Competitors shows how you do when an agent has to choose between you and the products it could use instead, and what tipped the choice.

​
Finding your competitors

Forge writes five short jobs a person would hand their agent where your product is one of the right answers, like “Add SMS verification to my signup flow”. Each job goes to an agent with web search, with no product named, and Forge notes what it recommends. The products a customer could use instead of yours become your competitors, up to six. This step only finds who’s in the running. Whether an agent names you unprompted depends on what it learned in training and how widely you’re written about, which no fix on your site changes, so it isn’t scored.

​
Head to head

Each job then goes to an agent again, this time with a shortlist: you and up to three competitors. The agent may only judge from what it can read on each site right now, such as the home page, pricing and docs. It picks one and notes, for every product, what helped and what it couldn’t find. Each head-to-head runs twice with the shortlist in opposite orders. When both runs agree, that product wins the job; when they don’t, it’s a split.
  • The headline is how many head-to-heads you won.
  • The scoreboard lists you and your competitors with head-to-heads won and each one’s agent experience score. Open a competitor to see why agents chose it, the jobs it won, and what it gives agents that you don’t.
  • Why agents chose someone else groups what agents couldn’t find on your site in the jobs you lost, such as pricing that isn’t public or docs that don’t cover the job. Each reason comes with What to change: one change to your site or docs, written from the agents’ notes and what helped the winners, with an instruction you can copy into Claude Code, Cursor or Codex. Where a Discovery fix answers it too, it’s linked.
  • The head-to-heads list each job with its result (a tick when you won, a cross when you lost, a dash for a split), its shortlist and who won. A strip of squares shows the whole record at a glance.
These are Forge’s tests of how agents choose, not your real agent traffic. To see your agent sessions, connect the SDK.

​
Keeping the list right

  • Not a competitor removes one. Later searches won’t suggest it again.
  • Add a competitor adds one the search missed. Forge checks its site, and it stays through later searches.
  • Search again finds competitors and runs the head-to-heads again, for example after you ship a fix.