A benchmark — a run — is one pass of your active prompt library across ChatGPT, Claude, Perplexity, and Gemini, scored end to end. Runs are the heartbeat of the product: each one is a fresh measurement, and the sequence of them is your score trend over time.

You manage runs from the Runs page in the console.

Trigger a run

Use Run all to start a benchmark immediately. You can pick which engines to include; leaving all four selected gives you the complete picture. The run executes in the background — you’ll see live progress and per-prompt logs as each engine responds and each response is scored. When it finishes, your report updates to reflect the latest numbers.

Trigger a manual run after you’ve shipped a fix and want to check whether AI has picked it up, or any time you want a fresh reading.

Put it on a schedule

Most of the value comes from a steady cadence, not one-off runs — a trend only exists if you measure repeatedly. Set a schedule to have KnitKnot run benchmarks for you: choose weekly, biweekly, monthly, or a custom cadence, or turn scheduling off entirely. Weekly is a sensible default for actively managing your presence.

Scheduled runs use the same active library and engines as a manual run, so results stay directly comparable to the ones you trigger by hand.

Runs, periods, and your trend

Runs group into periods, and each completed run writes a snapshot of your headline numbers. That’s what the score trend on your home page reads from — the same formulas as the report, captured per run, so the line you see is the real measurement history, not a re-estimate.

Because runs are what get measured and billed, treat them as deliberate events. There’s no benefit to hammering the Run all button repeatedly — back-to-back runs against unchanged content just measure the same thing twice. Run when something has changed: new content shipped, a product added, the library or competitor set updated.

Next: once a run completes, read your report and work through the issues and playbooks it surfaces.