AI shopping product feeds: ChatGPT, Google and Perplexity (2026)

AI shopping product feeds give shopping engines structured product facts, including price, availability, images and the exact item a customer can buy. They complement your website. Google accepts Merchant Center data, while ChatGPT supports direct feeds and commerce integrations and Perplexity runs a merchant program for large retailers. Each route has different requirements. For Australian and US retailers, start with accurate product identities and market-specific offers, then connect each eligible destination. Stock accuracy comes before longer descriptions.

In summary

AI shopping product feeds work best when the feed, product page and checkout describe the same offer.

• Keep a separate record for each purchasable variant, with stable identifiers and current stock.
• Google now documents conversational attributes for product questions, manuals and related products.
• ChatGPT discovery feeds and checkout integrations have separate requirements.
• Check market eligibility separately: a US shopping launch doesn't establish Australian availability.
• Measure correct product appearances and completed orders, alongside feed acceptance.

Documentation review: 27 October 2026 (AEST).

How do ChatGPT, Google AI Mode and Perplexity get product data?

AI shopping engines combine product information through different routes, including merchant feeds, platform integrations and web content. A feed supplies facts in predictable fields. Product selection still depends on how the engine interprets the shopper's request and the available information about each item and its sellers.

Sources: OpenAI shopping documentation, Google shopping announcement and Perplexity's merchant announcement.

Separate the product from the offer. A product identifies the manufactured item; an offer describes a seller's price, stock and purchase conditions for that item.

That distinction matters for manufacturers. Your brand might appear in an AI recommendation while a distributor gets the sale. Record the merchant link too.

Searchmaxxed's technical SEO work covers the website side: crawl access, indexation, product-page rendering and structured data. Your commerce team controls merchant submissions.

What does a ChatGPT shopping feed require?

OpenAI's Stable file-upload specification requires nine basic fields for product discovery, with a separate row for each purchasable item or variant.

Start with the supported schema. OpenAI labels its Draft schema as planning material, rather than a production integration.

Source: OpenAI's Stable product specification, checked 27 October 2026.

Variants need more detail. OpenAI specifies a shared group_id, listing_has_variations=true and a variant_dict containing the selected options. Each variant keeps its own offer.

Keep genuine identifiers too. A GTIN, or Global Trade Item Number, identifies a specific trade item across sellers. An MPN is the manufacturer's part number. Copy them from verified product records.

OpenAI makes GTIN and MPN optional in this schema, while variant-group fields become required when you submit variants. Optional doesn't mean useless.

How do you deliver and update the feed?

OpenAI recommends at least daily updates. Its file-upload route uses full catalogue snapshots through SFTP and accepts Parquet alongside compressed JSONL, CSV or TSV files for your catalogue.

Keep filenames stable. OpenAI's delivery documentation says to overwrite the existing snapshot, rather than create a fresh filename on every run.

There's an important removal rule. OpenAI can retain a product's last processed record for up to 14 days after it disappears from a snapshot. Omission takes time to take effect.

For the next processed snapshot, is_eligible_search=false makes the item ineligible for search. Stock changes belong in availability.

Test withdrawal before rewriting titles. Otherwise, a polished product recommendation can send a shopper straight to your support team to ask why the item no longer exists.

Do Shopify merchants need another application?

OpenAI says Shopify and Etsy catalogues already integrate with ChatGPT, without an additional merchant application. Other merchants can apply for direct feed access through its merchant page.

Check your existing route first. An integration already carrying your catalogue changes the job from building another connection to improving the information it receives.

Discovery and checkout also differ. OpenAI's merchant page prioritises merchant-owned checkout; its product schema treats checkout as a separately enabled integration. A discovery feed doesn't implement payment processing.

How should you optimise Merchant Center for AI Mode shopping?

Optimise Merchant Center by fixing core product data first, then adding useful conversational attributes. Google's newer fields complement the existing specification rather than replace it.

Begin with the offer. Google's product data specification identifies conflicting website and feed data as a cause of disapprovals or limited eligibility.

Requirements vary by category and destination. Have the feed owner validate against Google's specification instead of treating this table as a universal required-field list.

Which conversational attributes should manufacturers add?

Google now documents six optional conversational attributes, including product questions, PDF documents and related-product relationships. Compatibility-heavy catalogues have plenty to contribute.

Source: Google's conversational attribute guidance, checked 27 October 2026.

Google recommends a supplemental data source, although a primary source or Merchant API can also carry these attributes. Existing information doesn't need another copy.

For example, Google says details already in description, product_highlight or product_detail don't need duplication in conversational attributes. Put each fact where it belongs.

This is where page work helps. Searchmaxxed builds and improves the product explanations and schema; the retailer's feed owner maps those verified facts into supported fields.

Can AI write the product feed?

AI can help draft descriptions, but the underlying specifications need a reliable source. Keep generated claims away from product identity, safety certifications and compatibility decisions.

Google also requires disclosure. Its AI-generated content rules specify structured_title and structured_description, with digital_source_type set to trained_algorithmic_media for generated text.

AI-generated images require identifying metadata too. Check that your image processing preserves the required tags before the files reach Merchant Center.

What can retailers submit to Perplexity?

Perplexity's Merchant Program lets large retailers share product specifications so its shopping experience can access current details. Shopify integrations supply product details too.

Confirm the submission contract directly. The official merchant announcement describes the program, but doesn't provide a public field-by-field upload specification.

Your existing catalogue is useful preparation. Confirm each field and delivery requirement. An existing Google feed may need changes before another platform can accept its records or keep the offers current.

Ask the platform or your existing provider to confirm:

• Supported markets and merchant eligibility.
• File or API format and required fields.
• Variant handling and seller identity.
• Stock refresh and product-removal behaviour.
• Available product-level reporting.

Keep checkout separate here too. Perplexity's later shopping announcement describes PayPal checkout and retailers remaining the merchant of record.

For Searchmaxxed, the useful work remains product-page clarity and access, rather than promising admission to somebody else's merchant program.

What changes between Australia and the US?

Check shopper markets and offers separately. A platform may support Australian users without offering every US shopping or checkout feature in Australia.

Sources: OpenAI merchants, Google's Australian launch, Google reporting availability and Perplexity shopping.

Build offers for the destination. Currency, stock, shipping restrictions and return terms should describe what that customer can purchase, rather than the retailer's headquarters.

A manufacturer selling through distributors needs another distinction: brand discovery versus seller selection. Our marketplace SEO guide explains both.

Do product feeds replace structured data and technical SEO?

Product feeds don't replace structured data or crawlable product pages. Google recommends supplying both website structured data and a Merchant Center feed to widen eligibility and verify information.

They do different jobs. Product schema describes the item, while Offer markup describes its commercial terms. The visible page must support those facts.

Google's Product structured data documentation also covers variant relationships and relevant shipping or return information. Our schema markup guide covers implementation.

Check the rendered page. A valid feed can send customers to a page that selects the wrong variant, hides stock information or redirects into another market.

Searchmaxxed fixes those website faults through technical SEO, product-page content and internal linking. A feed upload alone can't repair them.

How would this work for a manufacturer selling in both markets?

Start with a representative product family and test its offers end to end before extending the same configuration across the rest of your catalogue. This manufacturer example is hypothetical.

Consider an industrial equipment manufacturer selling USB and Bluetooth barcode scanners through its own store and distributors, with a charging cradle sold separately.

Compatibility decides the order. A buyer asking for a scanner compatible with an existing workstation needs more than a rewritten product title.

Run an acceptance test before expanding:

1. Open the feed URL without a logged-in session and confirm the intended market and variant.
2. Compare the product ID, price and stock with the page, structured data and checkout.
3. Follow the manual and accessory links to check model compatibility.
4. Change stock in a testable product record and verify the downstream update.
5. Test withdrawal separately from stock changes, including OpenAI's retained-record behaviour.
6. Check whether an AI result recommends the product, the manufacturer or a distributor.

Assign an owner to each mismatch. Product information management owns specifications, while commerce systems own transactional facts. Website and feed teams own their outputs.

That's a workable division. Searchmaxxed can implement the page and schema work without pretending to own warehouse stock or the retailer's merchant relationship.

How do you measure whether AI shopping feeds work?

Measure feed health, visibility and sales separately. An accepted record proves ingestion, while a completed order tells you that somebody could find the product and buy it from a seller.

Google now has dedicated reporting. Its AI performance insights cover English-language shopping queries in Australia and the US, among other supported markets.

Find the report under Analytics → Products → AI performance in Merchant Center to review your organic AI traffic separately from paid advertising. Paid ads don't appear here.

Read the definitions carefully. Google says “Products showing” counts cumulative appearances, so the same product can contribute repeatedly during the selected period.

A second trap matters more. The report can display 100% share of voice when competitor data is unavailable. Check the competitor data first.

Keep category and country consistent. Keep the same buying-question set for repeat tests, and save the product cards and merchant URLs.

Searchmaxxed's managed search work connects site changes with Search Console, GA4 and AI-answer tracking. For retailers, agree the product and transaction measures before implementation starts.

Is Searchmaxxed right for your AI shopping project?

Searchmaxxed suits retailers and manufacturers that need technical SEO and AI search improvements implemented on their website. The fit is strongest when product information, schema and page access need work together.

We build and fix pages. That includes crawl and indexing faults, rendering, structured data, internal links and the product explanations customers need before purchasing.

The delivery arrangements matter too. You keep ownership of the website, repository, content and data, with direct access to the people doing the work.

Growth System and Agentic Search Engineering combine SEO and AI search work. Custom & Enterprise can cover multiple markets, dedicated engineering and custom reporting.

Scope the feed connection explicitly. These plans don't automatically establish an existing OpenAI or Perplexity partnership, merchant approval or an integration with your product information system.

If your pages are sound and you only need scheduled catalogue exports, your existing commerce or feed provider may cover the work. Fix conflicting source records first.

Frequently asked questions

What should you do next?

Start with the product family where incorrect details would cost the most sales or support time. Bring its pages, current feed and target markets to the discussion.

We'll agree responsibilities before proposing the scope.

Let's talk pricing.