How to make your products discoverable to AI shopping agents (2026 Malaysia guide)

Key takeaways:

• Discovery is going machine-first. AI agents don't browse websites: they query product feeds. If your data isn't structured and contextual, agents can't find you.

• Merchants can more easily integrate with major shopping surfaces through built-in connectors such as Google Merchant Center feeds and third-party integrations, while others may need to configure product feeds and APIs manually.

• Airwallex Checkout supports API-based and tokenised payment flows that can integrate with AI shopping agent–initiated checkouts, enabling merchants to accept payments when an agent brings a buyer to purchase.

AI product discovery is changing how customers find and buy products in Malaysia.

Instead of typing keywords into a search bar and scrolling through results, more shoppers are starting their journey through AI shopping agents that can compare options, check availability, and even place orders on their behalf.

This shift is already underway: AI-referred traffic to retailers grew 393% year-on-year in Q1 2026¹, and AI-referred shoppers convert 42% better than search-referred shoppers².

In this article, we'll break down how AI shopping agents are reshaping product discovery, why structured product data matters more than ever, and what Malaysian merchants need to do to stay visible in agent-driven commerce.

Why agents can't find most products today

When a human shops, they navigate visually. They respond to imagery, layout, and brand cues.

When an AI agent shops, it does none of this. It queries structured data — product attributes, pricing, availability, delivery speed — and makes decisions based on how completely and accurately that data answers the user's request.

Most merchant product data was built for humans and search engines, not agents. Here are three reasons why AI agents can't find products today:

1. Catalog gaps

Missing or thin attributes are the most common problem. An agent asked to "find a lightweight running jacket under RM120 suitable for humid weather" needs material, weight, use-case context, and price as structured fields, not buried in a product description paragraph.

Agents evaluate completeness as a proxy for reliability. A well-described product consistently outperforms a poorly described one, regardless of brand.

2. Discoverability gaps

Even with complete catalog data, products that aren't exposed through an ACP product feed, Google Merchant Center, or an MCP endpoint are invisible to agents.

Agents consume structured feeds and API responses. If your products aren't in those feeds, they don't exist for agent-driven discovery.

3. Checkout gaps

Getting discovered is only half the problem. Most checkout flows are designed for humans: form fields, page redirects, CAPTCHAs, manual address entry.

When an agent reaches your checkout, it expects a protocol-compliant endpoint that accepts delegated payment tokens, not a human-facing payment page. Merchants who fix their discoverability but leave their checkout unchanged still lose the sale.

Airwallex Checkout is built to handle agent-initiated payments, and it supports FPX, DuitNow QR, Touch 'n Go eWallet, GrabPay, and other local payment methods Malaysian shoppers expect. Learn more about Airwallex Checkout or sign up for free to start using it.

4 things agents need from your product catalog

Here's what agents need to be able to find your product catalog:

1. Machine-readable product data

Agents query product feeds with constraints set by the user. Standard attributes like the price, availability, and delivery speed are the baseline. What separates agent-optimised listings from generic ones are contextual attributes: occasions, material, use case, size constraints, environmental conditions.

These are the fields that let an agent match "waterproof jacket for hiking in Borneo" to your product rather than a competitor's.

The practical starting point is schema.org Product structured data on every product detail page. The dateModified property matters more than most merchants realise: LLM crawlers use freshness signals as a tiebreaker when two products are otherwise comparable.

Required vs recommended fields for Product structured data:

Beyond PDPs, implement FAQPage structured data for common product questions.

FAQ structured data can help search engines and some AI systems better interpret Q&A content, though how different agents use this markup varies by platform.

2. ACP/UCP-compliant product endpoints

Structured data on your website helps with organic discovery as AI agents crawl and understand your content.

But for agents that operate through chat interfaces like ChatGPT Shopping, Google AI Mode, and Perplexity, you also need your product catalog to be available through the feeds and APIs those platforms rely on. There are two main paths:

ACP vs Google Merchant Center:

For most MY merchants, maintaining both feeds gives the broadest agent coverage: Google Merchant Center for Google AI Mode and Google-indexed agents, ACP for ChatGPT and other OpenAI-protocol agents.

3. Delegated payment token support

When an agent is ready to complete a purchase, it presents a scoped payment token issued by infrastructure like Mastercard Agent Pay or Visa Intelligent Commerce, rather than raw card details.

These tokens are limited by design: single merchant, single currency, maximum spend amount, and a short expiration window (often minutes). They're the agent commerce equivalent of a purchase order: pre-authorised, bounded, and traceable.

Merchants need checkout infrastructure that can receive and process these tokens without routing them through a human-facing payment page. This means protocol-compliant checkout endpoints (either ACP checkout flow or UCP-equivalent) that accept the mandate and complete the transaction programmatically.

4. Agent traffic analytics

Tag agent-sourced sessions and transactions separately from human traffic in your analytics setup. Without this, you can't measure agentic commerce's impact on revenue, identify which products agents surface most often, or know whether your feed optimisations are working.

Agent sessions have a distinctive fingerprint: high-frequency product API queries, near-zero page dwell time, direct checkout initiation, and referrer strings from known AI platforms.

The most common mistake is flagging this as bot traffic and blocking it: merchants who do this are turning away sales.

How to test whether agents can actually find your products

Once you've addressed catalog data and feed submission, do a few quick checks to validate whether the work is visible to agents in practice.

1. Manual agent test

Open ChatGPT, Perplexity, or Google AI Mode and run ten natural language queries relevant to your product category — the kind of request a real customer would make, not keyword searches.

If your products don't surface in categories where they should, you have a feed gap or a structured data gap. Note which queries return competitors and which return nothing; this tells you whether the problem is coverage (you're not in the feed) or relevance (your attributes aren't matching the query intent).

2. Structured data validation

Run your product detail pages through Google's Rich Results Test and the schema.org validator. Missing required fields, incorrect data types, and stale dateModified values are the most common issues.

Fix these before troubleshooting feed-level problems: agents that discover your product via crawling use this data directly.

3. AI evaluations

Write test prompts that mirror your target customer's language and build simple test agents to simulate discovery and purchasing against your catalog.

This surfaces gaps in metadata that validation tools don't catch — for example, a product that passes schema validation but still doesn't match "gifts for runners" because the use-case context is missing from the description.

Airwallex's product team uses this approach internally to surface catalog readiness issues; the same method works for any merchant using freely available agent testing tools.

What agent-ready checkout looks like for Malaysian merchants

Getting your products discovered by agents is the first problem. The second is making sure the checkout can complete the sale when an agent arrives.

Agent-initiated checkout has three requirements that differ from standard eCommerce:

1. Protocol-compliant endpoints

Agents transact programmatically: they call a checkout API, not a payment page.

Your checkout needs to expose ACP or UCP-compliant endpoints that accept an agent's checkout request, create a session, handle the mandate, and confirm the order. This is what "agent-ready checkout" means at the technical layer.

2. Scoped payment token processing

As covered above, agents pay with delegated tokens, not raw card numbers. Your payment infrastructure needs to validate and process these tokens, including checking that the token's scope (merchant, currency, amount, expiry) matches the transaction.

Standard card processing infrastructure doesn't handle this by default.

3. Fraud screening that understands agent traffic

Agent-initiated transactions look different to fraud models trained on human behaviour: no browsing history, direct-to-checkout, high-frequency queries from the same agent.

Merchants need risk systems that can distinguish legitimate agent traffic from malicious bots operating on similar patterns. Blocking one means blocking the other without the right detection layer in place.

For Malaysian merchants, there's an additional consideration: APAC has an average cart abandonment rate of 87%, significantly higher than the global average of 70%³.

For human checkouts, this is driven by payment friction: missing local methods, forced redirects. For agent checkouts, it's driven by protocol friction. Merchants who address both have the strongest conversion profile across both traffic types.

How Airwallex Checkout supports AI product discovery for Malaysian merchants

Improving AI product discovery means having a checkout that can complete the sale when an agent brings a buyer to you. Here's how Airwallex Checkout helps:

• Agent-ready payment acceptance. Airwallex Checkout is built to accept delegated payment tokens from AI agents across ACP, UCP, and other protocols. When an agent initiates a purchase on behalf of a shopper, the transaction completes without friction.

• Local payment methods for human shoppers. Airwallex supports 160+ local payment methods including FPX, DuitNow QR, Touch 'n Go eWallet, GrabPay, Apple Pay, and Google Pay, so the same checkout converts both agent-driven and human traffic without separate integrations.

• Multi-currency settlement across SEA. Hold and settle in up to 12 currencies without forced conversion back to MYR. When agent-referred orders come from across the region, margins don't erode to FX fees.

• One-click checkout via Airi. Shoppers who save their payment details with Airi can complete checkout in a single click on participating merchants.

Frequently asked questions (FAQs)

What's the difference between traditional SEO and AI product discovery?

Traditional SEO optimises for keyword ranking on search results pages: position, click-through rate, meta tags. AI product discovery optimises for being selected and cited by AI agents during their product search. The signals are different: structured data completeness, attribute richness, feed format compliance, and real-time availability matter more than backlinks or domain authority. A product with a complete schema.org markup, accurate inventory data, and contextual use-case descriptions will consistently outperform a better-ranked page with thin structured data when agents are doing the selecting.

Do I need to be on Shopify to appear in ChatGPT Shopping?

No. Any merchant can appear in ChatGPT Shopping by submitting a product feed via the ACP spec or Google Merchant Center. Merchants with existing platform integrations may find this easier through built-in connectors, while others will need to submit feeds manually or via platform-specific tools.

What local payment methods do Malaysian AI agents support?

Mastercard is rolling out Agent Pay in select markets as part of its move into agentic commerce, with early activity across ASEAN. In Malaysia, payments are anchored by PayNet’s DuitNow rail, FPX, and widely used e-wallets like Touch ’n Go and GrabPay. Today, agentic payments are still primarily card-token based, with broader integration into local real-time rails like DuitNow expected as the ecosystem matures. Providers such as Airwallex already support key Malaysian payment methods like cards and FPX and are positioning for emerging agent-driven checkout flows.

How do I know if an agent has tried to buy from my store?

Tag agentic traffic separately in your analytics. ACP and UCP-compliant agents send identifiable headers — look for x-agent-initiated and platform-specific referrer strings from ChatGPT, Google AI Mode, and Perplexity. At the session level, agent traffic has a consistent signature: high-frequency product API calls, no dwell time, and direct checkout initiation with no browsing path. If you're seeing sessions that match this pattern but treating them as bot traffic and blocking them, you're turning away sales.

Sources

1. https://business.adobe.com/blog/ai-traffic-surge-retail-sites-not-machine-readable

2.  https://business.adobe.com/blog/ai-traffic-surge-retail-sites-not-machine-readable

3. https://www.sellerscommerce.com/blog/shopping-cart-abandonment-statistics/  and https://www.statista.com/statistics/477804/online-shopping-cart-abandonment-rate-worldwide/

This publication does not constitute legal, tax, or professional advice from Airwallex nor substitute seeking such advice, and makes no express or implied representations / warranties / guarantees regarding content accuracy, completeness, or currency. If you would like to request an update, feel free to contact us at [[email protected]]. Airwallex (Malaysia) Sdn. Bhd., a company incorporated under the laws of Malaysia with company registration number 201801007747 (1269761-X), is regulated as a licensed remittance business under the Money Services Business Act 2011 (Licence number 00743 with an expiry date of 3 August 2028, an E-Money Issuer and a registered merchant acquirer under the Financial Services Act 2013.)