---
title: Idea builder for Optimizely Web Experimentation
slug: web-experimentation/docs/idea-builder-for-optimizely-web-experimentation
docTags: 
createdAt: 2026-07-20T15:45:40.383Z
---

:::hint{type="success"}
**Products**

- Optimizely Web Experimentation
- Optimizely Feature Experimentation
:::

Idea builder is a dedicated tool in Optimizely Experimentation that helps you generate experiment ideas using Agent Platform, Optimizely's AI assistant. Provide structured context about your page, goal, and research, and Mark returns specific, actionable test ideas you can save, refine, and act on. Mark can analyze a single page or a multi-step flow.

Idea builder works best when you also have additional context ready, such as a heatmap image, qualitative research documents, or notes from your analytics tool. These inputs are optional but improve the quality of generated ideas.

## Access Idea builder

Click **Ideas**.

![](https://api.archbee.com/api/optimize/BrbvVORKJ5iIeSMZye5kI/jJPIxi4DDTnxK9_RXwy_M_ideas-ideas-nav-item-highlighted.png)

The Ideas page displays two tabs: **Draft** and **Saved**.

The context panel contains three collapsible sections: **Test Setup**, **Research**, and **Sources**.

### Provide page context

The **Test Setup** section is where you define where the test runs and what you want to improve.

Use the **Target by** drop-down list to select one of the following:

- **URL** – Enter a page URL for Mark to analyze. To analyze a multi-step flow, click **+** add up to four URLs, such as a product detail page, cart, and checkout page. Mark evaluates the full flow and suggests the best ideas across all the pages.
- **Saved page** – Select a page already saved in your experimentation account. This option is not available in Optimizely Feature Experimentation.
- **Screenshot** – Upload an image of the page you want to analyze. Drag and drop the image or click **Add** to upload.

In the **Goal** field, enter the primary outcome you want this experiment to achieve. For example, enter *Increase demo bookings* or *Improve step 1 to step 2 conversion*. A clear goal helps Agent Platform focus ideas on the behavior you want to change.

![](https://api.archbee.com/api/optimize/BrbvVORKJ5iIeSMZye5kI/s5u_MxBD8xZGnvT3LS6n3_ideas-test-setup-url-and-goal-filled-in.png)

### Add optional context

The following sections are optional but improve the relevance of generated ideas:

### Research

The Research section lets you add qualitative insights or files that explain user behavior and problem areas. Click **Research** to expand it.

![](https://api.archbee.com/api/optimize/BrbvVORKJ5iIeSMZye5kI/wk8toJ7LWjrnXcMWG62g7_ideas-research-additional-context-upload-area.png)

Upload files such as heatmaps, analytics reports, or user research documents to give Agent Platform visual evidence of user interactions. Drag and drop files or click **Add** to browse. Each file must be 5 MB or smaller.

In the text field, enter analytics notes, data points, or observations not captured in an uploaded file. For example, enter *Bounce rate is 80% for enterprise customers and 50% for small business customers*.

### Sources

The **Sources** section lets you reference past tests to inform new ideas and avoid duplication. Click **Sources** to expand it.

Use the **Previous Tests** search field to search for and select up to four past experiments. This helps Mark build on existing test results and avoid repeating experiments that did not work.

![](https://api.archbee.com/api/optimize/BrbvVORKJ5iIeSMZye5kI/HOajHM-GnrubVO44Zc2RO_ideas-sources-previous-tests-selection.png)

To remove a test, click **X** on each test card. To reset the entire context panel, click **Clear**.

### Generate ideas

After providing your page context and any optional inputs, click **Get ideas**.

![](https://api.archbee.com/api/optimize/BrbvVORKJ5iIeSMZye5kI/YfOhadnM0J6VwqJiOX_Wa_ideas-get-test-ideas-button-highlighted.png)

Mark analyzes the context you provided and generates experiment ideas. While ideas are generating, Agent Platform displays its reasoning process.

Generated ideas display as individual cards.

![](https://api.archbee.com/api/optimize/BrbvVORKJ5iIeSMZye5kI/w4ly1sGd_qT51O1T3ewm4_ideas-generated-test-ideas-list.png)

Click **View Details** to expand a card and see the full details.

![](https://api.archbee.com/api/optimize/BrbvVORKJ5iIeSMZye5kI/wd7h4LxXVzC1rmn6Nlnwt_ideas-high-contrast-red-cta-idea-details-expanded.png)

Each expanded card includes the following information:

- **Title** – The name for the experiment idea.
- **Description** – A summary of what the experiment would change.
- **Problem** – The user behavior or friction point the idea addresses.
- **Hypothesis** – A prediction of how the change will affect user behavior and the primary metric.
- **Why this will work** – The rationale behind the idea, including the conversion principle or heuristic that supports it.
- **Sources** – The page observations or principles Agent Platform used to generate this idea.

To remove an idea, click **X** on the idea card. To regenerate ideas, adjust the fields in the context panel and click **Update**.

## Refine an idea

Click **Refine** on any idea card. Click **Refine** on multiple cards to refine several ideas in one conversation. To deselect a card, click **X** next to its **Refine** button.

The Mark chat panel opens with suggestions tailored to Idea Builder, such as *Generate bolder ideas,* *more variations,* or *similar ideas.*

Enter a natural language prompt to refine the selected ideas. For example:

- *Make this more focused on mobile users.*
- *Generate three more ideas like this.*
- *Merge these two ideas together.*

Click **Submit**.

![](https://api.archbee.com/api/optimize/BrbvVORKJ5iIeSMZye5kI/qbIx6AGXfI1PIC0FsBSsr_opal-opal-chat-generate-more-ideas-prompt.png)

Mark generates updated or additional ideas based on your prompt. Refined ideas display at the bottom of the ideas list.

![](https://api.archbee.com/api/optimize/BrbvVORKJ5iIeSMZye5kI/LetVGkxFzPfw0uA_fXnPS_opal-opal-generated-three-additional-cta-ideas.png)

## Save ideas

To save an idea for later, click **Save** on the idea card.

![](https://api.archbee.com/api/optimize/BrbvVORKJ5iIeSMZye5kI/GQ1PmhSqifiJuGuW9tOhT_ideas-idea-successfully-saved-confirmation.png)

Saved ideas are available on the **Saved** tab.

![](https://api.archbee.com/api/optimize/BrbvVORKJ5iIeSMZye5kI/I4Vfno0jRs9W5wIUpRbSv_ideas-saved-ideas-list-tab.png)

To delete a saved idea, click **More (…) > Delete**.

## Build an experiment from an idea

:::hint{type="success"}
**Products**

- Optimizely Web Experimentation
:::

Build an experiment from an idea to turn that idea into a complete A/B test you can review and start. Mark does not only create the variation, it also creates the metric and the audience the test needs. You act on an idea without configuring each component yourself.

The following video shows the complete flow, from an idea card to an experiment ready to start:

[Idea to Build demo](https://dwhqmt.files.cmp.optimizely.com/download/4dda2cdea07011f181edc61de1503089?attachment=false)

1. Go to the **Ideas** page.
2. On the **Draft** or **Saved** tab, find the idea you want to test.
3. Click **Build** on the idea card. The Visual Editor opens in a new tab. Optimizely creates the experiment and adds **Variation #1**. Agent Platform receives the idea as a prompt, so you do not need to write a prompt yourself.
4. The New Idea pop-pup shows the idea details along with the Build Prompt. Click **Build Variation**.
5. The Visual Editor opens in a new tab. Optimizely creates the experiment and adds **Variation #1**. Agent Platform receives the idea as a prompt, so you do not need to write a prompt yourself.
6. Review the change Agent Platform applied to the variation. Mark lists what it changed under **What I built**. The summary covers anything Agent Platform decided differently from the idea.
7. Click **Save All** to keep the changes. To discard the changes instead, click **Undo All**.
8. Wait for Mark to check the experiment configuration. Mark reads the metrics and audiences that already exist in your project. Mark also reads the pages the experiment activates on. Mark reuses a matching event or audience instead of creating a duplicate.
9. Select a primary metric. Mark suggests metrics based on the idea. Click the checkbox for each metric you want, then click submit. The first metric you select becomes the primary metric. To describe a metric in your own words, click **Something else**. To continue without a metric, click **Skip**.
10. Select an audience. Mark suggests audiences, recommends one, and gives a short reason for each option. Select one option, then click submit. To describe an audience in your own words, click **Something else**. To continue without an audience, click **Skip**.
11. Choose what to do next. Mark creates the metric and the audience you selected and attaches them to the experiment. Mark then offers three options:
    - **Review experiment** – Opens the experiment **Summary** page in a new tab.
    - **Preview variation** – Opens the variation the way a visitor sees it.
    - **Generate another variation** – Adds another variation to the same experiment.
12. Click **Review experiment**, then review the pages, audience, variations, traffic allocation, and primary metric on the **Summary** page.
13. Click **Start Experiment** to run the test.

Mark-generated code sometimes contains mistakes. Review the variation and the experiment configuration before you start the test. Preview the variation at the viewport sizes your visitors use. A change that fits on desktop sometimes wraps or hides elements on mobile.

### What Mark builds

Mark creates each component the experiment needs to run:

- **Experiment** – Mark creates an A/B test and takes the experiment name and description from the idea.
- **Variation** – Mark adds **Variation #1** with the change it applied in the Visual Editor. Optimizely splits traffic evenly between the original and the variation.
- **Metric** – Agent Platform sets the metric you select as the primary metric. Mark reuses a matching event when one already exists in your project.
- **Audience** – Mark adds the audience you select. Mark reuses a matching audience when one already exists in your project.
- **Activation** – Agent Platform targets the pages you set as context, either the URL or the saved page.

You can change any of these from the experiment settings, the same way you would for an experiment you create yourself.