> ## Documentation Index > Fetch the complete documentation index at: https://docs.permutive.com/llms.txt > Use this file to discover all available pages before exploring further. # Creating Clean Room Audiences > How to create matched and modeled audiences in the Clean Room ## Overview This guide walks you through creating Clean Room audiences. Depending on your workflow: * **Publisher-driven workflows**: Publishers create audiences on behalf of advertisers * **Permutive-driven workflows**: Advertisers create their own audiences Both use the same interface described below. **Prerequisites:** * Data sources uploaded and permissioned between parties * Publisher connection established ## Steps In the left menu of your demand-side workspace, click **Cohorts**. Click **+ Add Cohort** in the top right. Enter a **Name** for your cohort (e.g., "ACME - Travel Intenders - Lookalike 70%"). Optionally add a **Description** to document the cohort's purpose. Build your **Query** using data from the available data sources: * Click the **+** button to add conditions * Select "FIRST PARTY" to access data source fields * Choose the data source (e.g., advertiser's uploaded data) * Select the specific field or segment you want to target * Add multiple conditions using "any" (OR) or "all" (AND) logic Select a **Country** to geo-lock the audience to a specific market. In the **Connections** dropdown, select which publisher connection(s) to deploy this cohort to. Click one of the buttons: * **Create Matched Cohort**: Creates an audience of users matched between the data sources. Segmentation begins immediately. * **Create Modeled Cohort**: Creates a lookalike audience by finding similar users. Requires model training (\~1 hour) followed by similarity selection before segmentation begins. ## Configuring Modeled Cohorts After you create a modeled cohort, the lookalike model will train for approximately 1 hour. Once training completes: Go back to the cohort in the Cohorts list. You'll see a curve showing the trade-off between audience similarity and reach. Choose a point on the curve based on your campaign goals: * **Higher similarity** (e.g., 90%): Smaller, more precise audience closely matching your seed data * **Lower similarity** (e.g., 70%): Larger audience with broader reach but less precise matching Confirm your selection. Segmentation will now begin and the cohort will be ready for targeting. **Important timing notes:** * **Matched cohorts** start segmenting users immediately upon creation * **Modeled cohorts** require: (1) \~1 hour for model training, (2) user selection of similarity-reach point, then (3) segmentation begins **Best practices:** * Use clear, descriptive naming conventions that include the advertiser name, audience description, and cohort type * Start with matched cohorts of at least 1,000 users for reliable targeting * For modeled cohorts, consider testing different model similarity levels (e.g., 70%, 80%, 90%) to balance reach and precision * Monitor cohort sizes and performance after creation to ensure they meet campaign requirements ## Next Steps Set up targeting in your ad server Return to product overview