AI Product Search for eCommerce with Orama and OpenAI

The Challenge: UX That Converts for Niche eCommerce

Our client runs a personalized padel t-shirt eCommerce site. They wanted:

• Faster and more relevant product search
• The ability to personalize empty state experiences
• Future integration with an AI chatbot

This wasn’t just about search — it was about making intelligent UX part of conversion rate optimization (CRO).

---

The Stack: Orama.search + OpenAI + Sanity

We chose Orama.search for its:

• Lightweight architecture and speed
• Native support for vector-based search (with OpenAI embeddings)
• Developer-friendly API (can be updated via REST, CLI, SDKs)

It let us go beyond standard search to implement answer engine logic, automations, and personalized search outputs — all while staying fully headless.

Dynamic Product Updates (Sanity + Cron + GROQ)

Instead of manually syncing product data to the vector database, we automated it with a cron job and a GROQ query from Sanity CMS:

// /api/update-orama-db.ts

const manager = new CloudManager({
	api_key: process.env.ORAMA_PRIVATE_API_KEY!,
});
const indexManager = manager.index('xxxxxxxxxxxxxxxxx'); // replace with your index id

export const GET = async () => {
	try {
		const products = await sanityClientRead.fetch(
			GET_ORAMA_PRODUCTS_FOR_UPDATING, // Consulta GROQ de Sanity
		);

		await indexManager.update(products);
		await indexManager.deploy();

		return NextResponse.json('OK', { status: 200 });
	} catch (error) {
		return NextResponse.json({ error: 'An error occurred', details: error }, { status: 500 });
	}
};

Bonus: This system supports updates via CMS actions, CLI tools, or custom scripts.

---

The Search Page: UX Meets Relevance

Here’s a snippet of how we used Orama to offer intelligent search with fallbacks:

const { search } = await searchParams; // Next.js 15 searchParams

const oramaResponse = await oramaClient.search({
	term: search as string,
	mode: 'hybrid',
});

if (!oramaResponse) {
	log.error('Products not found', { oramaResponse });
}

Bonus: This same idea is used on the custom 404 page to search for related products

---

We also leveraged past behavior and wishlist data to:

• Prioritize last-minute-viewed products
• Insert smart call-to-actions inside search result lists
• Suggest relevant bundles for quick checkout boosts

---

CRO Techniques Enabled

By combining AI-powered search with UX intelligence, we implemented:

• Personalized 404 and empty search pages that suggest real products instead of a dead end
• Wishlist-based smart carousels on product and cart pages
• Product slug disambiguation so a typo lands on the right product instead of a 404
• Search result re-ranking based on past behavior, not only text match

These micro-improvements reduced bounce rate and nudged more sessions into purchase flow.

What We Measured

Search work is easy to argue about and hard to prove, so we tracked four numbers before and after:

• Search exit rate: sessions that left from the search page without a product view.
• Zero-result rate: queries that returned nothing, grouped by term to find missing catalog entries or synonyms.
• Search-to-product click rate: how often a result list produced a product page visit.
• Conversion rate of search sessions against the site average.

The zero-result list was the most useful output. Half of the queries were plurals, typos or synonyms the index did not know, which is a data problem, not a search-engine problem. Fixing the index beat changing the engine.

More on the storefront side: 8 tips to optimize your eCommerce.

---

Why Orama Over Meilisearch/Typesense/Algolia?

• Open-source, lightweight and cost-effective
• Built-in support for vector search + OpenAI embeddings
• Native REST + CLI + SDK support for automation
• Developer-focused design and instant local setup

It gave us AI-enhanced flexibility without the SaaS pricing headaches.

---

What’s Next: AI Chatbot Integration

The next phase of this project will bridge Orama with OpenAI to create an in-site product assistant. It will:

• Guide users via chat toward relevant products
• Pull real-time insights from the search index
• Respond to wishlist/product queries

With Orama’s answer engine and embeddings already in place, this will be a smooth evolution.

---

Want to Build Something Like This?

I help eCommerce brands implement intelligent UX with modern stacks — from search and chatbots to CRO and headless architectures.

Let’s Talk | See More Projects