Product feed ads in ChatGPT: how to prepare your e-commerce business

ChatGPT is moving closer to an area online stores know well: the use of catalogues, product attributes, availability, pricing and purchase-intent signals to present relevant commercial offers. The introduction, in beta, of advertising campaigns created from product feeds in ChatGPT Ads represents more than the arrival of a new advertising format. It signals that structured catalogue data will play a decisive role in product discovery within conversational experiences.

For retail brands, manufacturers, distributors and businesses with e-commerce operations, this development requires a change in perspective. A catalogue should no longer be regarded merely as an internal database or a technical requirement for advertising platforms. It should be treated as a structured representation of the commercial offering, capable of explaining to artificial intelligence systems what each product is, who it is intended for, what problem it solves, how much it costs and under what conditions it can be purchased.

In practice, catalogue quality may influence the ability of a conversational platform to identify, compare and present relevant products. Incomplete titles, generic descriptions, outdated images, poorly grouped variants or inconsistent prices are no longer merely operational shortcomings. They become obstacles to visibility, advertising effectiveness and conversion.

What are product feed ads?

Product feed campaigns allow an advertiser to upload a structured catalogue to an advertising management platform. Instead of creating each advert individually, the business supplies product data and configures rules that help the system select the most suitable items for each context.

The model is similar to solutions already used by advertising platforms, price comparison websites, marketplaces and recommendation systems. The difference lies in the environment in which the advert may appear. In a traditional search engine, the user tends to enter a relatively short query. In a conversation with artificial intelligence, they can describe a much more detailed requirement.

For example, someone may ask for lightweight running shoes suitable for half-marathon training on asphalt, with a particular level of cushioning and within a specific budget. To identify a relevant product, the system needs to understand more than a category. It must interpret technical attributes, intended use, variants, materials, price, stock and trust signals.

For this reason, a feed prepared for conversational environments should not be limited to the minimum required fields. The greater the semantic quality of the data, the better the system will be able to connect a product with a specific need.

The catalogue becomes a commercial source of truth

For many years, numerous businesses treated their catalogue as a fragmented structure. The e-commerce platform held certain data, the ERP contained other information, images were stored in a separate system and each advertising channel received a different version of the data.

This model creates inconsistencies. A product may appear with one price in an advert, another on the landing page and a third condition in the billing system. It may be presented as available when it is already out of stock or use an image that does not match the selected variant.

In environments powered by artificial intelligence, these shortcomings become even more problematic. AI needs consistent sources to interpret an offering correctly. When it finds contradictory, incomplete or outdated information, confidence in the product decreases and the risk of delivering a poor experience increases.

The catalogue should therefore operate as a commercial source of truth. Prices, stock, titles, descriptions, variants, return policies, delivery times and seller information should follow clear rules and be updated through automated processes.

Campaign size and availability

According to the information made available during the beta phase, catalogue-based campaigns are primarily intended for advertisers with extensive or frequently changing product ranges. The feed should contain at least 1,000 eligible products and may include up to 2 million items.

This range benefits medium-sized and large online stores, distributors, brands with numerous variants, international operations and specialist marketplaces. Sectors such as fashion, electronics, sport, home decoration, beauty, automotive parts, industrial equipment and business supplies may find this an efficient way to activate large catalogues.

According to the reference material, ChatGPT Ads was in a beta phase with availability limited to selected markets. This limitation should not lead businesses to postpone their preparations. Cleaning and standardising catalogue data are valuable investments for multiple channels, even before any potential expansion of the platform.

A business that prepares its data today will be in a stronger position to integrate with new advertising ecosystems, AI-powered discovery tools, comparison services, recommendation systems and conversational commerce experiences.

How a catalogue campaign works

The process begins with the creation of a feed representing the available catalogue. This file is uploaded to the advertising platform through a secure connection and submitted for validation. Once the data has been processed, the advertiser can create a campaign, select the catalogue and organise the products into groups.

Groups may use filters based on category, brand, margin, availability, price, collection, product line, country or another relevant property. This segmentation prevents every item from receiving the same level of investment and helps align advertising with business priorities.

The platform may use the title, description, image, price and landing page available in the catalogue to compose adverts. This automation increases scale, but it also amplifies any error present at the source. When a title is incorrect across thousands of variants, the problem will be reproduced throughout the campaign.

Before activation, the team should confirm the number of eligible products, image quality, landing-page consistency, segmentation rules, budget, bidding strategy and conversion measurement.

Essential technical feed requirements

Catalogue delivery should follow a stable and predictable structure. The indicated model corresponds to a complete export, or full snapshot, in which each update represents the current state of all eligible products. Instead of sending only changes, the business replaces the previous version with a new complete snapshot of the catalogue.

The update should take place at least once a day. Businesses with highly dynamic prices, short-term promotions or rapid stock turnover may need to update their data more frequently.

The most relevant technical considerations include:

• Secure delivery: configuration of an SFTP connection with appropriate authentication.
• Encoding: use of UTF-8 to preserve accents, symbols and multilingual content.
• Format: preference for Parquet, although compressed formats such as JSONL, CSV or TSV may also be used.
• Stable filename: use of the same filename for every update.
• Sharding: division of very large catalogues into files of a controlled size.
• Validation: an initial test using a small sample before carrying out the full export.

A prudent approach is to begin with approximately 100 representative products. The sample should include simple items, variants, discounted products, out-of-stock items, pre-order products and different categories. This variety helps identify errors before millions of records are processed.

Required fields that deserve attention

An effective feed must identify every product unambiguously. The item identifier should be unique, stable and maintained over time. Changing this value unnecessarily may break performance history and make it harder to match information across systems.

The main information blocks include:

1. Identification: unique identifier, title, description and product-page address.
2. Brand: the correct and consistent name of the manufacturer or commercial brand.
3. Image: an accessible, clear main image that represents the variant.
4. Price: the current amount accompanied by the appropriate currency code.
5. Availability: a clear indication of whether the product is in stock, unavailable, on pre-order or on back order.
6. Seller: the name of the commercial entity and the address of its website.
7. Returns: the policy applicable to the product or store.
8. Markets: the countries in which the product may be advertised and purchased.

Titles should be clear, readable and distinctive. Excessive capitalisation, keyword repetition or the inclusion of promotional messages without context harms readability and reduces data quality.

A strong structure may include the product type, brand, model and a distinguishing attribute. However, there is no universal formula. A fashion store may need to include gender, garment type, collection and colour, while a technical distributor may benefit from including the reference, compatibility and capacity.

Descriptions should address real intentions

A product description prepared for artificial intelligence should explain the item accurately. A succession of adjectives or a repetition of the category name is not enough. The text should help both the system and the customer understand when the product is useful.

It is advisable to include tangible benefits, usage context, materials, dimensions, compatibility, care instructions, limitations and differences from similar variants. The aim is not to produce an excessively long text, but to provide enough information to answer the questions that arise before a purchase.

This approach brings catalogue management closer to SEO focused on organic traffic. Both search engines and generative systems need to understand the relationship between a page, an entity and a need. The difference lies in how the answer may be presented to the user.

Optional fields that increase relevance

Optional fields can distinguish a basic catalogue from a genuinely competitive one. Product categorisation, materials, dimensions, weight, additional images, videos, reviews and frequently asked questions provide context that may improve the match between a product and an intention.

Variant data is also essential. Size, colour, gender, capacity, finish or voltage should be organised consistently. Each variant needs its own identifier, although it remains connected to the corresponding product group.

Other useful fields include:

• Promotional price and promotion validity period.
• Shipping costs, methods and delivery times.
• Number of reviews and average rating.
• Related products, accessories and compatible alternatives.
• Region-specific availability and pricing.
• Internal metadata for advertising segmentation.

These elements make it possible to create more intelligent campaigns. A business may, for example, assign greater priority to products with stronger margins, reduce investment in items with limited stock or highlight a seasonal collection in selected markets.

Catalogue quality: the first audit

Before considering campaigns, the business should assess the quality of its information. The audit should identify duplicate titles, descriptions that are too short, missing images, broken URLs, products without a brand, inconsistent categories and variants that do not follow the same naming model.

It is also important to confirm that the data matches what the customer finds on the product page. An advert displaying a different price from the website creates mistrust. An image showing a different colour increases the likelihood of abandonment. A product advertised as available when it is out of stock wastes budget and damages the experience.

The audit should include automated validation rules. Instead of relying exclusively on manual checks, the business can establish alerts for missing prices, unusual stock reductions, inaccessible images, empty required fields or sudden changes in the number of exported products.

Data architecture and system integration

A catalogue may be supplied by the e-commerce platform, an ERP, a PIM, a DAM or an integration layer. The best architecture depends on the scale of the operation, the number of markets and the complexity of the products.

The main objective is to define which system is responsible for each piece of data. The ERP may control pricing and stock, while the PIM stores titles, descriptions and attributes. The e-commerce platform presents the information to customers, and the DAM manages images and videos.

Without this definition, channels receive contradictory versions. The creation of a feed for ChatGPT Ads should therefore form part of a broader data-governance strategy.

For Shopify stores, this preparation may involve applications, custom functions, integrations with external systems and specific rules for markets, currencies and variants. BYDAS's experience in Shopify integrations makes it possible to connect catalogue structure with commercial, logistics and advertising processes.

SEO, GEO and conversational commerce

The use of catalogues in conversational advertising is connected to the evolution of SEO for generative environments. GEO, or Generative Engine Optimization, seeks to improve the ability of artificial intelligence systems to understand, select and use information from a brand.

In a product context, the question is no longer limited to the position of a page for a particular search term. It also concerns whether an AI system can recognise the product as a suitable solution to a need expressed in natural language.

To achieve this, the store needs a logical category architecture, standardised attributes, structured data, useful content and pages that answer specific questions. The product page should explain the item and reduce purchase uncertainty.

Although products uploaded for beta campaigns were not automatically included in organic responses, catalogue preparation remains relevant. The same principles of clarity, consistency and structure will be useful for other search mechanisms, recommendation systems and future AI-assisted shopping experiences.

The landing page still determines conversion

A relevant advert can attract a potential customer, but the sale depends on the post-click experience. The landing page should remain fully consistent with the data presented in the advert.

The price, variant, availability and promotion must match. The page needs to load quickly, adapt to mobile devices and display essential information without requiring a lengthy search.

A strong product page should include:

• Clear images representing each variant.
• A description focused on benefits and use.
• Price, promotion and conditions presented without ambiguity.
• Stock and estimated delivery time.
• Exchange and returns policy.
• Reviews, frequently asked questions and trust signals.
• A visible purchase button and a straightforward checkout journey.

When the page fails to deliver on the promise of the advert, the campaign may generate traffic without producing results. Feed optimisation and improvements to the shopping experience should therefore progress together.

Measurement focused on profitability

Standard advertising metrics, such as impressions, clicks, click-through rate, cost per click and conversions, help assess performance. However, an e-commerce analysis should not end with these indicators.

The campaign must be connected to margin, logistics costs, return rate, repeat purchase and customer lifetime value. Two products with the same sales volume may deliver very different levels of profitability.

UTM parameters on landing pages help identify campaign traffic. Conversion-event configuration should cover the entire journey, from the product-page visit to the purchase, without overlooking intermediate actions such as adding an item to the basket or starting checkout.

A mature strategy also uses commercial data to decide which products should receive greater investment. Items with strong margins, sufficient stock and a high conversion rate may be placed in priority groups. Products with high return rates, limited availability or expensive shipping may require more restrictive rules.

Common mistakes when preparing feeds

One of the most common mistakes is to regard the feed as nothing more than a file to be exported. This view ignores the relationship between data, technology, content and commercial strategy.

Other recurring problems include:

• Using identifiers that change with every update.
• Using identical titles for different variants.
• Copying manufacturer descriptions without adaptation.
• Using low-resolution images, unsuitable backgrounds or incorrect products.
• Failing to synchronise price, promotion and stock.
• Failing to create rules that exclude unprofitable or unavailable products.
• Using URLs that redirect to categories or non-existent pages.
• Completing fields with generic values that add no context.
• Relying on manual validation without automated alerts.

Another mistake is to optimise the catalogue for only one channel. The business should maintain a central information layer and adapt the export to the rules of each platform. This avoids duplicated work and reduces the risk of discrepancies.

A step-by-step preparation plan

Preparation for conversational advertising can be organised into five stages. The first consists of mapping data sources and identifying the systems responsible for pricing, stock, content, images and variants.

The second stage is the audit. The team analyses catalogue quality, defines required fields and creates validation rules. At this stage, the percentage of complete products should be calculated and the categories with the most shortcomings identified.

In the third stage, the business standardises titles, categories, attributes and identifiers. It also reviews descriptions, images, product pages and commercial policies.

The fourth stage focuses on automation. The catalogue is exported at an appropriate frequency through a secure, monitored process capable of generating alerts.

Finally, the fifth stage includes sample testing, technical validation, the creation of product groups, the definition of commercial priorities and the implementation of measurement.

Checklist for online stores

Before making a catalogue available to a conversational platform, the business should answer the following questions:

• Are there enough products to meet the eligibility requirements?
• Does every product and variant have a unique, stable identifier?
• Do the titles distinguish the items correctly?
• Do the descriptions explain benefits, uses and limitations?
• Do the images match the displayed variant?
• Do the pages use HTTPS and load without errors?
• Are the price and currency up to date?
• Does the stock information match actual availability?
• Are variants grouped consistently?
• Are shipping and return policies clear?
• Are relevant reviews and frequently asked questions available?
• Can products be segmented by margin, category or priority?
• Is catalogue updating automated?
• Do landing pages remain consistent with the feed?
• Is click, conversion and revenue measurement configured?

The answers help determine whether the business is ready to launch campaigns or whether it first needs to improve its catalogue infrastructure.

The feed as a new conversational shop window

Catalogue ads in ChatGPT show that digital commerce is preparing for a new stage. Customers no longer depend exclusively on short searches, category filters or recommendations on social networks. They can describe what they are looking for, explain their constraints and ask for help comparing alternatives.

In this environment, the catalogue functions as a shop window that artificial intelligence can interpret. For a product to be considered, the information must be clear, complete, up to date and consistent with the experience provided on the website.

Competitive advantage will not depend solely on advertising budget. It will depend on the ability to combine data, technology, content, user experience and profitability analysis. Businesses that treat the catalogue as a strategic asset will be better prepared to take advantage of search and shopping channels that are still developing.

BYDAS helps brands and online stores organise catalogues, integrate platforms, optimise product pages and create performance-focused campaigns. Discover our digital marketing solutions and prepare your business for the next generation of search, advertising and conversational commerce.