Niche research before you launch

Market recon before you spend money — and not from scratch every time. AI + search collect competitors, their prices and promises, pains from real reviews, and help you find an open niche. And so you don't copy prompts over and over, the whole recon can be packaged into a ready Claude Skill (a SKILL.md file) that runs from a single phrase. The output is a short report and a "go / no-go / go in differently" decision.

What you get out of it

• A market map: who the competitors are, what they promise, at what prices, who they target.
• A list of pains and complaints from real reviews (where competitors fall short).
• An open position for you: how to be different instead of "one more of the same".
• A demand estimate and a quick SWOT, so you can decide whether to enter at all.
• A short 1-page summary report: "go / no-go / go in differently".
• A ready Claude Skill (SKILL.md): the same recon runs from a single phrase, with no manual prompt reassembly.

What you'll need

Tools from the database (by name): Perplexity (search with sources — competitors, prices, market), ChatGPT/Claude (synthesis, SWOT, positioning), the prompt "Testing a business hypothesis (before you build)", the prompt "Positioning and USP", the MCP Exa Search MCP (if you need "smart" web search from an AI agent), vidIQ (if the niche is about video/YouTube — keyword demand and competitor channel breakdowns). Accounts/money/hardware:

• The free route: Perplexity (free), ChatGPT/Claude (free), any PC; browsing sites and review aggregators by hand.
• The fast route: Perplexity Pro (~$20/mo — deeper and more sources), vidIQ Pro (for video niches).
• The repeatable route: package the recon into a Claude Skill (SKILL.md) — a file in Claude Code for free, or a zip upload to claude.ai (see the separate section below).

Step by step

1. Competitor map. Tool: Perplexity (it gives you sources). Action: collect the players in the market. Ready-made prompt:

I want to enter the niche: "<niche/product>", market <geo/language>.
Find 7-10 key competitors. For each: what they offer, their main
promise (USP), who they target, price range, strengths.
Give links to sources for every point.

How to tell it worked: you have a list of real competitors with prices and promises, and the facts have links (not "general talk"). 2. Pains from reviews. Tool: Perplexity/ChatGPT. Action: find what competitors' customers complain about — that's your opening. Ready-made prompt:

Find the typical complaints and grievances customers have about these products/competitors:
<list>. Sources: reviews, forums, marketplaces, social media.
Group the pains by frequency: what annoys people most and what they feel is missing.

How to tell it worked: you get a grouped list of real pains — the "holes" you can fill. 3. Demand estimate. Tool: Perplexity (or vidIQ for video niches). Action: check that people actually search for / watch this niche. How to tell it worked: you see signs of demand (query volume, discussions, number of competitors) — there's a market, not "nobody needs this". 4. SWOT and the open position. Tool: ChatGPT/Claude. Action: synthesize what you collected into a SWOT and find the unoccupied angle. Ready-made prompt:

Here's the niche data: competitors <…>, their promises <…>, prices <…>, customer pains <…>.
Produce: 1) a short SWOT for MY entry (my profile: <who I am>);
2) 3 open positions/angles where competitors are weak and I can stand out;
3) for each — why this one is worth taking.

How to tell it worked: you see 2-3 realistic angles of difference tied to the competitor pains you found, not "abstractly better". 5. Hypothesis check and USP. Tool: the prompts "Testing a business hypothesis" + "Positioning and USP" in ChatGPT/Claude. Action: formulate the hypothesis for the chosen angle and stress-test it. How to tell it worked: you have a clear USP and a list of risks/assumptions to test cheaply before investing. 6. Summary report. Action: ask for everything condensed to 1 page: market, competitors, pains, your position, demand, the "go / no-go / go in differently" verdict and 3 first steps. How to tell it worked: the report fits on one screen and you can make a decision from it without rereading the whole research.

Package the recon into a ready Claude Skill (SKILL.md)

If you do this kind of recon regularly, don't copy the prompts every time. A Skill (Agent Skill) is Claude's official format: a folder with a SKILL.md file that Claude picks up on its own when a request matches its description. The metadata (name + description) always sits in the system prompt (~100 tokens per skill), and the body of the instructions only loads once the skill fires — that's "progressive disclosure" from Anthropic's official docs. The format is open (the anthropics/skills repository, 161k★ on GitHub, Apache 2.0 for most skills).

A minimal SKILL.md for competitor analysis:

---
name: niche-competitor-analysis
description: Niche and competitor recon before launch. Collects a competitor map (USP, prices, audience), pains from reviews, a demand estimate, a SWOT and the open position, plus a "go / no-go / go in differently" verdict. Use when the user is entering a new niche, testing a business hypothesis or asking for competitor analysis.
---

## When to apply
The user is entering a niche, testing a hypothesis or asking for a competitor breakdown.

## What to collect (step by step)
1. Competitor map: 7-10 players, and for each — USP, price, audience, strengths. Source links are mandatory.
2. Pains from reviews: group complaints by frequency — those are the open "holes".
3. Demand: the signs (query volume, discussions, number of players).
4. A SWOT for this specific user's entry.
5. The open position: 2-3 unoccupied angles with reasoning.
6. A 1-page report: market, pains, position, verdict, 3 first steps.

## Rules
- Every fact comes with a link. Web search is mandatory, invented numbers are forbidden.
- No niche/geo in the request — ask first.
- The finish line is a decision and 3 steps, not 40 pages.

Format limits (from the official docs): name — up to 64 characters, lowercase letters/digits/hyphens only, and it must NOT contain the words "claude"/"anthropic" (which is why the skill is called niche-competitor-analysis and not "claude-…"); description — up to 1024 characters, and it must state BOTH what it does AND when to trigger it — Claude uses it to decide whether to call the skill.

Where to put it:

• Claude Code (free, just a file): ~/.claude/skills/niche-competitor-analysis/SKILL.md (personal) or .claude/skills/… inside the project — Claude finds the skill by itself, no upload needed. Here the skill has full network access, so live web search for competitors works.
• claude.ai: zip the skill folder and upload it in Settings → Features (you need a Pro/Max/Team/Enterprise plan with code execution enabled).

After that just write: "analyze the niche <…> in " — Claude will pull in the skill and walk the steps.

Free route vs fast (paid)

Basic recon really can be done for free in one evening. Paying speeds up collection and gives you more sources, but output quality depends on your questions, not your plan.

Common problems and fixes

• The AI produces plausible but invented facts. Use Perplexity/Exa (they answer with links) and verify key figures/prices against sources — don't trust a "bare" answer with no links.
• Too broad a niche → mush. Narrow it to a specific segment and geo; "coffee shops" → "specialty takeaway coffee shops in ".
• You found lots of competitors and "it's all taken". Lots of competitors = there's demand; look not for "empty space" but for customer dissatisfaction (step 2) — that's where your position is.
• Analysis for its own sake ("paralysis"). The goal is a decision and 3 first steps, not 40 pages; limit yourself to one evening and a one-page report.
• The skill doesn't search the internet. On the bare API a skill runs in a sandbox with no network — live web search doesn't work there. For recon, run the skill in Claude Code (full network access) or in claude.ai with web search enabled.
• The skill "disappeared" on another surface. Skills don't sync between claude.ai, the API and Claude Code — put the file in each one separately.
• Someone else's skill is a risk. Only install skills from trusted sources: a skill can execute code, so read a third-party SKILL.md in full first.
• Stale data. Ask for fresh sources and note the date; markets and prices change.

How much time and money it takes

• Time: basic recon — 2-4 hours (competitors + pains + SWOT + report); deep recon — 1-2 days. Assembling the skill from ready prompts — about 20 minutes, once.
• Money: free (Perplexity free + ChatGPT/Claude free) or ~$20/mo for Perplexity Pro for depth. The skill itself is a free text file; you only pay for access to Claude (plan/tokens).
• Honestly: AI speeds up collection and structuring, but it doesn't replace going out and talking to 5 real customers. The best niche test is your first sales/pre-orders, not a report.