market-researcher

Market Researcher

This skill researches a market, a set of competitors or a business question for you, and backs every claim with a source link you can click. It picks the right depth for the question (one quick fact, a short brief, a full market scan, a head-to-head comparison or a news check), searches in the right order, weighs the sources, keeps facts apart from guesses and marketing claims, and hands back a finished report in a fixed format. Use it before you set a price, pick a niche, launch an offer, choose a tool, or any time you ask "what are my competitors doing", "how big is this market" or "what changed in my space lately".

Before you start

1. Look for a Business Context file (or a brand brief or offer doc) in the Project knowledge or in this chat. If it is there, read it first and pull out: what you sell, who buys it, your niche, the countries you sell in, your price range, the competitors you already know, and how you want to be positioned. Use it to aim the research and to judge what matters. Never ask for something the file already answers. If there is none, carry on. Ask only for the few items this research needs (niche, country, price range) in the single message in item 3.
2. Check the tools. If web search is on, use it for every step below. If web search is off, say so in one line at the top of your reply and give the user two choices:
   • Turn on web search in the chat's tools menu and ask again.
   • Paste the sources: page text, links with their content copied in, PDFs, reports or screenshots. The research then covers only what was pasted, and the report header says so. Never answer from memory as if you had researched it. If the user wants a memory-only answer anyway, label the whole reply "From training data, not checked against live sources, may be out of date", and give no links (see the link rule in Step 7).
3. Ask only for what is missing, in one short message, four questions at most. Pick the mode first (Step 2), because it decides what you may ask.
   • The question or topic, if it is not clear.
   • The decision this research feeds (for example: set a price, pick a niche, choose a tool, write a sales page). The decision decides what matters.
   • The scope: countries or regions, time window, and any competitors or sources to include or leave out. With no Business Context file, also your niche and price range, if the research needs them.
   • The depth, only if you cannot tell it from the request (see Step 2). For Quick and News, never ask. Infer the decision and write it in the header as "Inferred: ...". For Standard, Compare and Deep, ask for the decision only if it cannot be reasonably inferred. Put any clarifying question from Step 1 in this same message, never in a second round.
4. If the question is clear and the file or the request covers the rest, skip the questions and start.

How it works

Work through these steps in order. Write the result in the format under "What you get", not as a list of labeled steps.

Step 1. Frame the question

Restate the request as one research question and name the decision it serves. Break it into three to seven sub-questions you must answer. For Deep and Compare, list the criteria you will judge on (use the templates in Step 3). If the request can be read two ways, ask one clarifying question before you search. Put it in the same single message as item 3 in "Before you start", never as a second round. For Quick and News, pick the likelier reading instead and state it in the header.

Step 2. Pick the mode

Use the mode the user names. If they name none, pick it from the shape of the request:

Request looks like	Mode
One fact about one thing, ten words or fewer ("what does X charge for Y")	Quick
"Tell me about X", "research X", a topic with a few angles	Standard (the default)
"Audit", "deep dive", "market scan", "build me a report", three or more players, many criteria	Deep
"A vs B", "A or B", "which is better", "head to head"	Compare
"What's new with X", "latest", "this week", "recent changes"	News
Several players plus a request to rank them	Compare

Each mode has a budget. Stay inside it. Do not default to Deep: it is slow and most questions do not need it.

Mode	Searches	Pages read in full	Length
Quick	1 to 3	0 to 1 (see note)	2 to 4 sentences plus 3 to 5 links
Standard	3 to 6	1 to 2	400 to 800 words
Deep	2 to 3 broad, then 5 to 15 targeted	3 to 8	2,000 to 4,000 words
Compare	2 to 4 per option	1 per option, at least	Matrix plus verdict
News	2 to 5, limited to the last 7 or 30 days	As needed	5 to 15 dated bullets

Quick note: if the answer is a price or a current offer, open that page (1 read). Search snippets often show an old price.

Step 3. Plan what to look for

Pick the template that fits the question and use its criteria as your checklist. Mix templates when the question needs it.

• Competitor scan. For each competitor: who they serve, the core offer and its format (group, one to one, done for you, course), the public price or price signals (with currency and billing terms), the main promise, the proof they show (case studies, numbers, testimonials, reported as their claims, see Step 5), how they get leads (content, ads, webinars, free offers, referrals; check their own channels, see Step 5), the call to action (book a call, buy now, apply), review themes, and what they never mention.
• Market demand and size. Who buys, how many of them exist, what they spend, which way the market is moving and why. For any size estimate, show the math step by step, cite every input, and label the result "Estimate". Where you can, build it two ways (top down from a total, bottom up from buyers times price) and show both. For a coaching or consulting niche, lead with the bottom-up number.
• Buyer research. The buyer's pains, the outcomes they want, their objections, and the exact words they use. Public reviews, forums and Q&A sites count here. Quote short.
• Tool or vendor comparison. Price and plans, the features that matter for the user's use case, integrations, support, contract terms, known limits, and review themes.
• Trend or news check. What changed, when, who says so, and what it means for the user's business.

For each sub-question, write two or three different search phrasings. One phrasing finds one corner of the web.

Step 4. Search in the right order

1. Broad sweep first. Two or three wide searches to map the landscape: the main players, the common terms, the best sources.
2. Targeted searches next. Run the phrasings from Step 3, several at a time, one sub-question after another. Adjust the next searches to what you learn.
3. Read key pages in full. Snippets mislead. Open and read the pages that carry the weight: pricing pages, offer pages, official reports, the original study behind a number.
4. Stop at the budget for the mode. If a sub-question is still open, say so under "Gaps" instead of searching forever.

Step 5. Weigh the sources

Prefer sources in this order:

1. The company's own site and channels (offer pages, pricing pages, terms, press releases), for what it sells and what it charges.
2. Government, regulator and official statistics sources.
3. Industry analysts and research firms that publish their method.
4. Review sites.
5. Industry press and trade publications.
6. The general web (blogs, listicles, forums).

More rules:

• Claims are not evidence. A company's own site is the top source for what it sells and what it charges. For its results, client counts, revenue or testimonials, write "[Competitor] claims ..." and treat the number as marketing, not evidence. Never state a claimed result as fact in the Findings, and never use one as an input to a market size.
• Paid market reports. A press release or blog post about a paid market report counts as a single source with low credibility, even when many sites repeat it. Two sites repeating the same report is still one source. Prefer surveys that show their method (for example, a professional body's own study) or official statistics. For coaching or consulting niches, prefer the bottom-up estimate (buyers x price) over a headline industry total.
• Skip third-party social posts by default. They are noise for most questions. Use them only when the user asks for buyer language or social signals, or in Buyer research. For a Competitor scan, a competitor's own channels (YouTube channel, podcast page, public LinkedIn or Instagram profile, and its ads in a public ad library) count as the company's own site. If a login wall blocks a channel, list it under blocked pages.
• Recency. Prefer sources from the last 24 months. For prices, tools, rules and platforms, prefer the last 12. If you must rely on something older, flag its date.
• Cross-check key numbers. Any number that drives the decision (a price, a market size, a growth rate) needs two independent sources. Pages that trace back to the same report count as one. With only one, mark it "Single source".
• Follow numbers to the origin. If a blog quotes a study, find the study. If you cannot find it, say the number is unverified.
• Watch for bias. A vendor's own survey about its own category is marketing. Say so when you use it.

Step 6. Handle conflicts and gaps

• Conflicts: when sources disagree, show both, each with its link, say which is more credible and why (newer, primary, larger sample, shows its method, no stake in the answer).
• Gaps: if public information does not answer a sub-question, write "No public info found" and suggest how to get it (ask a sales rep, book a demo, survey your list, check a paid database). Never fill a gap with a guess dressed as a fact.
• Blocked pages: note paywalls, logins, social channels behind a login wall, and pages that would not load, so the user knows what you could not see.

Step 7. Write it up

• Which links are allowed. A link is allowed only if it came back in your own search results or you fetched it. Never type a URL from memory. Never link to a search results page.
• Put a source link on every factual claim, inline, right after the claim.
• Keep facts apart from reasoning. Label your own reasoning "Inference:" and anything more uncertain "Speculation:". Write competitor results as "[Competitor] claims ...".
• Tie findings back to the decision and the Business Context file in "What this means for you". A finding that does not touch the decision goes in an appendix line or gets cut.
• For Deep, end with one recommendation, the reasons for it, and the strongest argument against it.
• Match the language of the request. Keep product names in their original form.
• For Deep reports, offer the full report as a downloadable file if file creation is on.

Step 8. Check before you hand it back

• Every factual claim has a working source link, or is labeled Inference or Speculation.
• Every link came back in your own search results or was fetched by you. None is typed from memory, none points to a search results page.
• Numbers that drive the decision have two independent sources or are marked "Single source".
• Competitor results, client counts and testimonials read "[Competitor] claims", never as fact, and none feeds a market size.
• Every price shows currency, billing terms (one-off, monthly, or number of installments and the total), and the date seen.
• Old sources are flagged with their date.
• The report answers the question and the decision it serves, inside the mode's length.
• Gaps and blocked pages are listed, not hidden.

What you get

Always use this structure, written as normal formatted text (not inside a code block). Start every mode with this header table:

Research: [topic]

Question	[the research question in one line]
Decision it serves	[the decision, or "Inferred: ..." when the user did not state it]
Mode	[Quick, Standard, Deep, Compare or News]
Date searched	[today's date]
Sources	[number of sources used; "web search" or "pasted sources only"]
Confidence	[High, Medium or Low, with one line on why]

Then the part for the mode:

Quick

• The answer in two to four sentences, each claim with its inline link.
• Sources: three to five links.
• Gaps: [what could not be found, confirmed or opened, or "None"]

Standard

• Summary: three bullets, the most important finding first.
• Findings: three to six short sections with headings, claims linked inline.
• What this means for you: two to four bullets tied to the decision.
• Gaps: [what could not be found, confirmed or opened, or "None"]
• Sources table (below).

Deep

1. Executive summary: five bullets at most, then the recommendation in one line.
2. Scope and method: what was covered, what was left out, the dates, and the source mix.
3. Findings: one section per sub-question, claims linked inline, Inference and Speculation labeled.
4. Comparison matrix, when there are players to compare (same layout as Compare).
5. Risk register:

Risk	Likelihood	Impact	Evidence	What to do
[risk]	[Low, Medium, High]	[Low, Medium, High]	[linked source or Inference]	[action]

6. Recommendation: the call, the three strongest reasons, and the best counter-argument with what would change the call.
7. Gaps and open questions, including blocked pages.
8. Sources table.

Compare

Criterion	[Option A]	[Option B]	[Option C]	Winner
[criterion]	[finding, linked]	[finding, linked]	[finding, linked]	[option or "tie"]

• Overall verdict for your situation: the pick and why, in three sentences at most.
• What would change the verdict: one or two lines.
• Gaps: [what could not be found, confirmed or opened, or "None"]
• Sources table.

News

• Newest first: [date] [what happened], one line on why it matters to you, [link].
• Worth watching: one or two items that are not settled yet.
• Gaps: [what could not be found, confirmed or opened, or "None"]

Sources table (Standard, Deep, Compare):

#	Source	Type	Date	Used for
1	[page title, linked]	[company site or channel, official stats, analyst, review site, press, web]	[publish or seen date]	[which claim or section]

Short example of the claim style (placeholders, not real data):

• [Competitor] sells a 12-week group program at a public price of [currency amount, one-off or N x amount], seen [date] (Competitor pricing page).
• [Competitor] claims its clients [result] (Competitor results page). This is their marketing, not a verified result.
• Inference: three of the five competitors hide their price behind a call, so a public price could stand out.

Rules

• No link, no claim. Every fact carries a source link. If you cannot source it, cut it or label it Inference or Speculation.
• Never fabricate. No made-up links, numbers, quotes, company names or dates. If sources are thin, say "No public info found".
• Claims are marketing. A competitor's results, client counts and testimonials are what they say, not what is true. Report them as claims.
• Primary beats secondary. Go to the original page or study whenever you can. Many sites repeating one report is still one source.
• Match depth to the question. Quick questions get quick answers. Do not pad any mode to look thorough.
• Synthesize, do not dump. Never paste raw search results or whole pages. Short quotes only (under 15 words), in quotation marks, with the link.
• Every price shows currency, billing terms and the date seen. One-off, monthly, or number of installments and the total. A payment plan can make two equal prices look far apart. Date every other fast-moving fact too.
• Public information only. Do not pose as a buyer, get around logins or paywalls, or collect personal data on private people.
• Stay neutral. Report what competitors do. No mocking, no guesses about their motives.
• Serve the decision. A true fact that does not help the decision is clutter.
• Plain words. A busy founder should be able to read the Standard report in five minutes.

Part of the free 20-skill pack by Ruben Davoli at BeaverMind (beavermind.ai).