make the AI attack your favourite idea

Research becomes dangerous when it only helps you explain why you are right. By this point you will probably have a promising angle. Resist the temptation to turn it straight into a strategy. Ask the model to behave like a great investor who wants the strongest case before looking for the fracture:

What's the strongest version of this argument, and where does it still break?

Follow the answer into specifics. What customer must exist for this idea to work? What behaviour must change? Which incumbent could copy it? What distribution advantage is required? What evidence would make you abandon the idea? Ask for the cheapest test of every important assumption.

The model has no ego about being wrong. Use that. Make it argue both sides, revise its conclusion when the evidence changes, and label uncertainty clearly. Its job is not to make your idea sound clever. Its job is to help you arrive at the market with sharper questions.

A three-hour research sprint

1. Build the dossier. Gather representative sources, remove duplicates, and organise them by competitor, customer type, and source type.
2. Map the market's hidden knowledge. Ask what winners understand, what customers reveal through behaviour, and where the sources contradict one another.
3. Challenge the consensus. List the market's shared assumptions and the conditions under which each one stops being true.
4. Form an initial GTM hypothesis. Choose a narrow customer, urgent problem, credible promise, and practical first channel.
5. Try to kill it. Steelman the opportunity, find where it breaks, and turn the biggest uncertainties into customer interview questions and small experiments.

Three hours gets you to the market, not around it

At the end of this process, you may have a very initial GTM strategy that reads like it came from someone who has spent years in the category. That is useful leverage, but it is not the same as having spent those years there.

The model has read the traces customers left behind. It has not watched someone struggle through a workflow, heard the hesitation before a buying decision, or learned which “must-have” problem is never urgent enough to fix. Only real conversations and behaviour can validate that.

So use AI to develop the hypothesis, then go out into the market. Talk to customers. Test the language. Show the idea. Notice where reality disagrees, update the dossier, and run the loop again. If you want help turning those early lessons into repeatable distribution work, OctoLoops can help you keep the loop moving.