---
title: "Configure Sitecore AI for visual search"
slug: "configure-sitecore-ai-for-visual-search"
abstract: "Understanding how AI-assisted visual search performs and what you can configure to control how it works."
---
# Configure Sitecore AI for visual search

> [!important]
> AI capabilities in Content Hub need to be enabled and, if necessary, configured by a *superuser* or a user with the necessary permissions granted to them through [user group policies](https://doc.sitecore.com/ch/en/users/content-hub/user-group-policies.html "User group policies"). Most AI capabilities require Sitecore Stream.

You can fine-tune how AI-assisted visual search performs, for example, you can change the number of results returned or filter out assets that are not relevant. If you are seeing a lot of returned results that are not relevant, adjust the settings or upload more assets. Having a wide variety and a large number of assets helps visual search return more valid results. For more information, see the [AI features FAQ](https://doc.sitecore.com/ch/en/users/content-hub/ai-features-faq.html "AI features FAQ") and the [Using visual search](https://doc.sitecore.com/ch/en/users/content-hub/using-visual-search.html "Using visual search") best practices.

To configure visual search:

1. On the menu bar, click **Manage** :mdiCogOutline:.
2. On the **Manage** page, click **Settings**.
3. On the **Settings** page, in the search box, enter *VisualSearch* and then select it.
4. In the right pane, click **Switch editor mode**, select **Text**, and update any of the following settings:
    - `nrResults` - impacts the number of results returned. If you want more results returned, increase this value. For fewer results, decrease this value, which limits results to the most relevant. Default: 50.
    - `nrCandidates` - determines the number of documents to consider in a search. Increase this value to improve search accuracy. This might slow down searches. Decrease this value to speed up searches. This might reduce search accuracy.
    - `relevanceScore` - impacts the number of results returned and filters out results that are not relevant. Visual search returns a default number of results sorted by their relevance (with more relevant results displayed first). A relevance score of 1 means that two vectors are identical. A vector is a numerical representation of an image or text that captures its key features in a way that the search engine can understand. Instead of comparing image pixels or relying only on image metadata, a vector search, which is part of the visual search feature, allows AI to find similar images based on content. The default relevance score is 0.2. To display fewer but more relevant results, increase the relevance score (closer to 1).
5. To add to the list of terms that cannot be used with visual search, in the `deniedList`, add additional words using the appropriate syntax, and click **Save**. If a denied word is included in the search criteria, no results will be returned.

    > [!important]
    > The `deniedList` contains words that might be considered offensive.

6. Click **Save**.

> [!important]
> Results will not be returned if the search criteria contains a denied word or if the AI model cannot find relevant results with a relevance score higher than the defined value.