Discovery through public pages
Accessible product pages, readable HTML, internal links, sitemaps, images and structured data give systems a public foundation to understand and verify.
Open sourceGuides
How to sell on ChatGPT: prepare ecommerce, catalogue and product feeds for AI
ChatGPT can show products, compare alternatives and direct a user to a merchant. Being there takes more than an “AI ready” badge: it requires accessible product pages, consistent data, an updatable catalogue and, when approved, a feed that follows OpenAI’s specifications.
The short answer
An ecommerce site can prepare on three levels: make products and pages understandable to ChatGPT search, structure a catalogue that can power reliable feeds, and apply for merchant onboarding when the market and programme allow it. A feed improves control and data freshness, but does not guarantee approval, placement or sales. Conversion should first be designed on the merchant’s own website.
Gian Luca Partengo
Web developer since 1995 · bespoke websites · Updated
What “selling on ChatGPT” actually means today
There is no universal switch that turns a store into a ChatGPT channel. There is a chain of data, selection and conversion.
- 01The user expresses purchase intent and ChatGPT interprets the need, constraints and context.
- 02The system can show products with images, details, prices and links to merchants.
- 03A product is selected according to relevance and available data, not because the merchant “installed ChatGPT”.
- 04Purchases normally complete on the merchant’s site or app; any integrated checkout remains limited and programme-dependent.
The three levels of AI commerce are not the same
A serious project separates what depends on the website, what needs a feed and what requires an approved commercial integration.
Direct, current catalogue data
A product feed communicates titles, descriptions, images, prices, availability and variants in a structured form. Onboarding is currently available to approved partners.
Open sourceCheckout and agentic integrations
Sessions, carts, payments, orders and webhooks need APIs, security, idempotency and testing. They are not an automatic feature of a product feed.
Open sourceThe right order
First build an ecommerce site that converts and maintains reliable data; then add distribution channels. Designing a feed before the catalogue merely automates inconsistency.
AI cannot rescue a badly built ecommerce site
ChatGPT can send a qualified visit. The site still has to earn trust, answer questions and complete the order without friction.
Design principle
The product page remains the commercial source
Title, price, availability, variants, shipping, returns and terms should describe the same reality on the page, in structured data, in the database and in the feed.
When sources diverge, this is not only an SEO problem: a customer may see one price and find another, select an unavailable variant or lose trust before checkout.
Indexable architecture
Categories, products and important variants need stable URLs, correct canonicals, internal links and a coherent sitemap.
Performance and mobile
Heavy images, unnecessary JavaScript and unstable interfaces hinder both understanding and conversion, especially on mobile traffic.
Credible checkout
Costs, timing, payment methods, errors and order confirmation must be clear before a customer hands over data and money.
Reachable policies
Privacy, cookies, terms, shipping, returns and contact details are not footer filler: they support commercial trust and the data required by integrations.
A product page should answer before the user asks
Generic descriptions make products and shops interchangeable. Specific information lets search engines and assistants understand who an item suits.
Stable name and identity
A clear title, stable SKU, brand and identifiers such as GTIN or MPN when they exist. Every purchasable variant should be identifiable without ambiguity.
Factual description
Materials, dimensions, compatibility, use, limitations and real differences. Avoid copied manufacturer text when it does not distinguish the offer.
Complete offer
Price and currency, actual availability, condition, delivery time and cost, returns, variants and a consistent purchase CTA.
Useful images
Clear images that represent the selected variant, are accessible and use efficient formats. Images should not compensate for missing written information.
Structured data: the page also speaks to machines
Markup does not replace visible content and must not contradict it. Its job is to state entities, offers and hierarchy explicitly.
| Type | Where | What it clarifies | Common error |
|---|---|---|---|
| Product | Single product or variant page | Name, images, description, SKU, brand and identifiers | Generic markup duplicated across products |
| Offer | Nested inside Product | Price, currency, availability, condition and purchase URL | Price or stock differs from visible content |
| Shipping and returns | Organization or Offer, depending on scope | Delivery services and return policy | Incomplete or unmaintained claims |
| BreadcrumbList | Categories and products | Catalogue hierarchy and context | Path conflicts with navigation and URLs |
| Organization | Seller entity | Name, website, contacts and relevant global policies | Unclear seller or conflicting data |
Product, Offer and Breadcrumb are already part of the Essential E-commerce standard for the agreed scope. No structured data guarantees a rich result, an AI citation or ranking.
A catalogue is a data system, not a grid of cards
If price, stock and variants live only in page copy, creating and maintaining a reliable feed becomes fragile.
Every purchasable product should derive from an authoritative data source. The HTML page, structured data, cart and exports should read the same values, with clear rules for currency, tax, promotions and availability.
OpenAI recommends modelling variants at row level: a stable identifier for the parent product and a unique identifier for each purchasable option. Title, URL, media, price and availability should be variant-specific whenever they differ.
Identifiers
IDs and SKUs do not change on every export. GTIN and MPN are used when they exist, never invented merely to fill a field.
Variants
Size, colour, material or configuration are explicit dimensions. Every purchasable combination has the correct stock, price and images.
Taxonomy
Categories and attributes are coherent across navigation, filters and data. Arbitrary labels make similar items harder to compare.
Commercial rules
Promotions, markets, currency, tax, shipping and returns have a source and validity. No export should have to guess.
One truth, multiple representations
The page, JSON-LD, cart and feed are not four catalogues to update manually. They are four outputs from the same data foundation.
OpenAI product feeds: what is actually required
A feed makes the catalogue more controllable and current, but it requires a pipeline, validation and approved access.
- 01Choose delivery: snapshot files or API, according to volume and update frequency.
- 02Map the official schema: required fields, canonical names, formats, allowed values and variants.
- 03Validate a sample and then the complete catalogue, blocking incomplete records, invalid URLs and inconsistent data.
- 04Update on a real cadence, monitor errors and maintain attribution and quality after launch.
A feed is not a plugin
It is a contract between catalogue and channel
OpenAI documents compressed files and APIs, field checks, valid public URLs and stable records. A one-off export is not enough: the feed is only valuable while it stays close to the store’s real state.
OpenAI evaluates merchant applications. Preparing the schema makes a project ready, but it does not entitle the merchant to onboarding or guarantee that every item will be displayed.
Stale prices and availability destroy trust
Quality is not measured only on the first upload. A recommended product should still be purchasable under the conditions shown.
Complete snapshot
The full feed represents the authoritative catalogue state and should be regenerated on a cadence that matches the store.
Fast updates
Stock, prices and product status may need intraday updates, particularly where availability changes frequently.
Cross-channel consistency
The price in the feed, page, JSON-LD and cart should match. Any difference should be treated as a product defect.
Monitoring
Logs, rejection reports and automated checks should find broken URLs, missing images and rejected records before customers do.
OpenAI warns that price and shipping updates may take time to propagate. Reducing errors at source remains the merchant’s responsibility.
OAI-SearchBot and GPTBot serve different purposes
Allowing ChatGPT search does not require allowing content to be used for training. The directives are independent.
Search
OAI-SearchBot
- It is used to surface websites in ChatGPT search features.
- Allow it in robots.txt and do not block it through the firewall or CDN if search eligibility matters.
- Rule changes can take about 24 hours to be reflected.
Training
GPTBot
- It concerns content that may contribute to training generative foundation models.
- It can be disabled independently from OAI-SearchBot.
- Blocking it is not the same as asking to be excluded from ChatGPT search.
Merchant application: prepare without selling promises
In July 2026 access and availability are not universal. The website roadmap should work before admission as well.
Direct feeds require approval
OpenAI documentation states that product-feed onboarding is available to approved partners through the merchant application.
Regional availability
The merchant page states that Shopping is available in the United States and will expand to other merchants and regions over time.
Shopify and Etsy
OpenAI says Shopify and Etsy catalogues are already integrated, with no extra application or setup required from individual merchants on those platforms.
What can be promised
You can design a ready catalogue, apply as a merchant and integrate a channel when available. You cannot promise access, product inclusion, ranking or an expansion date for a particular country.
Checkout inside ChatGPT: treat it as a scenario, not the foundation
Official pages describe a transition. The project should assume that the order completes on the merchant’s website.
Status on 24 July 2026
OpenAI sources do not describe a uniform rollout
The Help Center says some eligible products and merchants may offer Instant Checkout. The merchant page says OpenAI is prioritising discovery and merchant-managed checkout, with purchases completed on merchant sites or apps.
The prudent choice is a complete first-party checkout, treating ACP, apps or integrated purchase as extensions subject to availability, admission and the current technical specification.
Authoritative sessions
Create, update and complete calls must return consistent cart contents, pricing, tax, shipping, discounts, status and messages.
Payments on your stack
The merchant keeps its PSP, authorisation, capture, settlement and obligations. An agent does not replace commercial responsibility.
Application security
HTTPS, authentication, signatures, idempotency, input validation, safe retries and error handling are foundations, not polish.
Orders and webhooks
Events should keep order, cancellations, shipping, fulfilment and refunds aligned across retries and interruptions.
Six mistakes that make a store “AI ready” in name only
Most begin before the feed and surface again in every channel.
Duplicate descriptions
The same copy across dozens of SKUs does not explain differences, use or why one variant fits.
Conflicting prices
The page, JSON-LD, feed and cart show different amounts or currencies because they derive from separate sources.
Flattened variants
Colours and sizes share an ID or page without specific stock and images, making the purchasable option ambiguous.
Catalogue hidden in JavaScript
Content and markup arrive only after fragile execution or interaction, leaving weak URLs and initial HTML.
One-off feed export
Nobody monitors rejections, stock, URLs or new rules after the first submission. The feed ages while the store changes.
Improvised external checkout
Costs and conditions appear late, errors do not help and AI-originated visits are not measured proportionately.
Essential E-commerce from €4,500: what it already prepares
The foundation is not a ChatGPT plugin. It is a hand-coded shop with a contained catalogue and data organised well enough to power new channels.
Website package
Essential E-commerce · from €4,500 before tax
The package starts with the Pro foundation and includes the agreed shop scope. Technical quality stays consistent with other websites; the catalogue and purchase flow are what change.
- Contained catalogue and product pages
- Categories, products, basket and checkout
- Payment gateway
- Essential order management
- Relevant Product, Offer, Breadcrumb and metadata
- No WordPress, WooCommerce or customer account
| Level | Scope | Package status | Price |
|---|---|---|---|
| Store | Pro foundation, contained catalogue, basket, checkout and essential orders | Included in the agreed scope | From €4,500 |
| SEO and data | URLs, metadata, sitemap and relevant Product, Offer and Breadcrumb | Included across the agreed catalogue | Included |
| OpenAI product feed | Mapping, export, validation, updates and monitoring against the specification | Only when agreed and access is available | Quoted separately |
| ACP or app | APIs, sessions, security, webhooks, orders and conformance testing | Separate advanced development | Quoted separately |
Marketplace, multivendor, customer accounts, booking, ERP, extended catalogues and unagreed integrations remain excluded. No service can guarantee admission or placement in ChatGPT.
Method: build today without depending on a future promise
Every phase produces value for the store even if a direct feed has not yet been approved.
-
01
Offer and catalogue audit
Define markets, products, variants, commercial rules, data sources and actual volume.
-
02
Data model
Organise IDs, SKUs, attributes, media, pricing, stock, shipping and policies in one coherent source.
-
03
Pages and conversion
Build accessible, fast and measurable categories, product pages, basket and checkout.
-
04
SEO, GEO and crawling
Verify HTML, internal links, canonicals, sitemap, robots and structured data on the published website.
-
05
Feed and application
Where relevant, map the OpenAI schema, validate a sample and apply without promising the outcome.
-
06
Updates and controls
Monitor errors, quality, freshness, clicks and conversions, adapting the pipeline to specifications actually available.
The advantage does not expire
A clean catalogue improves the website, SEO, support, campaigns and future feeds. Preparing for AI commerce does not mean becoming dependent on one provider.
Compare the ecommerce website pricing , read how to appear on ChatGPT and AI search or tell me about your catalogue.
Verified primary sources
Features, specifications and availability change quickly. These official sources distinguish current status, requirements and product direction.
-
OpenAI Help Center — Shopping with ChatGPT SearchOpen source
Product results, selection, pricing, merchants, feeds and Instant Checkout.
-
OpenAI — Merchant pageOpen source
Applications, regional availability, integrated platforms, product discovery and merchant-managed checkout.
-
OpenAI Developers — Product feed get startedOpen source
Approved partner onboarding, integration path, files, APIs, updates and prohibited products.
-
OpenAI Developers — Best practicesOpen source
Content quality, URLs, seller, policies, variants and attribution.
-
OpenAI Developers — File upload overviewOpen source
Snapshots, frequency, formats, compression, sharding, removals and validation.
-
OpenAI Developers — Product feed schemaOpen source
Required and optional fields, identifiers, descriptions, URLs and validation rules.
-
OpenAI Developers — Agentic Checkout SpecOpen source
Sessions, endpoints, authoritative cart, security, payments, orders and webhooks.
-
OpenAI Developers — CrawlersOpen source
OAI-SearchBot, GPTBot and ChatGPT-User, robots.txt and published IP ranges.
-
Google Search Central — Merchant listingsOpen source
Product, Offer, price, availability, images, shipping, returns and requirements for purchasable pages.
-
Schema.org — ProductOpen source
Public reference vocabulary for products, offers, identifiers and properties.
Sources checked on 24 July 2026. Availability, admission, specifications and feature behaviour can change: always verify the official pages before integrating.
Frequently asked questions
Direct answers about the limitations most likely to become misleading promises.
Can I already sell products on ChatGPT from outside the United States?
You can prepare the website, catalogue and application, but the merchant page checked on 24 July 2026 states that Shopping is available in the United States and will expand over time. Direct-feed access is also limited to approved partners.
Is a product feed mandatory for appearing?
No. OpenAI says ChatGPT can crawl the website. A feed provides more control, catalogue coverage and fresher data when the merchant is approved.
Does a compliant feed guarantee that products will be shown?
No. Technical compliance, onboarding and structured data do not guarantee selection, ordering, ranking, traffic or sales.
Are ChatGPT shopping results advertisements?
OpenAI states that product results are independently selected and separate from ads. Policies and the product can evolve, so this distinction should always be checked again.
Can users always pay inside the conversation?
No. Official sources describe limited availability and a transition towards merchant-managed checkout. The project foundation should assume purchase on the seller’s website or app.
Are WooCommerce or Shopify required?
No. A feed is a data format, not a CMS. Shopify and Etsy have integrations announced by OpenAI, but a custom ecommerce site can generate compliant pages, structured data and exports without WordPress or WooCommerce.
What minimum data does an AI-ready catalogue need?
Stable identifiers, factual titles and descriptions, public URLs, images, variants, price, currency, availability, seller and consistent policies. Exact fields depend on the current schema and integration method.
Are Product and Offer JSON-LD enough?
No. They make page data more explicit, but do not replace content, crawling, catalogue structure, feeds, commercial quality or updates.
How often should the feed be updated?
OpenAI generally recommends a complete daily file and updates throughout the day through the API. The real cadence should reflect how quickly stock, price and promotions change.
Is the product feed included in Essential E-commerce from €4,500?
The package includes the agreed catalogue, product pages, basket, checkout and relevant Product, Offer and metadata. OpenAI feed mapping and management, APIs or ACP are separate developments unless explicitly included in the quote.
Can feed traffic be measured without Google Analytics?
Yes. OpenAI suggests consistent attribution parameters in feed URLs. Privacy-first analytics can measure them in proportion to the website’s goals and configuration.
Can you guarantee that OpenAI will approve my catalogue?
No. I can prepare data, pages, feeds and technical controls, submit a coherent application and integrate available channels. Approval, selection and rollout remain OpenAI decisions.
Next step
Do you want an ecommerce site ready for new AI channels?
Start with catalogue and conversion: a fast, readable, hand-coded store with structured data and a foundation ready for feeds and integrations when they actually make sense.