AI assistants are becoming a place where people shop. A customer asks ChatGPT or Google's AI Mode to find a product, and the assistant returns real options — with prices, availability and a way to buy. The question every merchant now faces is simple: when an assistant looks for products like yours, does it find them?
The answer comes down to your product catalog. This is a practical guide to making it discoverable — and choosable — by AI assistants.
Why AI assistants can (or can't) see your products
An AI shopping assistant doesn't crawl your website the way a shopper browses it. It reads a structured product feed — a machine-readable file describing each product — supplied to the platforms the assistant draws from. If that feed is missing, incomplete or stale, your products are effectively invisible to the assistant, no matter how strong your brand or how good the product.
This is a bigger gap than most merchants realise. PayPal's Agentic Commerce Pulse Report (498 US merchants, February–March 2026) found that nearly all merchants report at least partial visibility into AI-agent traffic, yet only about one in five have 80% or more of their catalog available as structured, machine-readable data. The traffic is arriving; the readiness usually isn't.
The reassuring part: the requirement is not exotic. It's the same structured, high-quality feed that already powers Google Shopping, Performance Max and Meta — applied to a new set of consumers.
The four things an AI assistant needs from your catalog
1. Descriptive, natural-language titles
Assistants match products to conversational requests like "a waterproof jacket, size M, under €150." A title built for an internal SKU system ("JKT-AW24-BLU-M") gives the assistant nothing to match. A title that states product type, key attributes and defining features ("Men's waterproof jacket, navy, size M, breathable") maps directly onto how people ask. Write titles for the question, not for the warehouse.
2. Complete, structured attributes
Colour, size, material, category, GTIN, brand — the more complete the structured data, the more confidently an assistant can match your product to a specific need and present it accurately. Missing attributes don't just lower relevance; they make the assistant less willing to surface an item it can't fully describe.
3. Compliant, high-quality images
Visuals are part of how the assistant presents options to the shopper. Images need to meet the platform's requirements and show the product clearly. A poor or non-compliant image can get a product filtered out before it's ever shown.
4. Accurate, fresh price and availability
An assistant that surfaces a product which is out of stock or wrongly priced erodes trust — in your catalog and in the assistant. Frequent synchronization of price and availability is what keeps you eligible to be recommended. Stale data is the fastest way to be dropped.
Where the catalog goes: the surfaces and how they connect
Different assistants connect to catalogs through different routes, but the input is always the same structured feed:
- ChatGPT surfaces products and, where enabled, completes checkout through the Agentic Commerce Protocol, the open standard co-developed by OpenAI and Stripe.
- Google AI Mode and Gemini make products discoverable and purchasable through Google's Universal Commerce Protocol.
- Payment providers such as PayPal connect a merchant's catalog to several of these surfaces at once — so a single, well-maintained feed can reach multiple assistants rather than requiring a separate build for each.
The practical implication: you don't optimize "for ChatGPT" or "for Gemini" separately. You maintain one excellent feed and distribute it through the services that connect to these surfaces.
From "technically correct" to "chosen by the agent"
Being present in the feed is the baseline. Being the product the assistant chooses is the goal — and it's decided on the same signals that drive performance everywhere else:
- Relevance of the title to real, natural-language queries.
- Completeness of attributes, so the assistant can match specific constraints.
- Image quality, so the product presents well when shown.
- Data freshness, so the assistant trusts the listing enough to surface it.
None of these is new to a mature feed operation. Agentic commerce simply raises the stakes: the same feed quality that improves ROAS on Shopping now determines whether you exist for a growing class of AI-mediated buyers.
How to prepare — a practical checklist
- Audit your feed for missing attributes, thin titles, non-compliant images and stale prices — the same issues that cause Merchant Center disapprovals also cause AI invisibility.
- Rewrite titles in natural language, leading with product type and key attributes.
- Enrich attributes so that every product carries complete, structured data (including variants).
- Fix image compliance and quality across the catalog.
- Increase sync frequency, so price and availability are always current.
- Distribute through a service that connects your feed to AI assistants, rather than building one-off integrations.
This is feed management, applied to a new frontier. Merchants who already run a disciplined feed operation are closest to being ready; those who don't will find the same weaknesses that hold back their ad performance also keep them out of AI results.
The bottom line
AI assistants shop from structured data, not from web pages. To be discoverable — and choosable — your catalog needs descriptive titles, complete attributes, compliant images and fresh prices, distributed to the surfaces where agents look. It's the discipline of product feed management, pointed at a new and fast-growing channel. Getting it right is how you make sure that when an assistant goes looking for products like yours, it finds — and recommends — you.
→ PayPal Agentic Commerce with Highstreet.io — connect the feed to AI assistants
→ What is agentic commerce and how AI is changing online shopping
→ PayPal Agentic Commerce: what it is and how it works
→ AI Enrichment: optimise product titles and descriptions with AI
→ Auditing & Monitoring: detect feed errors before they cost you visibility