Market and TAM Research with AI (Without the Hallucinated Numbers)

Once you know what you are testing, you need a sense of the market. Is this a niche of a few thousand people or a category of millions? Is demand growing or fading? Who already spends money to solve this problem? AI can compress days of research into an afternoon. It can also confidently invent numbers that do not exist, and a fabricated market size is worse than no number at all, because it feels like evidence. This lesson teaches you to get the speed without getting fooled.

What You'll Learn

• What TAM, SAM, and SOM mean in plain language
• How to research market size and trends with AI
• Why AI invents plausible-looking statistics, and how to catch it
• A verification workflow so every number you keep is traceable

TAM, SAM, SOM in Plain Language

You do not need a finance background. Three nested numbers describe how big your opportunity is, from the whole world down to what you could realistically win.

• TAM (Total Addressable Market): everyone in the world who has the problem you solve. The biggest possible circle.
• SAM (Serviceable Available Market): the portion you can reach given your product, geography, and channels.
• SOM (Serviceable Obtainable Market): the realistic share you could capture in the next year or two against existing competition.

For a pre-launch founder, SOM matters most. A huge TAM is meaningless if your obtainable slice is tiny. AI tends to anchor on the giant TAM number because it is the one most loudly repeated online, so you have to steer it down to SOM deliberately.

How AI Invents Numbers (and Why It Is Convincing)

A large language model predicts plausible text. When you ask "how big is the market for X," it will happily produce a precise figure like "$4.7 billion, growing at 12% a year" even when no such study exists, because that sounds like the kind of answer that belongs there. Researchers and consultants have documented AI generating market-size figures, growth rates, and even fake citations that look entirely real. The most dangerous output is not an obvious mistake. It is a believable statistic with a confident tone and a citation that turns out to point nowhere.

This is the single most important habit in this course: treat every number AI gives you as a lead to verify, never as a fact to quote.

A Verification Workflow That Actually Works

The fix is not to avoid AI. It is to use it for finding and structuring, then verify the numbers yourself. Follow this loop.

Step 1 — Use a research mode that browses the web. ChatGPT, Claude, and Gemini all offer a "deep research" style mode that searches the live web and attaches citations, rather than answering from memory. Turn that on. A model answering from memory alone is far more likely to fabricate a figure. As of mid-2026 these research modes search many sources and return a structured report with links, which is exactly what you want.

Step 2 — Force it to cite. Add this to your prompt:

Research the market for [your idea]. For every number you give, include the specific source and a working link. If you cannot find a credible source for a figure, say "no source found" instead of estimating. Prefer government data, industry reports, trade associations, and reputable news over blogs.

Step 3 — Open every source. Click through. Does the page actually contain the number? Is the study recent, or is it a decade old being recycled? Does the source have a reason to inflate the figure (for example, a vendor selling into that market)? If the link is dead or the number is not there, the number does not exist for your purposes.

Step 4 — Keep only traceable numbers. Anything you cannot trace to a real source gets cut or clearly labeled as a rough guess. Your validation deserves honest inputs.

Build a Bottom-Up Estimate Instead

Top-down numbers from reports are easy to fabricate and easy to misread. A bottom-up estimate is harder to fake and often more honest, and AI is genuinely useful for building one because the logic is transparent. Try:

Help me build a bottom-up market estimate for [idea]. Walk through the math step by step: roughly how many people fit my target customer, what share plausibly has this problem, how many might pay, and a realistic annual price. Show every assumption as a separate number I can challenge, and flag which numbers I should verify.

Because every step is visible, you can sanity-check each input and swap in real figures as you verify them. A defensible "about 8,000 likely customers in my region paying around $40 a month" beats a hand-wavy "$5 billion market" every time.

Spot the Trend, Not Just the Size

Size is a snapshot; trend is the direction. Ask AI to summarize whether interest in your category is rising or falling, what is driving it, and what could disrupt it, then verify the same way. Free tools like search-interest trackers and public industry news give you signals you can confirm with your own eyes. A small but fast-growing market can be a far better bet than a large, shrinking one.

Key Takeaways

• TAM is the whole problem, SAM is what you can serve, SOM is what you can realistically win. SOM matters most pre-launch.
• AI invents plausible market numbers and fake citations. Treat every figure as a lead, not a fact.
• Use a web-browsing research mode and force the model to cite a source for every number.
• Open every source yourself; if you cannot trace the number, cut it.
• Build a bottom-up estimate so each assumption is visible and checkable.
• Track the trend, not just the size. Direction often matters more than the headline figure.