How to sell on ChatGPT: a guide to preparing your product catalogue

ChatGPT is beginning to occupy a new position in the buying journey. A shopper can explain a need, add constraints, compare alternatives and request recommendations in one conversation. This creates a new discovery surface for manufacturers and retailers: products can appear while the need is still being defined, before the shopper has selected a store or even a brand.

However, selling on ChatGPT does not currently mean opening a separate shop inside the assistant. OpenAI is prioritising product discovery and merchant-managed checkout: ChatGPT helps people explore and compare, while the purchase is completed on the seller’s website or app. According to the official information available when this article was published, shopping is available to ChatGPT users in the United States, with expansion to other markets planned.

The opportunity is therefore to prepare the catalogue, infrastructure and measurement now. Sitelicon helps brands and manufacturers do that: we turn scattered information into a reliable feed, connect the systems that keep it current and design an experience that turns a recommendation into a measurable sale.

What it means to surface products in ChatGPT

When someone asks a commercially relevant question, ChatGPT can present suitable options with images, prices and key details. Selection depends on the conversational context and the information available for each product. This is not a static listing or a keyword exercise: the system needs to understand what the item is, who it is for, which variants exist and where it is useful.

The result can link directly to the product page, where the shopper confirms the conditions and completes the transaction. OpenAI says it does not charge merchants a fee for purchases that begin in ChatGPT and finish on their website or app.

It is also important to distinguish two different capabilities. Organic catalogue discovery is not the same as ChatGPT Ads. The feed describes the offer and supports product representation; advertising is a paid campaign system with separate requirements and metrics.

The feed: connecting your catalogue to ChatGPT

OpenAI may discover information on public pages, but a feed gives the seller more control, accuracy and freshness. It communicates in a structured form which products exist, their price and availability, the image to use and the URL to which a shopper should be directed.

The stable feed specification requires one row per purchasable item or variant and nine core fields:

• Stable item identifier.
• Title.
• Description.
• Product URL.
• Brand.
• Seller name.
• Image URL.
• Availability.
• Price.

These fields are the technical minimum, not the final quality target. Category, colour, size, material, dimensions, GTIN or MPN, return terms, shipping, promotions and reviews can all help describe and differentiate the offer. Identifiers must be genuine and stable; they should never be invented merely to populate a field.

Variants need particular care. A size or colour with its own price, stock or image should be represented by a separate row and retain its relationship to the parent product. Otherwise, a recommendation may lead to an option that does not match the landing page or is no longer available.

Catalogue quality makes products understandable

A catalogue designed for AI shares many requirements with a well-managed PIM, a marketplace and sound SEO:

1. Precise titles. Identify the product without artificial promotional language or keyword stuffing.
2. Factual descriptions. Explain attributes, benefits, compatibility and use cases clearly.
3. Public, representative images. Each variant should show the item the shopper will find after clicking.
4. Current price and availability. A recommendation loses value if it leads to a different price or an unavailable product.
5. Accessible canonical URLs. Product and image addresses must be absolute, public and correctly encoded.
6. Consistent policies. Shipping, returns and seller details should agree across the feed and store.

Technical eligibility does not guarantee display. The system decides which results are relevant to each conversation. Meeting the specification is therefore the starting point; improving depth and consistency is what makes the catalogue more useful.

A connected and enriched PIM makes this easier by centralising attributes, variants, content and images before distributing a controlled version to ChatGPT, ecommerce platforms, marketplaces and other destinations.

Keeping price and stock synchronised

A feed prepared once becomes obsolete quickly. OpenAI recommends combining a complete catalogue snapshot, submitted at least daily, with API updates during the day when price, availability or other data changes frequently. Small catalogues can handle both full and incremental updates through the API.

Integration can use SFTP files or APIs. The right choice depends on volume, architecture and update frequency. In every case, the company should define a source of truth rather than manually editing several versions of the same item.

Before automating the whole catalogue, validate a representative sample: simple products, variants, offers, out-of-stock items and complex attributes. This exposes missing fields, broken URLs, invalid price formats and discrepancies between the feed and the ecommerce site.

Measuring business that begins in a conversation

Visibility matters only when it can be linked to outcomes. Feed URLs can include attribution parameters so analytics can identify sessions, behaviour and conversions originating in ChatGPT. This must be integrated with the company’s consent, analytics and commercial measurement framework.

Useful measures include products and categories receiving visits, session quality, cart and checkout actions, purchases, revenue and margin, market or device differences, and any price, availability or destination errors discovered after the click.

The aim is not to attribute the entire journey to one touchpoint, but to understand the role ChatGPT plays in consideration and conversion.

A practical preparation plan

Product feed onboarding is currently available to approved merchants. Stores whose catalogues are already integrated through Shopify or Etsy do not need to apply separately; other businesses can register their interest on the official merchant page.

While approval or availability in a new country is pending, a brand can make progress through work that also improves its other channels:

1. Audit product data, variants, identifiers, images, pricing and availability.
2. Define the PIM, ERP or ecommerce platform as the source of truth for each field.
3. Improve titles and descriptions for clarity and accuracy.
4. Produce a representative sample using OpenAI’s stable specification.
5. Validate landing pages, canonicals, structured data and policies.
6. Design full catalogue loads and incremental updates.
7. Add attribution and reporting for traffic, conversion and quality.
8. Apply for access and plan a gradual implementation.

How Sitelicon helps companies sell on ChatGPT

Sitelicon combines more than twenty years of experience in ecommerce, marketplaces, product information, integrations and digital acquisition. That cross-functional perspective is valuable in a channel that cannot be solved by one marketing action: reliable data, technology, content, operations and conversion must work together.

Our service can cover catalogue auditing, data modelling, PIM implementation or integration, format transformation, file and API automation, quality checks, product page preparation and subsequent analytics.

We also coordinate GEO/AIO strategy so brands and products can be understood in generative environments. The goal is not merely to satisfy a technical schema. We give each item the context that supports an appropriate recommendation and provide a destination experience capable of turning interest into business.

Start before the channel becomes mainstream

Availability is still limited and the specifications will continue to evolve. This does not remove the opportunity; it defines the responsible way to approach it. Early preparation lets a company resolve long-standing catalogue problems, automate processes and learn how to measure a conversational buying journey before a mass launch creates urgency.

Businesses that complete this work will not only be better prepared to sell on ChatGPT. They will have a stronger product foundation for search, marketplaces, advertising, distributors and every future AI-based interface.

Sitelicon can help you assess catalogue readiness and build the integration in stages: audit, feed pilot, automation, measurement and continuous improvement.