market-research

market-research: Comprehensive Market Research for Any Idea

Research ANY product or business idea across Web, Reddit, and X (Twitter). Validate the problem, size the market, map the competition, profile the customer, and identify distribution channels. Produce a structured markdown report with citations. No API keys required.

Modes

Parse the user's invocation for depth flags:

• --quick → Faster, fewer queries (~15 total). Good for a quick gut check.
• (default) → Balanced (~25 total). Solid coverage across all categories.
• --deep → Comprehensive (~35 total). Leave no stone unturned.

Store: DEPTH_MODE = quick | standard | deep

---

PHASE 1: Intake

Step 1: Gather the idea

If the user provided their idea inline (e.g., /market-research AI scheduling tool for plumbers), parse it directly. Otherwise, use AskUserQuestion:

Tell me about your business/product idea. Include as much as you can:

- What's the product or service? (one-liner)
- What problem does it solve?
- Who's the target customer?
- What industry/category is it in?
- Any competitors you already know about?

Step 2: Parse into variables

Extract and store:

• IDEA_NAME — short name for the product/idea
• IDEA_ONELINER — one-sentence description
• PROBLEM — the core problem it solves
• TARGET_AUDIENCE — who it's for (be specific)
• INDUSTRY — industry or market category
• PRODUCT_CATEGORY — the type of product (e.g., "scheduling software", "marketplace")
• KNOWN_COMPETITORS — list of known competitors (may be empty)
• IDEA_SLUG — lowercase, hyphenated, max 40 chars (for filename)

Step 3: Confirm and start

If all variables are clear, display the parsed intent and proceed:

Researching: {IDEA_NAME}

Parsed intent:
- Idea: {IDEA_ONELINER}
- Problem: {PROBLEM}
- Audience: {TARGET_AUDIENCE}
- Industry: {INDUSTRY}
- Known competitors: {KNOWN_COMPETITORS or "none provided"}
- Mode: {DEPTH_MODE}

Starting research across Web, Reddit, and X. This typically takes 3-10 minutes.

If PROBLEM or TARGET_AUDIENCE cannot be inferred, ask ONE follow-up AskUserQuestion to fill the gaps. Maximum 2 total questions before research begins.

---

PHASE 2: Query Generation

Read the query template library:

Read file: ~/.claude/skills/market-research/references/search-queries.md

Substitute the parsed variables into the templates. Select queries based on DEPTH_MODE per the depth table in the reference file.

Identify 3-6 relevant subreddits using the subreddit discovery table. If the domain is unclear, run one WebSearch: "{INDUSTRY}" site:reddit.com to find active communities.

---

PHASE 3: Parallel Research Execution

Maximize parallelism. Run independent searches in the same tool-call batch.

Step 0: Check Reddit MCP availability

Before running any Reddit queries, test MCP availability with a single probe call:

mcp__reddit__search_reddit(query="test", search_in="posts", limit=1)

• If it succeeds → set REDDIT_MODE = mcp and use MCP tools for all Reddit queries.
• If it errors (tool not found, connection refused, timeout) → set REDDIT_MODE = websearch and use the WebSearch fallback for all Reddit queries.

WebSearch fallback mapping (use when REDDIT_MODE = websearch):

MCP Call	WebSearch Fallback
mcp__reddit__search_reddit(query=Q)	WebSearch("{Q} site:reddit.com")
mcp__reddit__scrape_subreddit(subreddit=S)	WebSearch("{PROBLEM} OR {PRODUCT_CATEGORY} site:reddit.com/r/{S}")
mcp__reddit__scrape_post(url=U)	WebFetch(url=U, prompt="Extract the post title, body, and top comments with upvote counts")
mcp__reddit__get_posts(target=S, search_query=Q)	WebSearch("{Q} site:reddit.com/r/{S}")

When using fallback mode, note in the final report: "Reddit data collected via web search (MCP unavailable). Thread-level comment depth may be limited."

Wave 1: Reddit + Web (parallel)

Reddit (use MCP tools or WebSearch fallback per REDDIT_MODE):

• Scrape top subreddits: mcp__reddit__scrape_subreddit(subreddit, limit=50, scrape_comments=true)
• Search for problem/industry: mcp__reddit__search_reddit(query, search_in="both", limit=30)

Web + X (WebSearch — parallel with Reddit):

• Run Problem Validation web queries
• Run Market Size web queries
• Run Competition web queries (including site:g2.com, site:producthunt.com)
• Run X queries using site:x.com scoping

Group all independent WebSearch calls into a single parallel batch.

Wave 2: Remaining categories

• Run Customer web + X queries
• Run Distribution web + X queries
• Run competitor-specific Reddit searches if KNOWN_COMPETITORS provided

Wave 3: Targeted follow-ups (if needed)

Review coverage. If any category is thin (< 3 data points):

• Scrape specific promising Reddit posts: mcp__reddit__scrape_post(url, scrape_comments=true) (or WebFetch fallback)
• Run broader web queries (use adjacent industry terms)
• Fetch high-value URLs with WebFetch

Track throughout:

• REDDIT_THREADS — count of Reddit threads/posts found
• X_POSTS — count of X results from site:x.com searches
• WEB_PAGES — count of web search queries run
• SUBREDDITS — list of subreddit names explored

---

PHASE 4: Synthesis & Report

Step 1: Synthesize per category

For each of the 5 categories, synthesize findings with source weighting:

• Reddit/X: HIGH weight. Real people, engagement signals (upvotes, likes, comments).
• Web articles: MEDIUM weight. Good for factual data (market size, pricing).
• Company websites: LOW weight for opinions (biased), HIGH for feature/pricing data.

Ground every finding in actual research data. Do not inject pre-existing knowledge as findings. If general knowledge adds useful context, label it: "[Note: based on general knowledge, not from this research]".

Step 2: Generate the report

Read the report template:

Read file: ~/.claude/skills/market-research/references/report-template.md

Fill in the template. Every claim needs a citation:

• Reddit: (r/{subreddit}, {N} upvotes)
• X: (@{handle} on X) or (X search: {N} posts mentioning this)
• Web: ({source name})

Step 3: Assign verdicts

Problem Validation verdict:

• Strong Signal: 10+ organic mentions, people actively seeking solutions, workarounds described
• Moderate Signal: 3-10 mentions, some discussion but not urgent
• Weak Signal: Under 3 mentions, or only in marketing contexts (not organic)

Competition verdict:

• Crowded: 5+ direct competitors with funding/traction
• Moderate: 2-4 direct competitors
• Blue Ocean: 0-1 direct competitors

Market verdict:

• Growing: CAGR > 10% or strong trend signals
• Stable: CAGR 2-10%
• Shrinking: CAGR < 2% or declining signals

Step 4: Save the report

Create output directory and save:

mkdir -p .tmp/market-research

Write to: .tmp/market-research/{IDEA_SLUG}-{YYYY-MM-DD}.md

Step 5: Display summary

After saving, display to the user:

---
Market research complete for {IDEA_NAME}.

## Key Findings
- [Top 3-5 findings as bullet points]

## Verdicts
- Problem Validation: {Strong/Moderate/Weak Signal}
- Market: {Growing/Stable/Shrinking} — {size if found}
- Competition: {Crowded/Moderate/Blue Ocean}
- Distribution: {Clear channels / Needs creative approach}

## Research Stats
├─ 🟠 Reddit: {N} threads across r/{sub1}, r/{sub2}, ...
├─ 🔵 X: {N} posts analyzed
├─ 🌐 Web: {N} pages searched
└─ 📄 Report: .tmp/market-research/{IDEA_SLUG}-{date}.md

Want me to dive deeper into any section, or help plan next steps?
---

---

PHASE 5: Follow-up

After the report, handle follow-up requests:

• "Go deeper on [section]" → Run targeted searches for that category, update report
• "Help me plan next steps" → Generate tactical action plan based on findings
• "Compare [competitor A] vs [competitor B]" → Deep-dive comparison
• "What subreddits should I monitor?" → List relevant subreddits with engagement levels

For follow-ups, do NOT re-run the entire pipeline. Use targeted searches only.

---

Error Handling

• Reddit MCP unavailable: Detected automatically in Phase 3, Step 0. Uses WebSearch fallback mapping. Noted in report Methodology section.
• Thin data in a category: Broaden to adjacent industry terms. Report honestly rather than fabricating.
• Very niche topic: Try broader industry terms and adjacent audiences. Note the pivot in the report.

Additional Resources

Reference Files

• references/search-queries.md — Full query template library organized by category and depth mode
• references/report-template.md — Structured markdown template for the output report