> ## Documentation Index
> Fetch the complete documentation index at: https://docs.frictionai.co/llms.txt
> Use this file to discover all available pages before exploring further.

# A/B Testing

> Create experiments that test whether prompt or content changes improve AI visibility.

# A/B Testing

A/B Testing lets you test a clear hypothesis against control and variant prompt buckets. Use it when you want to know whether a messaging, content, or positioning change affects AI visibility.

<Frame>
  <img src="https://mintcdn.com/friction/GwaVVezuSSMLCKiT/images/experiments-lifecycle.svg?fit=max&auto=format&n=GwaVVezuSSMLCKiT&q=85&s=28f433aa303c79fc11c48f9a32face74" alt="Experiment lifecycle from draft to prompt buckets, scheduled run, completed result, and decision modal" width="960" height="520" data-path="images/experiments-lifecycle.svg" />
</Frame>

<Info>
  A/B Testing is available on Professional and Enterprise plans.
</Info>

## When to run an experiment

Run an experiment when all of these are true:

* You have a clear hypothesis
* The prompts reflect one intent area
* Control and variant prompts are comparable
* You can wait for the run to complete and measure the result
* You are willing to record a decision after the run

## Create an experiment

<Steps>
  <Step title="Open A/B Testing">
    Go to Tracking -> A/B Testing and select **New Experiment**.
  </Step>

  <Step title="Name the experiment">
    Use a name that captures the content or positioning change.
  </Step>

  <Step title="Write the hypothesis">
    Example: "Use-case-focused prompts will improve visibility versus feature-focused prompts."
  </Step>

  <Step title="Select the model">
    Choose one available provider/model for the experiment.
  </Step>

  <Step title="Assign prompts">
    Add prompts to control and variant buckets. The app enforces minimum prompt requirements before a started run.
  </Step>

  <Step title="Save or start">
    Save a draft while building. Start when the buckets are ready.
  </Step>
</Steps>

## Experiment status

<Tabs>
  <Tab title="Draft">
    The experiment exists but has not been scheduled. You can edit details and prompt assignments.
  </Tab>

  <Tab title="Scheduled">
    The experiment is waiting to run.
  </Tab>

  <Tab title="Running">
    The selected AI model is processing the prompt buckets and collecting results.
  </Tab>

  <Tab title="Completed">
    Results are ready for review. Use the decision modal to record whether to ship, iterate, or archive.
  </Tab>

  <Tab title="Failed">
    The run did not complete. Review the error and create or retry the experiment depending on the failure.
  </Tab>
</Tabs>

## Good experiment design

<CardGroup cols={2}>
  <Card title="One hypothesis" icon="target">
    Test one change at a time. Mixing product copy, pricing, and page structure makes results hard to explain.
  </Card>

  <Card title="Comparable buckets" icon="scale-balanced">
    Control and variant prompts should have the same intent and difficulty.
  </Card>

  <Card title="Enough prompts" icon="list-ordered">
    Use enough prompts to reduce noise, but keep the experiment focused.
  </Card>

  <Card title="Decision discipline" icon="check-circle">
    Record a decision after completion so the experiment history remains useful.
  </Card>
</CardGroup>

<Warning>
  Experiments measure AI response behavior for the configured prompts and model. They do not prove that every AI assistant will change in the same way.
</Warning>