OpenAI Commerce for Magento 2
5.0 1 Based on 1 reviewTest out this module for Free
OpenAI Commerce
Check your email
We've sent a confirmation link. Click it to spin up your demo.
Setting up your demo...
This usually takes a few minutes. You can close this modal or wait for the result — we'll also send you an email when it's ready.
You're in the queue
Position: #
All demo slots are currently in use. You'll receive an email as soon as your demo is ready.
Your demo is ready!
Expires:
Credentials have also been sent to your email.
Something went wrong
We couldn't set up your demo. Please try again or contact support.





Powerful Magento 2 OpenAI Commerce Extension for AI Product Discovery
Connect your Magento 2 catalog to ChatGPT and AI commerce platforms
AI-driven commerce is rapidly changing how customers discover, compare, and purchase products online. The OpenAI Commerce extension for Magento 2 helps merchants prepare for this new shopping landscape by generating structured product feeds optimized for ChatGPT and future OpenAI commerce integrations.
The extension automatically generates AI-ready product feeds directly from Magento 2, including product titles, descriptions, images, pricing, stock information, and structured product URLs. Everything is designed to remain publicly accessible, crawlable, and optimized for AI-driven discovery.
Built for merchants, marketers, and Magento developers, the module combines lightweight feed generation, cron automation, diagnostics, and future-ready commerce functionality in one streamlined Magento 2 extension.
Whether you want to increase visibility in ChatGPT product discovery or prepare your store for future conversational commerce experiences, this module provides the technical foundation without unnecessary complexity.

Create a feed

Set seller details

Enrich ChatGPT search


Automatically Generate an OpenAI-Compatible Product Feed
The OpenAI Commerce extension transforms your Magento 2 catalog into a structured product feed optimized for AI-driven product discovery.
The generated feed includes essential product attributes such as titles, descriptions, prices, stock status, image URLs, and structured product links. Product information is synchronized directly from Magento, helping ensure that ChatGPT and future OpenAI systems always receive accurate and current product data.
Merchants can configure category filters, stock rules, store views, and product selections directly from the Magento admin. This allows complete flexibility over which products become discoverable inside AI commerce environments.
Combined with Magento cron automation, feed generation remains fully automated and easy to maintain, even for larger catalogs.
Conversational Purchases Through ChatGPT
The upcoming Buy it in ChatGPT functionality introduces a completely new way for customers to discover and purchase products directly through AI conversations. Instead of browsing through traditional webshop navigation, customers can interact naturally with ChatGPT to compare products, ask questions, receive recommendations, and complete purchases in a conversational environment.
Magento 2 remains the central commerce platform while ChatGPT becomes the intelligent discovery and shopping layer. Product information, pricing, inventory, categories, and order processing continue to be managed entirely within Magento, while ChatGPT provides the customer-facing AI shopping experience.
This creates a more natural buying journey where product discovery, recommendations, and checkout happen within one seamless flow. Customers no longer need to jump between multiple pages or search manually through large catalogs. Instead, they can simply ask questions and receive contextual product suggestions based on their intent.
Simple activation
Enable conversational commerce directly from the Magento admin.
Real-time synchronization
Product data, pricing, and stock availability remain synchronized with Magento 2.
Magento order processing
Orders created through ChatGPT flow back into Magento automatically.
Improved customer experience
Customers can browse and purchase products without leaving the conversation.
This upcoming integration positions Magento 2 merchants for the next generation of AI-powered commerce, where conversational interfaces increasingly become part of the online shopping journey.

Automatic llms.txt Generation for AI Crawlers and Language Models
The OpenAI Commerce extension now includes automatic llms.txt generation to help AI crawlers and large language models better understand your Magento 2 store.
Comparable to robots.txt, the generated llms.txt file acts as a structured Markdown roadmap for AI systems. It can contain references to your store description, CMS pages, categories, and OpenAI product feeds, making important storefront content easier to discover and interpret.
Each Magento store view receives its own dedicated llms.txt file, automatically served through a clean public URL such as yourstore.com/llms.txt.
This helps merchants prepare their Magento 2 stores for AI-native search engines and conversational commerce systems.

Multistore llms.txt generation
Automatically generate separate llms.txt files for every Magento store view.
Dedicated public endpoints
Serve clean public URLs such as yourstore.com/llms.txt for AI crawler access.
Flexible content selection
Choose which CMS pages, categories, and feeds are included in the output.
Admin overview and controls
Manage generation status, previews, and actions directly from Magento.
Scheduled regeneration
Keep llms.txt content synchronized automatically through Magento cron.
Built for AI discovery
Help AI crawlers and large language models understand your storefront structure.
Future-Proof Your Magento 2 Store for AI Commerce
The OpenAI Commerce extension is designed to prepare Magento 2 stores for the next generation of AI-powered shopping experiences.
Whether OpenAI introduces direct merchant submissions, public commerce APIs, or deeper ChatGPT shopping integrations, your product data is already structured, crawlable, and accessible through a public endpoint.
The extension focuses on scalability, flexibility, and long-term compatibility. Product feeds, llms.txt support, crawler diagnostics, and automated synchronization work together to create a stable AI commerce foundation directly inside Magento 2.
This allows merchants to prepare early for conversational commerce while maintaining full control over catalog structure, visibility, and product data quality.

Set It Up Once and Stay AI-Ready
The OpenAI Commerce extension is fully managed from the Magento 2 admin panel. No external middleware, custom API development, or manual feed formatting is required.
Simply install the extension, configure your preferred store views, apply optional product filters, and enable feed generation. The extension automatically handles product mapping, structured feed generation, UTM tracking, and public endpoint generation.
Feeds are updated automatically using Magento cron, while built-in diagnostics help validate whether OpenAI crawlers such as OAI-SearchBot can correctly access your feed.
The lightweight architecture makes the extension suitable for merchants, agencies, and developers looking for a practical and future-ready AI commerce integration.
OpenAI-compatible feeds
Generate structured AI-ready product feeds directly from Magento 2.
Public JSON endpoints
Expose crawlable product feeds for AI crawlers and discovery systems.
Automatic llms.txt support
Generate AI-focused discovery files automatically for every store view.
Built-in crawler diagnostics
Validate whether OAI-SearchBot can access your product feed correctly.
Magento store-view support
Configure separate feeds and filters for each Magento storefront.
Automated cron synchronization
Keep product feeds continuously updated without manual actions.
Technical Details and Who This Magento 2 Extension Is For
The OpenAI Commerce extension is configured entirely from the Magento 2 admin panel. Merchants can manage feed settings, crawler accessibility, product filters, store views, llms.txt content, and feed visibility without custom code.
Feed generation is optimized for AI crawler compatibility and conversational commerce environments. Public feed endpoints and llms.txt files are automatically generated and updated through Magento cron.
The extension supports Magento store-view configurations, structured product feeds, public crawler accessibility, and AI-focused content discovery. Built-in diagnostics help validate that OpenAI crawlers can access your storefront content correctly.
This Magento 2 extension is ideal for merchants preparing for AI commerce, agencies building innovative storefronts, and developers looking for a lightweight OpenAI integration with scalable feed generation and AI crawler support.
SEO, AI Visibility, and Business Impact
AI-powered search and conversational commerce are becoming increasingly important discovery channels for e-commerce merchants. Structured OpenAI-compatible product feeds help improve visibility inside AI-driven shopping experiences such as ChatGPT.
By combining structured product feeds, llms.txt support, automated synchronization, and crawler validation, the extension helps Magento 2 stores become easier to discover, interpret, and surface inside AI-powered recommendation systems.
For merchants, this means better visibility across emerging AI shopping environments. For developers and agencies, it provides a scalable and future-ready Magento 2 foundation for AI commerce integrations.
Frequently
Asked Questions
A list of the most frequently asked questions from the OpenAI Commerce extension for form.
How does the OpenAI Commerce Feed work with Magento 2?
Can I control which store views generate an OpenAI feed?
What format does the feed use?
Is the feed updated automatically?
1 of many customers
Rate this plugin with 10 out of 10
5/5 - Does what needs to be done
The OpenAI Commerce for Magento 2 makes it easy to generate product feeds to submit to OpenAI. Adjusting the mapping in the extension is easy, meaning that you can setup a feed in 5 minutes of configuration. Does exactly what needs to be done (just like all extensions from Magmodules). Highly recommended!
Knowledgebase
All familiar Help articles
A list of the most frequently asked questions from the OpenAI Commerce extension for . We've tried our best to help you with the installation, configuration and troubleshooting. If you do not find the answer you were looking for, you can always send us a message through this form.
Installation articles
Installation using Composer (recommended) Installation using the Adobe MarketplaceChangelog 1.4.0 is released on 17 November 2026
Our plugins are regularly updated to ensure compatibility with the latest Magento release.
-
Features:
- A new Feed Format setting writes the feed as JSON Lines, one product object per line, which is the shape OpenAI's merchant onboarding reads from a hosted URL; the single JSON document the specification shows cannot be submitted there. A second setting writes the file gzip compressed. The filename extension follows both, so the URL to register is `.json`, `.jsonl` or `.jsonl.gz`.
- Conversions reach OpenAI Ads from the server instead of the browser alone: an order queues its event, cron delivers the queue every five minutes, and the result is visible under Marketing > OpenAI Commerce > Ads Conversions with payload and API response. A browser pixel loses the conversions an ad blocker, a privacy mode or a closed tab takes away.
- Seven event types are supported, each with a real Magento trigger: order_created, checkout_started, items_added, contents_viewed, page_viewed, registration_completed and lead_created. A multiselect decides which ones are sent and it defaults to purchases only, so nothing starts flowing without being asked for.
- A visitor arriving from an ad carries an oppref parameter on the landing URL, which a storefront script stores in a first-party cookie so it survives full page cache; the queued row holds that identifier next to the pixel's own cookies, the IP address and the user agent.
- A config button posts a validate_only sample event to check the credentials, and `mm-openai:conversions:process` sends the queue from the command line exactly as the cron does. Improvements:
- Events go out in batches, and a batch the API rejects over a single invalid event is retried event by event, so one bad order cannot block the rest. Transport errors, 5xx and 429 are retried up to five times; other 4xx are terminal.
- The order increment ID is the event ID and therefore OpenAI's deduplication key, so resending an event is safe. Events older than seven days are skipped because the API rejects them anyway, and a missing Pixel ID or API key leaves the queue untouched instead of burning attempts.
- Events other than purchases store a snapshot of their own, and that snapshot holds only SHA-256 digests, so no plain email addresses or names are written to the queue table. Hashing follows the documented normalisation rules.
- The configuration sections sort alphabetically within the O block: Ads Conversions, Feed Configuration, LLM Configuration.
- A JSON Lines feed carries no `metadata` block, because a reader takes every line as a product; the seller fields it holds are written on each product as well, so nothing is lost. Bugfixes:
- The admin buttons of this module only act inside its own configuration form. The click handler and the modal lookup were bound to the whole admin page, so they also fired on the configuration page of another Magmodules module.
- The spinner on an admin button now disappears even when the request fails or the response cannot be read; a request that never arrives says so instead of leaving the modal empty.
- A configurable product takes the price a customer can actually pay for its cheapest buyable variant. The indexed minimum price can sit below every variant and the indexed final price above it, and both cases published a price nothing sells for.
- The CLI commands take their configuration through a proxy, so `setup:install` with this module in place no longer touches configuration before the store exists.
-
The error log button's template left a `` unclosed.
- Every configuration group opens expanded, Debug excepted. All of them shipped collapsed, so every setting took a click to reach before it could be read or changed.
- A telephone number is only hashed when it carries a country calling code, which a leading `+` or `00` says it does. A national number was stripped of its leading zero and hashed as if that zero were the country code, so `06 12345678` went out as `612345678` and matched nobody; it is left out now.
- The source URL of a page view, a cart addition or a started checkout reports https for a shop that terminates TLS at a proxy. It read `$_SERVER['HTTPS']`, which such a setup does not set, so these events reported http while the orders of the same shop reported https.
- Downloading, deleting and cleaning up a feed looked for a file ending in `.json` whatever it was written as, so a feed in another format was neither served nor removed, and the cleanup counted it as removed anyway.
Tests:- Architecture tests on the admin button script (form scope, loader in `finally`, changelog through the sanitizer), on CLI dependencies being proxied, on `db_schema_whitelist.json` agreeing with `db_schema.xml` in both directions, on section sort order, and on template script syntax.
- The end-to-end suite type-checks as its own gate (`npm run typecheck`), and the baseline fixture names the Ads Conversions section and its CLI command.
- Unit tests on the JSON Lines writer: one product per line, no metadata block, a trailing newline, an empty catalogue, the same sanitising as the document, and gzip that decodes back to the same feed. Plus the resolver that finds a feed file by whatever extension it was written with.
- Unit tests on the conversions logic: the normalisation and hashing of every customer identifier, the minor unit per currency group, which answers from the API are worth retrying, the batch that is retried event by event after a rejection, the queue a shop without credentials keeps untouched, and the event id as the deduplication key.
- End-to-end coverage of conversions: a real order queues exactly one event, nothing is queued with either switch off, and delivery is proven against a stand-in for the API - credentials on the request, a rate limit leaving the row pending until the next run delivers it, a rejection terminal on one attempt, one invalid event in a batch not taking the rest down with it, and test mode sending validate_only.
Compatibility:- Magento 2.4.x and up
- PHP 8.1 and up
Version 1.3.0 3 November 2026-
Features:
- A new Enable Checkout setting under Seller Data marks every product as eligible for checkout in ChatGPT; it is off by default.
- Every product in the feed carries the seller name, the policy pages and the return window, as the OpenAI feed specification reads them; the weight is also written as `weight` with its `item_weight_unit`.
- Product weights stored in grams or ounces are converted to kilograms or pounds in the feed. A new setting under Advanced says which unit the catalogue keeps; left on the default nothing changes.
- Generating a feed from the configuration page shows a progress bar. The run happens in the background while the row reports how many products have been written, the row says when it finished, and the page keeps its state. Improvements:
- The feed carries `url`, `image_url` and `additional_image_urls` as the specification names them, alongside the fields it has always written.
- A product with no GTIN attribute mapped sends none. The default was the SKU, which is never a valid GTIN, so every item carried one that fails validation; a shop that mapped a real GTIN keeps its setting.
- Generating llms.txt while it is switched off names the switch that is off instead of offering both as possible causes.
- The products in the feed are written as a list, one record per row, the way the specification reads them.
- The offer id of a product is its id and the seller, so it stays the same when the price changes.
- The result message of a feed run shows the time in the admin's timezone.
- The llms.txt texts in the admin and the documentation say where the file lives and that it is served at /llms.txt, and that the module's own Enable switch has to be on.
- The message after generating llms.txt shows the URL the file is served at instead of its path on the server.
- llms.txt lists categories with their path (Men > Tops) and links the home page to the store root.
- The llms.txt cron is on out of the box, like the feed cron; it only runs for store views that have llms.txt enabled.
- A description with line breaks, headings or a list after an unclosed paragraph keeps a line break at every boundary, instead of running the sentences together.
- The Seller Name and Return Window settings explain what they are for, and the action dropdown of the feed tables is wide enough for its placeholder.
- A product set to No for OpenAI Search Enabled is written with `enable_search` false; the setting was overwritten with Yes on every product.
- The feed's product data runs on the shared drop-in that the other feed modules use; what differs per channel is passed in at the call site.
- "Use Default" is available on all 86 website- and store-scoped settings.
- The cron schedule lives in the field the administrator edits, and the cron jobs run in their own group with their own process.
- A configurable parent carries no `item_group_id`; only its variants do, all under the parent SKU.
- The admin header logo comes from the Magmodules CDN for the extension and version.
- The dead `inventory_types` and `inventory_filter` settings are gone.
- `composer.json` states Magento 2.4 and PHP 8.1, the floor the code and the CI matrix have had for a while. Bugfixes:
- Descriptions keep their named characters: a quotation mark written as an entity reached the feed as `“`.
- A list item that is never closed, which the sample data writes, starts its own line instead of being glued onto the one before it.
- The Run Self-test button is back on the Debug section; the template rendered the result window without the button.
- Product links in the feed and the llms.txt link in the admin carry the store code when "Add Store Code to Urls" is on. Without it every link of a non-default store view opened the default store.
- A field left empty after post-processing is no longer written as an empty value on every item.
- Children of a bundle or grouped product no longer receive the parent's SKU as item_group_id; only a configurable's variants form a group.
- The feed grid mass delete removes the selected feeds; a select-all could remove every feed.
- The self-test and version-check screens require the module's own permission, so access can be granted per admin role.
- Feed generation is on out of the box; the default landed on a path the field does not use.
- A product with no price of its own keeps its own price. In a batch it could be published with the price of whichever product came before it.
- HTML entities in catalogue text are decoded before the feed is written, so a consumer reads the character rather than the entity.
- The feed cleanup reports how many feed files and records it removed and how old they had to be.
- The llms.txt schedule stays in its own field; it was also written into Magento's default cron group, where it left a job with no class behind.
- The feed file name is validated to a plain name so it cannot leave the feed directory.
- AJAX responses in the admin modals are escaped, and the changelog from the version host is reduced to its text formatting tags.
- An extra feed field named with the `g:` prefix no longer comes out as `g:g:field`.
- The `ProductAttributes` data patch returns itself, as Magento requires. Tests:
- The OpenAI product feed specification is pinned in the module, and a unit test checks every field name the feed writes against it.
- An end-to-end test generates the feed with "Add Store Code to Urls" on and checks every link; unit tests pin the store code in rewrite and fallback links, that an empty field is not written, and that only configurable parents form an item group.
- End-to-end coverage of the feed itself: the CLI writes a feed with products in it, a simple product carries the fields the channel needs, a disabled product is left out, the visibility filter keeps what is selected, the variants of a configurable carry the parent SKU, an out of stock product is reported or filtered, and nothing is written while generation is off. Plus the baseline layer: everything on, everything off, every grid, every admin route, on Magento 2.4.7 and 2.4.9.
- Architecture tests on Api contracts, admin ACL resources, cron wiring, data patches, grid wiring, namespace casing and every `config.xml` default landing on a declared admin field. PHPStan runs at level 8 without a baseline, and CodeSniffer runs the module standard at severity 6; the lint job checked out nothing and passed whatever was pushed. Compatibility:
- Magento 2.4.x and up
- PHP 8.1 and up
Version 1.2.0 25 May 2026-
Features:
- Added llms.txt generation per store view, including links to key store content such as categories, CMS pages, and product feeds.
- Added multistore support for llms.txt files with clean public URLs per store view.
- Added admin files overview with Generate and View actions per store.
- Added configurable cron schedule for automatic llms.txt regeneration.
- Added Magento 2.4.9 and PHP 8.5 support. Improvements:
- Bumped the minimum PHP version to 8.1. Bugfixes:
- Fixes Not Visible Individually parent products not being excluded when product type is set to simple products only.
- Fixes configurable product price logic not supporting minimum and maximum price calculation.
- Fixes weight unit conversion for g-to-kg and oz-to-lb not being applied before the generic weight handler.
- Fixes Header block Context type to Template\Context. Minimum requirements:
- Magento 2.3.x & PHP 8.1
Version 1.1.2 26 March 2026-
Bugfixes:
- Fix store-scoped attribute config not passing storeId to weight flag
- Fix NVI exclusion to respect visibility filter settings Improvements:
- Optimize AttributeMapper option resolution for cron performance
- Extract shared feed processing logic into ImageProcessor and ActionProcessor service classes Minimum requirements:
- Magento 2.3.x & PHP 7.4
Version 1.1.1 17 October 2025-
Bugfixes:
- Fixed undefined store_code when Seller Name is not set Minimum requirements:
- Magento 2.3.x & PHP 7.4
Version 1.1.0 6 October 2025-
Features:
- Added OpenAI Search Enabled attribute. Improvements:
- Use CMS page sources for privacy, terms of service, and policy. Bugfixes:
- Fixed issue with content-type in preview. Minimum requirements:
- Magento 2.3.x & PHP 7.4
Version 1.0.0 3 October 2025-
Features:
- OpenAI Commerce Feed Minimum requirements:
- Magento 2.3.x & PHP 7.4
Version 1.0.0 3 October 2025-
Features:
- OpenAI Commerce Feed Minimum requirements:
- Magento 2.3.x & PHP 7.4
100% Unencrypted Code quality guaranteed.

Adobe Commerce Marketplace Approved
This extension is listed on the official Adobe Commerce Marketplace and has passed their review process for code quality, performance and security.
Adobe Marketplace approval means the extension follows Magento coding standards and integrates properly with the platform. Updates go through the same review process.

Hyvä Compatible Plugin
This extension includes full support for Hyva Themes. All frontend components, widgets and templates work natively with Hyva's Alpine.js and Tailwind CSS stack.
The extension works out of the box on Hyva and is tested with every new Hyva release to ensure continued compatibility.

GraphQL Made Ready
This extension exposes its data and configuration through Magento's GraphQL API. You can query and mutate data directly from headless frontends or custom applications.
The GraphQL schema follows Magento's conventions and covers all key functionality of the extension. This makes it ready for PWA Studio, Vue Storefront or any other headless setup.

Magento Cloud Support
This extension is fully compatible with Adobe Commerce Cloud (Magento Cloud). It works with the read-only filesystem, the build and deploy pipeline and cloud-specific caching layers.
No additional configuration or workarounds are needed for cloud environments. The extension handles static content deployment and config management the same way as on-premise installations.

Private Composer Repo
After purchase you get access to a private Composer repository with personal credentials. Install and update the extension with a single composer command, just like any other dependency.
Your Composer credentials are available in your account dashboard. Updates are published to the repository as soon as a new version is released, so you always have access to the latest version.
Playwright E2E Tests (36)
This extension is covered by 36 end-to-end tests using Playwright. These tests run automatically on every code change through our CI/CD pipeline.
The tests simulate real user interactions across checkout flows, configuration changes and edge cases to make sure everything works before every release.