Competitor Research

Competitor Research

Help the user decide what to learn from a competitor. A useful result distinguishes what happened, what might explain it, and what the user can test. Do not promise a complete growth history for every domain.

Start with the decision

Identify the exact product/domain, the research question, and any known market, budget, or stage. Ask only for missing context that would materially change the answer; otherwise begin with a stated scope. If the user's own product is unknown, make recommendations conditional rather than inventing a business context.

Select an available evidence route:

• Public web: use the host's available search/browser tools. No account or MCP is required by this skill. Research capability and model usage come from the host.
• Supplied materials: analyze the user's files, extracts, or reports. A URL alone is not its contents. Without browsing, request the contents needed to proceed and label the report as supplied-materials-only.
• Optional connected data: if the user has a suitable research connector, use it within their authorized scope. Keep the public or supplied-materials route useful even when no connector exists.

Never install a connector, create an account, submit a directory listing, contact a founder, or incur paid research charges merely because research was requested. Existing explicit authorization for a scoped budget remains valid.

Gather evidence

Read references/evidence.md before interpreting growth claims.

Start with the product's own site, pricing, documentation, changelog, and dated launch announcements. For a narrowly historical question, omit irrelevant current pricing or feature research. Then look for independent accounts, public launches, founder explanations, and channel evidence relevant to the question. Search snippets are discovery aids; read the supporting passage before treating its claim as verified. A successful page open without useful body text is not verification. If the passage is inaccessible, attribute the snippet and lower confidence.

For each material claim record: claim, source URL or supplied filename, source type, event date (if known), publication date (if distinct), observation date, and limitations. Record precise dates only when supported. Distinguish a source's claim from independent corroboration; duplicated syndication is one underlying source.

Focus on:

• Audience, job to be done, positioning, pricing unit, and free-to-paid boundary.
• A bounded timeline of supported events, not a fabricated continuous history.
• Evidence of acquisition: a published page, listing, integration, campaign, mention, or referral. Separate channel presence from traffic and conversion evidence.
• What is transferable at the user's stage, including distribution access, costs, existing audience, and product prerequisites.

For a first brief, prioritize a few high-value sources. When two targeted search passes add no material evidence, stop, state the gaps, and deliver the supported findings. For a famous company, narrow to the requested period/product rather than substituting its general company history. For a sparse site, do not replace missing evidence with a generic startup narrative. Expand only when the user's question requires it.