The semantic algorithm is an AI-powered recommendation algorithm available through Algorithm Studio that helps you create themed product recommendation experiences using natural language prompts.
Instead of manually curating recommendation collections or relying primarily on historical shopper behavior, you can describe the type of experience you want in your own words. This ensures you can still deliver relevant, personalized experiences even when behavioral or contextual data is unavailable, such as for new or anonymous visitors.
For example, you can create recommendation experiences around ideas such as:
- Back-to-school essentials for students
- Minimalist work-from-home setup
- Weekend travel packing list
- Home fitness starter kit
This is especially useful for:
- New or low-traffic products with limited behavioral data
- Seasonal or campaign-based collections
- Lifestyle-oriented merchandising
- Brand storytelling and thematic experiences
Notes:
- You must have Algorithm Studio to use this algorithm.
- You can create up to 30 semantic algorithms.
How it works
A semantic algorithm finds products that match the idea behind your theme. For each prompt, the algorithm:
- Breaks the prompt into intents.
- Applies each intent to discovery engines.
- Surfaces the most relevant items based on semantic understanding of product data, including text and images.
For example, with the prompt "Back to school," the algorithm surfaces products that best fit the idea of going back to school, both in how they look and in how they're described, without you needing to specify any items or item categories.
Create a semantic algorithm
- In Algorithm Studio, click Create Algorithm and then select Semantic.
-
In the What should your algorithm recommend? field, enter a theme description or select a suggested theme.
- Review the preview. The preview displays products with their image, name, SKU, groupid, and price, so you can understand the type of products the algorithm serves. Refresh the results or adjust the theme as needed.
Coming soon: Boost your semantic algorithm with a ranking algorithm - Click Save Algorithm.
After you save the algorithm, it needs to sync for up to 24 hours before it's available for use in recommendation strategies.
Use your semantic algorithm in a strategy
- In Experience OS, go to Assets › Strategies, and create a new strategy.
- From the algorithm selector, in the Algorithm Studio section, select your semantic algorithm.
- Optional: Add filter rules on top of the algorithm to narrow the results. See example that follows.
- When you're done configuring the strategy, click Save.
Example: To show only products under $20 from a "Back to school" semantic algorithm, select the algorithm in the strategy and add a filter rule for a price under $20.
Important to know: The semantic algorithm understands thematic ideas only. It doesn't understand business logic, so a prompt such as "Back to school under $20" doesn't work. Enter the theme ("Back to school") in the algorithm, and add business conditions such as price ("under $20") as filter rules in the strategy.
Boost semantic results with a ranking algorithm
Note: This capability is planned for release around the end of November.
By default, semantic algorithms order products by semantic relevance. However, for more control over the results, you can use a ranking algorithm as a boosting factor to control how products are ordered within your theme. The semantic algorithm still decides which products fit your theme, and the ranking algorithm decides the order in which they appear.
Ranking algorithms are a separate algorithm type that you create in Algorithm Studio. Only ranking algorithms with an Active status are available as boosting factors.
- In the preview, open the ranking algorithm drop-down next to the theme. By default, it displays No Ranking Algorithm.
- Select an active ranking algorithm.
The preview updates to show the products in ranked order. - Click Save Algorithm.
In the Algorithm Studio dashboard, the Used By column shows that a ranking algorithm is in use when a semantic algorithm uses it.
Best practices
- Use descriptive, specific themes instead of broad concepts
- Include style, mood, season, or use-case language when relevant
- Use filter rules for business conditions such as price instead of adding them to the theme
- Ensure your catalog includes rich product titles, descriptions, and attributes
- Review generated previews before saving algorithms