AI Commerce Ready: preparing Magento and Shopify stores for AI-driven shopping
HUB LLC · 12 September 2026 · Commerce
Consider a request a customer might give an AI assistant: a quiet dishwasher under €600, no wider than 45 cm, available in Riga and deliverable this week. To answer it with confidence, software needs a product type, a price, a width, a noise level, stock by location and a delivery estimate.
Most stores hold that information somewhere. It is simply spread across the storefront, an ERP, a PIM, supplier spreadsheets and PDF data sheets. AI Commerce Ready is how HUB LLC structures the work of turning that fragmented environment into a commerce foundation that people, search engines and AI-driven shopping systems can all use. This article explains the problem, the five layers of the work and what we deliberately do not promise.
Why a good product page is no longer enough
The familiar journey of keyword, search result, category page, product page and checkout still matters, and conventional search should not be neglected. What is changing is that more of the comparison can happen before anyone visits a store. A conversational system narrows options by specific requirements, and it can do that only when those requirements exist as reliable data.
An attractive page that states the noise level in an image, or reveals the delivery estimate only after an address is entered at checkout, gives such a system nothing to work with.
What the platforms are building
As of September 2026, several parallel efforts show where the requirements are heading:
- Google introduced the Universal Commerce Protocol in January 2026 as an open standard for agents, businesses and payment providers, co-developed with Shopify, Etsy, Wayfair, Target and Walmart. The same announcement described new Merchant Center data attributes for conversational discovery. Source: Google.
- Google Merchant Center documents optional conversational attributes, including questions and answers, related products, document links and variant options, to help AI-driven surfaces such as AI Mode understand products. Source: Google Merchant Center Help.
- Shopify serves a managed agent discovery file on storefronts that lists the store’s Universal Commerce Protocol and Model Context Protocol endpoints, its policies and guidance for shopping assistants. Source: Shopify developer documentation.
- OpenAI documents the Agentic Commerce Protocol and a product feed through which merchants share titles, descriptions, images, prices and availability with ChatGPT. Onboarding product feeds is currently limited to approved partners. Source: OpenAI Agentic Commerce documentation.
These are different interfaces controlled by different companies, and all of them will keep changing. They depend on the same inputs: accurate identifiers, complete attributes, current price and stock, clear policies and systems that can export that information reliably. Building those inputs well is a safer investment than building for any single channel.
The five layers of AI Commerce Ready
We organize the work in five connected layers. Each one produces something a client can verify.
Discover: make commercial information reachable
Robots and crawler directives, XML sitemaps, canonical tags, hreflang and country structure, redirects, URL architecture, internal linking, pagination, JavaScript rendering risks, metadata and breadcrumbs. The aim is not to create special pages written for AI. It is to make the real store technically clear and consistent for every discovery channel.
Structure: turn the catalog into reliable commerce data
Product name, brand, SKU, GTIN or MPN where applicable, category, variants, size, color, material, compatibility, technical parameters, price, currency, availability, condition, images and shipping information, modeled as data and exposed consistently in Product, Offer, BreadcrumbList and Organization markup that matches what customers see.
For large catalogs the goal is systematic correctness. Fixing ten product pages by hand does nothing for a store with 50,000 products. The deliverables are rules, templates, validation and pipelines that work at catalog scale.
Enrich: improve catalog quality without losing control
Incomplete supplier data, inconsistent titles and duplicated manufacturer descriptions are well suited to carefully controlled AI automation. A typical pipeline:
- Ingest supplier feeds, spreadsheets, PIM records or existing catalog content.
- Validate structure and required fields, and quarantine malformed rows.
- Classify products and map them to the store’s attribute sets.
- Extract attributes from titles, descriptions and data sheets, recording the source of each value.
- Normalize units, terminology and naming.
- Generate descriptions, metadata and translations from verified attributes.
- Apply business rules and confidence thresholds.
- Send uncertain or high-impact results for human review.
- Publish to Magento, Shopify or the PIM, and monitor for recurring errors.
The difference between useful and risky automation is governance. AI output should not be published blindly. A value taken from a manufacturer’s data sheet and a value inferred by a model are not equally trustworthy, and the workflow should treat them differently.
Connect: prepare the platform for agents and new channels
Product feed architecture, catalog APIs, inventory and price synchronization, merchant policy data, webhook and event design, and an assessment of readiness for relevant open commerce protocols or platform agent capabilities.
We do not recommend that every store build an autonomous purchasing agent today. The practical goal is a clean integration layer, so the business can adopt a useful new channel without rebuilding its catalog and back office each time one appears.
Measure: track what actually changes
Start with a baseline, then follow catalog completeness, structured data errors, crawl health, performance, feed errors, referral traffic, including AI referrals where analytics can identify them, and conversion. A recurring report shows whether high-priority problems stayed fixed and whether new imports reintroduce old errors. Visibility in AI answers can be sampled separately; see how to monitor AI visibility.
A worked example: the dishwasher request
Return to the opening request. The table maps each requirement to the data a system would need. The status column illustrates a typical gap analysis; it does not describe a specific store.
| Requirement | Data needed | Example status |
|---|---|---|
| Dishwasher | Product type and category | Present |
| Under €600 | Price and currency in page, markup and feed | Present |
| No wider than 45 cm | Width as a numeric attribute with a unit | Present |
| Quiet | Noise level in dB as an attribute | Missing: only in the PDF data sheet |
| Available in Riga | Stock by location | Present |
| Delivered this week | Delivery estimate by region | Missing: shown only at checkout |
An AI assistant cannot confidently recommend a product on the strength of information the catalog does not provide. The two missing rows are the work: extracting noise levels from data sheets into a structured attribute, and exposing delivery estimates in a form that pages, feeds and integrations can share.
Platform notes
Magento and Adobe Commerce
Product information is often distributed across attribute sets, configurable products, custom modules, extensions, store views and ERP or PIM integrations. The work typically involves attribute architecture, variant relationships, structured data templates, indexing and caching behavior, multi-store language structure, feeds, cron processes and custom APIs. Nordic agencies can have this delivered white-label, with the agency keeping the client relationship.
Shopify
The platform provides strong infrastructure, including the managed agent discovery file described above, but it does not fix the catalog. Merchants still control product and variant data quality, collection structure, theme output, app-related performance, international content and external integrations.
Custom commerce and marketplaces
On custom platforms the data model, API design, search architecture and inventory sources are part of the review. Marketplaces add another challenge: many sellers supplying inconsistent titles, categories and images. Validation and AI-assisted normalization before listings go live raise the quality of the whole catalog rather than one seller at a time.
How the work is scoped
- Audit. Public technical analysis, a catalog sample, structured data and discoverability review, prioritized findings and an implementation roadmap. See what a readiness audit should check.
- Implementation. The agreed improvements across templates, catalog data, feeds and integrations, with testing and before-and-after validation.
- Continuous improvement. For stores with changing catalogs and several integrations: monitoring, catalog quality controls, recurring technical reviews, enrichment workflows and ongoing engineering capacity.
Scope depends on platform complexity, catalog size, the number of languages, integrations and the quality of source data, so we estimate after the audit rather than publishing a fixed price.
What we will not promise
We cannot guarantee that a third-party AI system will cite or recommend a particular merchant. We can improve the technical accessibility, data quality, structured information and integrations that make a business easier for digital systems and customers to understand.
Independent AI systems control their own outputs and change their behavior over time. Treat any offer to make an assistant recommend your products, or to rank first in AI, with suspicion.
Value that does not depend on one AI provider
This work pays off even if AI shopping develops differently from current expectations. Better product data improves filters, onsite search, feeds, marketplace listings and customer support. Accurate structured data supports conventional search. Better attributes improve comparison and conversion. Cleaner APIs lower integration costs, catalog automation reduces manual work and consistent multilingual data makes international expansion easier.
HUB combines eCommerce engineering with AI-assisted development and automation, and experienced engineers remain responsible for architecture, security, business logic, testing and deployment. If you would like to know which of these layers matters most for your store, tell us about your platform and catalog, or see the AI Commerce Ready service. Stores that also want a shopping or customer service assistant can read about our AI assistants.