KnitKnot

# Run your first benchmark

Confirm the setup inputs, choose the engines available to you, and follow the first run.

## At a glance

  1. 1. **Confirm the setup**: Review products, competitors, capabilities, buyer roles, topics, and prompts.
  2. 2. **Start a run**: Open Benchmarks, choose available engines under Run all, and select Run.
  3. 3. **Inspect results**: Follow the run, then open the captured answers behind the result.

A benchmark runs your active prompts against the AI engines you select. Check the inputs first: the questions and competitor set determine what this reading can tell you.

## 1. Review Benchmark setup

Open **Get started** in the [console](https://app.knitknot.ai/onboarding) while setup is available. The steps are **Company & products**, **Competitors**, **Capabilities**, **Decision criteria**, **Buyer roles**, **Topics**, and **Prompts**. Confirm the items you want included as you move through them. A product can have its own competitor and prompt scope; the company remains the brand identity.

If setup is complete, use **Facts** to check the company record and [Prompt library](https://app.knitknot.ai/prompts) to review topics and prompts. Use the **Prompts** view for the question list. You can add a prompt, star one that matters, or archive one that should no longer run. See [Manage benchmark questions](/docs/manage-questions/) for those controls.

## 2. Start the run

Open [Benchmarks → Results](https://app.knitknot.ai/benchmarks). Select **Run all**, check the engines you want, then select **Run**. An engine marked **Paid** is unavailable on your current plan. A run starts in the background, so you can leave the page and return to **Benchmarks → Runs** to check its state.

Choose the same active prompts and engines for later measurements when you want a close comparison. A change to either can change the result even if your published content stays the same.

## 3. Read the answers

When scoring completes, open the result and inspect the answers behind it. Head-to-head prompts can show which vendor was recommended and why. Claim checks compare material statements with available company facts and receipts. A source is attributed to a claim only when the captured answer provides a reliable citation signal; missing attribution stays missing.

Some answers or checks may fail. Read the run status before treating a partial result as a complete benchmark. The [report guide](/docs/read-your-report/) explains the score and the drilldowns.

Docs navigation
Docs › Getting started

Run your first benchmark

Confirm the setup inputs, choose the engines available to you, and follow the first run.

Updated

At a glance

  1. Confirm the setup

    Review products, competitors, capabilities, buyer roles, topics, and prompts.

  2. Start a run

    Open Benchmarks, choose available engines under Run all, and select Run.

  3. Inspect results

    Follow the run, then open the captured answers behind the result.

A benchmark runs your active prompts against the AI engines you select. Check the inputs first: the questions and competitor set determine what this reading can tell you.

1. Review Benchmark setup

Open Get started in the console while setup is available. The steps are Company & products, Competitors, Capabilities, Decision criteria, Buyer roles, Topics, and Prompts. Confirm the items you want included as you move through them. A product can have its own competitor and prompt scope; the company remains the brand identity.

If setup is complete, use Facts to check the company record and Prompt library to review topics and prompts. Use the Prompts view for the question list. You can add a prompt, star one that matters, or archive one that should no longer run. See Manage benchmark questions for those controls.

2. Start the run

Open Benchmarks → Results. Select Run all, check the engines you want, then select Run. An engine marked Paid is unavailable on your current plan. A run starts in the background, so you can leave the page and return to Benchmarks → Runs to check its state.

Choose the same active prompts and engines for later measurements when you want a close comparison. A change to either can change the result even if your published content stays the same.

3. Read the answers

When scoring completes, open the result and inspect the answers behind it. Head-to-head prompts can show which vendor was recommended and why. Claim checks compare material statements with available company facts and receipts. A source is attributed to a claim only when the captured answer provides a reliable citation signal; missing attribution stays missing.

Some answers or checks may fail. Read the run status before treating a partial result as a complete benchmark. The report guide explains the score and the drilldowns.