Competitive Analysis

Evidence-graded teardowns, inside ChatGPT.

Ask a model to "research this company" and you get the marketing site, paraphrased. Every claim arrives with the same confidence whether it came from a filing or a launch tweet, nothing is dated, and the gaps go unmentioned because absence doesn't look like a finding.

Competitive Analysis is a ChatGPT plugin that runs the teardown the way I'd want a strategy analyst to run it. Ask it for a "competitive analysis of company" and it comes back with a verdict first, then the handful of findings that decide it, and every material claim carrying a link, an evidence date, and one of three labels:

  • VERIFIED: it read the primary source itself.
  • COMPANY-CLAIMED: the company said it, or press repeated the company. Funding announcements live here.
  • ESTIMATE: its own triangulation, with the method shown, or a headcount figure from a data tool you connected, with the tool named.

Four methods are required on every run, because that's where the alpha is:

  • Positioning history from the Wayback Machine. What the homepage and pricing page used to say, and the day they changed.
  • Job-listing mining. What a company is actually building shows up in job descriptions months before it ships.
  • Contradictions between sources, including the company's own pages. A discrepancy is a finding, not noise.
  • Absence as evidence. The review site with no reviews. The blog posts that now 404.

It ends with one of four verdicts: compelling, promising but unproven, structurally challenged, or unclear. If it couldn't reach something, it says what and why, instead of filling the hole with prose.

Bring your own research tools, or don't

The plugin holds no keys and runs no server. It's a set of instructions ChatGPT follows with its own browser. If you have Clay connected, it spends a little of your Clay credit on two things: the company's headcount history and its latest funding round. It never looks up the people who work there. A team that shrank right through a pivot tells you a lot. Without Clay, it runs on built-in browsing and tells you where that left gaps. Reddit coverage needs a community connector; without one, the report says "Reddit not searched" rather than guessing at sentiment.

One thing I learned in testing: with a search plugin like Exa installed, ChatGPT sends every page fetch through it, and those fetches couldn't open the Wayback Machine or the job boards. Exa's search turned up more competitors, but the history and hiring checks came back empty, so I run this one without it.

Where it came from

This plugin is the public version of the competitive-analysis skill I run in Claude Code. Mine does the same job with a few more moving parts (four parallel researchers, a paid Reddit gateway, a report that lands in a file). The plugin is the same method, written for a model that has a browser and whatever plugins you've connected. The source for the plugin itself is at github.com/amr05008/competitive-analysis.

Support

Questions, a bug, or a teardown that missed something? Email me at aaron@aaronroy.com with the company you asked about and I'll take a look.

Privacy ยท Terms