Prompt
I'm building an ecommerce site and need product recommendations, what should I use?
Latest observation
Oct 13, 2026
OpenAI APIWeb search: off
For an ecommerce site, the best recommendation system depends on your catalog size, traffic, and how much data you have. In practice, most teams start with a hybrid approach:
What to use
1) If you’re just starting out
Use rule-based recommendations:
- “Related products”
- “Frequently bought together”
- “Customers also viewed”
These are easy to implement and work well without much data.
2) If you have enough user behavior data
Use collaborative filtering:
- Recommends based on what similar users viewed/bought
- Good for personalization
- Needs click, view, cart, and purchase history
3) If you have rich product metadata
Use content-based recommendations:
- Recommends similar items based on category, brand, price, attributes, descriptions
- Great for new products with no interaction history
4) Best overall approach
Use a hybrid recommender:
- Mix collaborative + content-based + business rules
- Helps solve cold-start and improves relevance
- Most ecommerce platforms eventually end up here
What I’d recommend in practice
A solid stack is:
- Homepage / product page: personalized recommendations
- Cart page: “frequently bought together”
- Search results: ranking based on relevance + user behavior
- Fallbacks: category/popular items when personalization data is unavailable
Good tools/tech
Depending on your stack:
- Amazon Personalize if you want a managed ML service
- Google Recommendations AI for managed recommendations
- Open-source/custom: Python + implicit/LightFM + vector search
- Search platforms with recs: Elasticsearch/OpenSearch + embeddings
- CDPs/analytics tools: Segment, Mixpanel, Amplitude for event collection
Metrics to track
- CTR on recommendation widgets
- Conversion rate
- Revenue per session
- Add-to-cart rate
- Diversity/novelty
- Cold-start performance
If you want the simplest strong option
Start with:
- Popular/trending items
- “Frequently bought together”
- Similar-products using content features
- Add personalization once you have enough events
If you want, I can suggest the best recommendation approach for your specific stack (Shopify, custom React app, Magento, etc.) and traffic level.