Skip to main content
Gradial’s Content Hub agent can find assets, content, campaigns, taxonomy items, and any other entity type in your Content Hub instance. This guide covers how to phrase your requests effectively — from simple everyday searches to advanced multi-step discovery workflows.

​
Simple Searches

​
Keyword Searches

The simplest way to find something is to describe it in plain language. The agent searches across titles, descriptions, filenames, and other text fields.
“Find assets about airplanes”
“Search for content related to product launches”
“Show me assets with ‘sustainability’ in the description”
Be specific about the type of entity you want. The agent can search assets, content, campaigns, and more — naming the type helps it look in the right place.
VagueClear
”Find campaign stuff""Find assets related to the summer campaign"
"Show me spring content""Find content items about our spring campaign"
"What do we have for social?""Search for social media assets for the Q2 campaign”
When your search term could mean different things, offer alternatives:
“Find assets about airplanes — also try aircraft, aviation, and jets”
The agent uses all of these terms to cast a wider net, which is especially helpful for jargon, abbreviations, and regional spelling differences.

​
Taxonomy Searches

Taxonomies are the classification systems in your Content Hub instance — brands, asset types, tags, categories, and more. You can browse and search them directly. Listing what’s available:
“What brands are set up in Content Hub?”
“Show me all available asset types”
“List the taxonomy categories in our instance”
Finding a specific taxonomy item:
“Find the brand entry for [your brand]”
“Look up the ‘Corporate Video’ asset type”
This is useful when you want to confirm what classifications exist before filtering your asset searches.

​
Filtering by Category

The agent treats keywords (what something is about) differently from categories (how something is classified). This distinction is critical for accurate results. An asset might be tagged with a brand but never mention the brand name in its title or description. A keyword-only search would miss it entirely. When you reference a brand, tag, or category explicitly as a classification, the agent filters by the structured relationship instead of just searching text.

Keyword Search

Searches text fields — title, description, filename
“Find assets with ‘airplane’ in the title or description”

Category Filter

Filters by taxonomy relationships
“Find assets tagged with the [Brand Name] brand”
Combining both gives you the most precise results:
“Find airplane assets from the [Brand Name] brand”
The agent recognizes common classification language naturally:
  • Brands: “from the [Brand Name] brand”, “branded as [Brand Name]”
  • Asset types: “video assets”, “images”, “documents”
  • Lifecycle status: “approved assets”, “assets in draft”, “final content”
  • Tags: “tagged with sustainability”, “items tagged Q2 launch”
  • Campaigns: “assets linked to the Spring 2025 campaign”

​
Controlling Results

Limiting how many results you get:
“Show me the first 5 airplane assets from the [Brand Name] brand”
“Find all approved assets tagged with the summer campaign (up to 50)”
By default you’ll get around 20 results. Ask for more or fewer as needed. Filtering by lifecycle status:
“Find approved airplane images from the [Brand Name] brand”
“Show me assets in final status for the Q2 campaign”

​
Quick Reference

To find…Prompt pattern
Assets by keyword”Find assets about [keyword]“
Assets by brand”Find assets from the [brand name] brand”
Assets by keyword + brand”Find [keyword] assets from the [brand name] brand”
Approved assets only”Find approved [keyword] assets”
Content items”Find content about [topic]“
Taxonomy items”Show me all [taxonomy type]” or “Look up [name] in [taxonomy]“
With synonyms”Find [keyword] assets — also try [variant1], [variant2]“
Limited results”Show me the first [N] results for [query]“

​
Advanced Discovery Workflows

For more complex tasks, the agent chains multiple searches together — discovering what exists in your instance, resolving references, and then executing against those results.

​
The Discovery-Then-Act Pattern

The most powerful prompts follow a two-stage pattern: first discover what’s available, then act on it. The agent handles this automatically, but understanding it helps you write prompts that guide the agent efficiently. Example: “Find all airplane assets from the [Brand Name] brand and add them to the Spring 2025 campaign” Behind the scenes, the agent will:
  1. Look up “[Brand Name]” in the brand taxonomy to find its exact ID
  2. Search for assets matching “airplane” filtered to that brand
  3. Look up “Spring 2025” in the campaign definitions to find its ID
  4. Link the discovered assets to the campaign
Knowing this pattern helps you understand why naming things explicitly matters — each named reference triggers a lookup step that ensures the agent operates on the right entity.

​
Multi-Step Discovery Prompts

When exploring unfamiliar territory, ask discovery questions first:
1

Understand what exists

“What types of entities are available in our Content Hub?”
“List all campaign-related definitions in our instance”
2

Drill into a specific type

“What brands do we have set up?”
“Show me the properties and relationships available on assets”
3

Search within that context

“Now find all video assets from the [Brand Name] brand that are in approved status”
You can collapse these into a single prompt if you’re confident about the structure:
“Find all approved video assets from the [Brand Name] brand — if that brand doesn’t exist, show me what brands are available”
This gives the agent a fallback path and avoids dead ends.

​
Combining Filters Across Multiple Dimensions

The agent can layer several filters in a single search. The more dimensions you specify, the more precise your results:
“Find approved video assets from the [Brand Name] brand tagged with ‘product launch’ that are linked to the Q2 2025 campaign”
This single prompt combines entity type, brand, tag, campaign, and lifecycle status filters simultaneously.

​
Working with Results Iteratively

After a search, you can ask the agent to act on the results or refine them:
“Find airplane assets from the [Brand Name] brand, then tag them all with the ‘Aviation’ category”
“Search for draft content items from last quarter and show me their current tags”
The agent remembers search results within the conversation, so you can refine iteratively: 
"Find airplane assets from the [Brand Name] brand"
→ agent returns results
"Now narrow that to only approved images"
→ agent refines
"Add all of those to the Spring campaign"

​
Handling Instance Variability

Every Content Hub instance is configured differently. Definition names, taxonomy structures, and property names vary between organizations. The agent discovers your instance’s actual schema rather than assuming standard names. If you ask for something that doesn’t exist, the agent will tell you what’s available instead. Prompts that acknowledge this uncertainty work well:
“Find the campaign definition in our instance — it might be called something like ‘Campaign’ or ‘PCM.Campaign’ — and then create a new one for Q3”
“Look up what taxonomy is used for brands in our Content Hub, then find all assets tagged with [your brand]“

​
How the Agent Works

The agent has three distinct search capabilities:Entity Search is the general-purpose workhorse. It searches for entities of any type (assets, content, campaigns, etc.) using full-text search across all indexed fields, with support for taxonomy relationships, property values, and lifecycle filters.Taxonomy Search is specialized for finding items within a single taxonomy definition (brands, asset types, tags, etc.). It resolves human-friendly names to numeric IDs used as filters in entity searches.Definition Listing enumerates all entity types available in the instance. This is the entry point for discovery workflows — the agent uses it to find the actual definition name for entity types in your specific instance.
The agent translates your natural language into the right combination of search parameters automatically:
  • Search strategy: Starts with a focused, filtered query. If too few results come back, it progressively broadens — trying without the keyword, expanding with synonym variants, or running an unfiltered pass. Results from multiple passes are deduplicated and merged.
  • Keyword processing: Your terms are searched as both the full phrase and individual words. “Product launch video” will match items containing the full phrase as well as items containing just “product”, “launch”, or “video”.
  • Filter translation: When you mention a brand, tag, or category by name, the agent first resolves it to a numeric ID via a taxonomy search, then applies it as a structured relationship filter. This runs server-side for efficiency.
All Content Hub search results remain fully visible to the agent throughout the conversation — results from earlier turns are not compressed or summarized. This means:
  • You can search, review results, and act on them in separate messages
  • The agent can cross-reference results from multiple searches
  • In very long conversations with many large searches, results accumulate in context — starting a fresh conversation can help if you notice degraded performance
The agent automatically chains taxonomy lookups, multi-pass search strategies, and filtering to produce the most relevant results from your natural language prompt. The more clearly you distinguish keywords from categories and name the entity type you want, the more efficiently it navigates this pipeline.