OM3.8: Admin > Assets > Semantic search

Overview

Semantic search lets users find assets by meaning, not just exact keywords. Instead of matching tag or file name text, it understands the intent behind a search query and returns conceptually relevant results.

For example, searching for “summer outdoor lifestyle” can surface relevant photos even if those words don’t appear in the asset’s tags or title.

This Menu Item is for Brandkit Staff Staff System Admins only.

How it works

Brandkit uses AI-powered embeddings (via OpenAI) to analyse and index your assets. When a user runs a semantic search, the query is matched against this index to find the closest conceptual matches.

Semantic search works alongside — not instead of — keyword search. Users can switch between the two from the search bar on the Grid.

Enabling semantic search

Semantic search requires OpenAI embedding to be active on your account. This generate Search Vectors for Assets and enabled Semantic Search in your account

To check or enable this:

1. Navigate to Admin > Settings > Automation
2. Check that Semantic Search is enabled

Once enabled, your new assets will automatically generate Search Vectors as they are uploaded. However existing Assets in your account prior to Semantic Search being enabled will have to be Indexed. The Admin > Assets > Semantic Search page can be used to generate Search Vectors for existing Assets.

Note: Semantic Search is not available to Public Users (any user not logged in) to help prevent rate limitation issues and abuse. Noting that every embedding process and every semantic search consumes some tokens.

Monitoring usage

AI Tokens used for the AI embedding process (generating Search Vectors) and tokens used for semantic search are recorded and displayed in the Reports section and your account dashboard.

To see how often semantic search is being used and the queries being run, go to:

• Admin > Reports > OpenAI calls
• Admin > Reports > OpenAI embedding calls

Tips

• The quality of semantic search improves with good asset metadata — titles, descriptions, and tags all contribute to the index.
• Assets without any metadata are harder to surface via semantic search. We recommend auto-tagging and auto-descriptions are enabled (AND run across all assets ) for all Asset Types before running Search Vector generation.