Understand Vertex AI Search tiers
Optimizely Configured Commerce's Search v3, built on Google Vertex AI Search for Commerce, offers advanced, AI-driven search capabilities designed to improve product discovery, increase conversion rates, and boost revenue.
Vertex AI Search uses a tiered ranking system, which you can think of as a ladder with four rungs. There are two parallel ladders: one for Text Search (like a user typing "stainless steel ball valve") and one for Browse Search (like a user going through "Valves & Ball Valves").
Each tier builds upon the data requirements and capabilities of the previous one. You cannot skip tiers, as the AI's learning and optimization at higher tiers depend on the foundational data that lower tiers establish. New implementations begin at Tier 1 and require sufficient time and data accumulation to advance.
See Optimizely's blog post for information: Climbing the Relevancy Ladder: Unlocking Vertex AI Search Tiers.
Catalog data quality
A foundation of high-quality catalog data is key to advancing beyond Tier 1. Poor data quality acts as a significant blocker to unlocking more intelligent search capabilities.
Vertex AI Search has specific standards for catalog data.
- Valid and accessible URIs – At least 95% of products must have valid, accessible Uniform Resource Identifiers (URIs) to enable AI crawling of web signals.
- Comprehensive descriptions – At least 90% of products require comprehensive descriptions for context and relevance.
- Meaningful titles – At least 80% of product titles must contain a minimum of two words for effective semantic matching.
- Unique titles – The system permits less than 50% duplicate titles to ensure unique product identification.
- Searchable attributes – Products should have at least five searchable attributes, such as manufacturer part number, model number, and material type.
- Multi-value words – Do not use multi-value words in exact searchable attributes, as this can impede tier advancement.
- B2B relevance – Include searchable attributes that are highly relevant to B2B queries (like product numbers, manufacturer part numbers, and material type).
Relevancy tiers
Tier 1: Relevance
- What it offers – The AI ranks search results purely by their semantic relevance to the user's query.
- What you need – Primarily, you need your product catalog and customer search queries.
- What you get – The AI orders results by how well they semantically match the query, without considering product popularity or sales history.
Tier 2: Relevance and popularity
- What it offers – This tier enhances semantic relevance by incorporating popularity signals. Among equally relevant items, the system prioritizes products that users click and purchase more frequently.
- What unlocks Tier 2
- Event volume – Over 100,000 text search or browse events within the last 90 days.
- Product join rate – At least 95% of events must successfully link to valid product IDs.
- Attribution tokens – At least 95% of search events require attribution tokens.
- Associated events – At least 70% of search requests should have associated user events.
- Real visitor IDs – Using real visitor IDs instead of hardcoded synthetic data.
- Best practice – Implement attribution tokens from the outset. These unique identifiers track which products the system displayed to users, letting the AI learn from subsequent user actions (like detail-page-view, add-to-cart and purchase-complete).
- Why it matters – Products that resonate with a broader user base rise in the rankings, making search results more useful.
Tier 3: Revenue-optimized ranking
- What it offers – This tier moves beyond popularity to rank results based on purchase likelihood. The AI learns site-wide user behavior patterns to surface products with a higher probability of conversion.
- What unlocks Tier 3
- Detail-Page-View events – Over 250,000 detail-page-view events following search events within the last 90 days.
- Search event volume – Over 200,000 search events within the last 90 days.
- User interaction – Over 250,000 search events with at least one user interaction (detail-page-view, add-to-cart, or purchase from the same visitor).
- Conversion ratios
- Add-to-cart / detail-page-view ≥ 0.02
- Purchase / add-to-cart ≥ 0.025
- Product pricing – At least 95% of products must have prices.
- Product views – Over 100 products must have at least one detail-page-view within the last 90 days.
- Best practice – Upload all user events, not just those directly attributable to searches. The AI learns more effectively from the complete customer journey.
- SEO/AEO connection – The Vertex AI model also incorporates web signals crawled from product URIs, including SEO metadata, Answer Engine Optimization (AEO) structured data, and rich product schemas, further enhancing search relevancy.
- Why it matters – This tier optimizes search results to directly contribute to revenue by prioritizing products most likely to convert based on learned behavioral patterns.
Tier 4: Personalized revenue-optimized ranking
- What it offers – This tier provides personalized search results tailored to the individual user's preferences and historical behavior patterns.
- What unlocks Tier 4
- Attributed search events – Over 100,000 search events with attribution tokens served by Vertex AI Search within the last 90 days.
- Visitor ID matching – Over 10% visitor ID matching between search requests and user events.
- User ID integration – Over 1% of events must have user IDs set for signed-in users (the system calculates this over the last seven days).
- User ID matching – Over 10% of search requests should have user IDs that match corresponding event user IDs.
- Caching avoidance – Less than 1% cached search results.
- Best practices
- For signed-in users, provide both user IDs and visitor IDs to enable seamless cross-device personalization.
- Excessive caching of search results can prevent effective personalization. The system should provide each user with fresh, dynamically personalized results.
- Why it matters – Delivers a highly intelligent and individualized shopping experience, significantly improving customer satisfaction and conversion for repeat visitors.
Browse search
Browse search has its own four-tier ladder that mirrors the Text Search tiers.
- Tier 1 – Random results.
- Tier 2 – Popularity (requires over 100,000 browse events in 90 days).
- Tier 3 – Revenue-optimized ranking (requires over 250,000 browse detail-page-views in 90 days).
- Tier 4 – Personalized revenue-optimized ranking.
You can see this comparison in the following chart:
A key distinction for browse search is the requirement for over 95% of browse requests and events to have exactly matching category and filter values. Monitor text and browse search tiers independently, as they can often be at different levels of advancement.
Get help
Optimizely can help you identify your current tier and overcome any blockers by conducting audits of the following:
- Catalog data quality.
- Event pipeline and attribution configuration.
- Tier-specific blockers and gap analysis
Contact your Customer Success Manager to initiate a Commerce Search v3 tier optimization audit.