Claude is good at the structure of market research and bad at the numbers in it. It will build you a sizing model, a segment map and a competitor matrix in minutes, and it will also fill any empty cell with a confident figure that has no source behind it. The working method is to make it show its arithmetic, name its assumptions, and label every number it could not verify - then go and verify those yourself. Structure from the model, figures from primary sources.
Why generic AI market research misleads
An unguided model fills gaps with plausible fiction. Ask it to size a market and it hands you a figure to two decimal places with no source, no method and no way to tell whether it is real. It produces market sizes, growth rates and competitor revenues that sound credible and may be entirely invented. Worse, it states all of it at the same confidence, so you cannot separate the well-grounded reasoning from the guesswork - which is exactly the distinction that matters when a decision or an investor question rides on it.
The fix is not a better-worded request for accuracy. It is a process that forces the model to show its arithmetic and mark its own uncertainty, so the parts that need a primary source arrive labelled as questions rather than facts.
What do you give it before it researches anything?
A bare "size the market for my product" produces a paragraph of impressions. Five inputs turn it into a model you can argue with:
- The product, in one concrete sentence. "A booking tool" and "scheduling software for independent physiotherapy clinics" size to numbers three orders of magnitude apart.
- The geography and the time frame. A market is always a market somewhere, in some year. Without both you get a global figure that flatters everything.
- Every real number you already have. Your own pricing, your conversion rate, a competitor's published revenue, an industry body's member count. Anything real anchors the model and reduces what it has to invent.
- The competitors you already know. Naming five keeps it from listing the obvious three and stopping.
- What the research is for. A seed deck, a board paper and a pricing decision need different depth and different sceptics in mind.
How do you get a sizing model instead of a number?
Demand both directions. A top-down figure alone is unfalsifiable; a bottom-up build from unit counts is checkable, and the gap between the two is the most informative output of the whole exercise.
The tagging instruction is the one that earns its keep. Without it every number arrives looking equally solid; with it, the model's own guesses are marked as guesses, and the three-sensitivity list tells you exactly which two or three figures are worth a day of real research.
Every rule on this page, loaded once: sizing built both directions, assumptions stated in the open, and the source-tagging pass applied to every figure without being asked. She holds your market, your competitors and your real numbers between sessions, so the second study starts where the first one ended.
Get Serena →How do you get a competitor matrix worth reading?
A list of competitors with a sentence each is not analysis. The useful version compares them on the dimensions a buyer actually chooses between, and says out loud where two of them are saying the same thing - because converging language is the clearest positioning gap you will ever be handed, and it is invisible when you read the sites one at a time.
The last instruction matters more here than anywhere else on the page. Competitor revenue and funding are exactly the sort of figure a model will produce fluently and wrongly, and they are also the sort a reader will quote back to you in a meeting.
Serena sizes the market; Ivan lives inside it. Battlecards, win/loss framing, pricing-page teardowns and the running record of what each rival changed this quarter - built to be handed to a sales team, not just read once.
Get Ivan →The verification pass, and why it is not optional
Run this against anything the model has already produced, before the figures leave your screen. It is the same discipline a research firm applies to its own drafts, and it takes about a minute.
That last question is the one people skip. A model will happily produce forty figures of equal apparent weight; asking which one carries the argument turns a report into a research plan.
Where Claude fails at market research
- It cannot read a paywalled report. The Gartner, IDC and Forrester figures that anchor most market sizing sit behind paywalls. What the model recalls is whatever leaked into press coverage, at whatever age.
- It has a training cutoff and markets move. A funding round, an acquisition, a pricing change or a new entrant from the last few months may simply not exist for it.
- It cannot verify a private company's revenue. Nobody can, from outside. Any specific figure for a private competitor is an estimate wearing a suit.
- It will not volunteer that a market is too small. Ask "is this market big enough" and you get encouragement. Ask "what would have to be true for this market to be too small for a venture-scale business" and you get the actual answer.
- It cannot do primary research. It has not spoken to your customers. Anything about why people actually buy is a hypothesis to test in interviews, not a finding.
- Sample sizes and statistics are its weakest ground. Confidence intervals, significance and representative sampling are arithmetic, and arithmetic is where language models quietly fail.
None of that makes the tool useless. It makes it a research scaffold rather than a research department: it builds the structure and writes the questions, and you fill in the cells that carry real weight.
Once the VERIFY list exists, someone has to work down it. Aiden handles the literature side: finding the primary source, reading it properly, and telling you whether it actually says what the press summary claimed.
Get Aiden →How do you install a research skill in Claude?
A KissMySkills skill is a ZIP file with a SKILL.md at its root. In the Claude desktop app, open Customize → Skills, add a skill, upload the ZIP, and toggle it on.
Skills run on Free, Pro, Max, Team and Enterprise - there is no paid-plan gate. The real requirement is code execution: switch on Code execution and file creation under Settings → Capabilities. On Team and Enterprise an owner can disable it for the whole organisation, which is why the Skills menu sometimes looks greyed out - an admin setting rather than your subscription.
On the free plan without code execution, or if you work in ChatGPT or Gemini, open the SKILL.md in a text editor and paste its contents into custom instructions: a Claude Project's instructions, a ChatGPT custom GPT, or a Gemini Gem. Same rules loaded, you just re-paste them per workspace.
The reason to load a skill rather than re-paste these prompts is memory across sessions. A loaded skill keeps your market definition, your competitor set and the real figures you have already verified, so the second study does not start from a blank page. For the rest of the go-to-market workflow, the Sales & Marketing Growth collection holds 120 skills covering positioning, messaging and campaign planning.
In summary
Size it both directions, make the model tag every number as known, published or assumed, and run the verification pass before the figures leave your screen. Sizing, segments and the research plan: Serena ($19). Competitors as an ongoing job: Ivan ($19). Working down the VERIFY list: Aiden ($19). All three work in Claude, ChatGPT and any AI chat, with a 30-day money-back guarantee. None of them can read a paywalled report or call your customers.
Serena - Market Researcher AI Skill
Market sizing built top-down and bottom-up, segment profiles, competitor matrices and a SWOT - with every figure that needs a primary source flagged rather than invented.
KissMySkills is a marketplace of 2,000+ AI skills, 300+ prompt packs, 100+ agents & free tools for Claude, ChatGPT & any AI chat. 30-day money-back guarantee.
