If you are building an AI shopping agent, clothes are where your users will test it hardest. A shopper asks for "a cotton kurta set in M that runs true to size, delivered to Pune by Friday, exchangeable if it doesn't fit", and every part of that request depends on data most product sources don't carry.
This guide is for teams building shopping agents and the platforms behind them. It covers where fashion product data comes from today, which fields a fashion agent needs and which ones standard feeds leave out, and what to check in any data source before you rely on it. The last part describes TailorTalk Shop, the fashion catalogue we are opening to shopping agents. It is in early access, so we say plainly what it covers and what isn't published yet.
The stakes are high: McKinsey puts US retail revenue orchestrated by agents at up to $1 trillion by 2030.
Why fashion is the hard category for shopping agents
An agent can compare a hundred phone chargers on wattage and price without asking anyone anything. Clothing breaks that in five ways:
- Size is not a fact, it is a mapping. An M at one label is an L at another, size systems differ by country (US, UK, EU, Indian), and "runs slim" is rarely written down anywhere a machine can read.
- Fabric decides the sale. Composition is sometimes in the feed; sheerness, lining, stretch and whether a colour bleeds almost never are.
- A lot of fashion is made or finished to order. Ethnic wear sells stitched, semi-stitched or unstitched, with custom measurements and a lead time that changes the delivery date.
- Stock lives at the variant level and moves daily. "Is the rust one still there in L?" has a different answer each afternoon.
- Delivery and exchange rules close the order. Outfits are bought for a date, and shoppers ask about cash on delivery, exchanges and sale-item returns before they pay.
Getting this wrong is expensive. Coresight Research estimates the average US online apparel return rate at 23.4% in 2025, and nearly 70% of shoppers who returned clothing bought online cited size and fit (research sponsored by Alvanon). An agent that guesses at fit adds to that number.
Where shopping agents get product data today
There are four main sources, and each was built for a different job:
| Source | What you get | Gaps for fashion |
| OpenAI product feeds | Brand-pushed snapshots over SFTP, at least daily; apparel fields such as size, size_system, material, color, gender and age_group; return and exchange flags | Approved partners only; no fields for fit notes, garment measurements or made-to-order options |
| Shopify Catalog | Global Catalog across Shopify merchants and Storefront Catalog for one store, with search_catalog, lookup_catalog and get_product tools | Only brands in Shopify Catalog (Shopify stores, and others on Shopify's Agentic plan); no caching; some fields are inferred by Shopify's AI |
| Google Merchant Center | Feeds behind Google Shopping and AI Mode, plus optional conversational attributes such as question_and_answer and variant_option | Built for Google's own surfaces; the conversational attributes are optional, so coverage varies by merchant |
| Product pages (schema.org) | Product markup with size (including SizeSpecification), shipping details and return policy | Only as complete as each site's markup; crawl lag on stock and price |
Some details worth knowing. OpenAI's feed spec has a material field and a size field with an optional size system, and it expects full snapshots at least daily, preferably as parquet. Shopify's catalog docs say some fields are inferred by Shopify's AI and "might not always be present or have varying accuracy", and they ask agents not to cache search results. Google describes its conversational attributes as "completely optional", so coverage depends on each merchant filling them in.
Open protocols are settling how agents talk to merchants. The Agentic Commerce Protocol (from Stripe and OpenAI) and the Universal Commerce Protocol (from Google with Shopify, Etsy, Wayfair, Target and Walmart) standardise discovery and checkout flows. They define how an agent asks; they don't make a brand's fit notes or stitching options exist. The data still has to come from the brand.
Then there is the coverage gap. Many independent fashion labels, and much of ethnic wear, sell mainly through Instagram and WhatsApp, with no Shopify store and no feed. To an agent that only reads feeds, those brands don't exist.
The fields a fashion shopping agent needs
Use this as a checklist for whatever source you build on. The right-hand column is a rough guide to what generic feeds carry today:
| Field | Why the agent needs it | In generic feeds? |
| Size label and size system | Map the shopper's size across labels and countries | Usually |
| Size chart and garment measurements | Answer "will it fit" with numbers, not guesses | Rarely |
| Fit notes (runs small, relaxed, true to size) | The question behind most size returns | Rarely |
| Fabric composition | Filter by cotton, linen, silk | Often |
| Fabric detail: weight, sheerness, lining, stretch | Answer what shoppers ask before buying | Rarely |
| Work and occasion (embroidery, festive, office) | Match intent such as "a wedding-guest outfit" | Rarely |
| Stitched, semi-stitched or unstitched; custom sizing; lead time | Quote a real delivery date for made-to-order pieces | Rarely |
| Stock by variant (size and colour) | Avoid recommending what can't ship | Usually, but often stale |
| Delivery time by destination | Answer "will it arrive by Friday" | Sometimes |
| Cash on delivery | Shoppers ask before paying wherever COD is common | Rarely |
| Exchange and return rules, including sale items | Close the order and avoid disputes | Sometimes |
How to judge a fashion data source
Four questions sort the sources worth integrating from the ones that will cost you returns and support tickets:
Clean data
Is each field supplied by the brand, or inferred by a model from images and titles? Inferred fields are useful for search but risky for fit and fabric answers. Ask which fields are brand-supplied, and whether values are normalised (one size system per product, one way of writing "unstitched").
Coverage
Which brands are in it, and how do they get there? A source limited to one commerce platform misses brands that sell on Instagram and WhatsApp. Ask how many brands are fashion brands, in which markets, and how products without a website get catalogued.
Reliability
How fresh is stock, and what happens when the data has no answer? A daily snapshot is fine for discovery but not for "is it in stock in L right now". Ask for the sync interval, and whether there is a way to get an answer from the brand when no field covers the question.
Ease of integration
Can your agent query it the way it already works, with clear rate limits and caching rules? Shopify's rules, for example, forbid caching search results, which changes how you design retrieval.
A simple flow for a fashion shopping agent
Whatever data you use, the steps that keep fashion answers right look like this:
- Capture the constraints first: size and the system it's in, budget, occasion, destination and the date it is needed by.
- Search the catalogue for intent (occasion, fabric, work), then filter by stock at the variant level.
- Check fit with measurements and fit notes, not the size label alone.
- Check the delivery date for the destination, adding lead time for anything made to order.
- Check the brand's policies the shopper cares about: cash on delivery, exchanges, sale items.
- If a question has no field, ask the brand instead of guessing, and say so to the shopper if no answer comes back.
- Hand off to the brand's checkout. ChatGPT now sends shoppers to the merchant's own checkout, and agentic checkout elsewhere is limited to eligible merchants.
Step 6 is where most stacks have nothing to call. That is the gap a seller agent fills, and the argument for it is in our piece on buyer agents and seller agents.
TailorTalk Shop: fashion brands, answerable by your agent (early access)
TailorTalk runs AI sales agents for fashion brands on Instagram, WhatsApp and website chat. Each brand's agent already answers shoppers from that brand's catalogue, live stock and policies; Samyakk, an ethnic wear brand, handles more than 1,000 Instagram inquiries a day this way. TailorTalk Shop opens the same catalogue and the same agent to shopping agents. What it covers:
- Catalogue data with the fashion detail generic feeds miss: fabric, work, occasion, stitched or unstitched, and the size chart for each piece.
- Prices, sizes in stock, delivery times, cash on delivery and exchange rules, kept in sync with the brand's Shopify, WooCommerce or Magento store, a Google Sheet or TailorTalk's AI Catalog.
- Brands without a website. Boutiques that sell only on Instagram and WhatsApp build their catalogue from photos with AI Catalog, and it becomes readable in the same way.
- Answers beyond the data. When a question has no field, the brand's own TailorTalk agent answers it, from its catalogue and policies only. It speaks only for that brand and never recommends another; if a piece is out of stock, it suggests the closest match from the brand's own range.
- Consent and privacy. Brands opt in and can switch it off at any time. Only what a product page would show is shared: products, photos, prices, sizes, stock and policies. Customer conversations, phone numbers, order history and sales numbers never are.
What isn't published yet: endpoints, authentication, rate limits and the full schema. TailorTalk is opening this in stages, starting with brands already selling through a TailorTalk agent, and we will publish the technical documentation as access opens. If you want to build on it now, contact us with what your agent does, the markets you serve, the interface your agent already uses and the volume you expect, and we'll bring you in as it opens.
If you run a fashion brand rather than an agent, the brand side is on the AI shopping assistants page, and our guide to getting your brand recommended by ChatGPT covers what you can do today.
Is there an API for fashion and clothing product data?
There are several sources, each limited in a different way. OpenAI accepts product feeds from approved partners, Shopify's Catalog lets agents search Shopify merchants' products, Google Merchant Center feeds power Google's shopping surfaces, and product pages carry schema.org markup. Few of them carry the fields fashion needs most, such as fit notes, garment measurements and stitching options. TailorTalk Shop, in early access, adds those for fashion brands on TailorTalk.
How do I build an AI shopping assistant for clothes?
Capture the shopper's size, budget, occasion, destination and date first; search the catalogue by intent and filter by stock at the size level; check fit with measurements and fit notes; confirm the delivery date and the brand's cash-on-delivery and exchange rules; and ask the brand when no field answers the question. Then hand off to the brand's checkout.
What is the difference between ACP and UCP?
The Agentic Commerce Protocol was developed by Stripe and OpenAI, and OpenAI was the first to implement it in ChatGPT. The Universal Commerce Protocol was launched by Google in January 2026 with Shopify, Etsy, Wayfair, Target and Walmart and covers the shopping journey from discovery to checkout. Both standardise how agents and merchants talk; neither supplies a brand's product data.
Is TailorTalk Shop live?
It is in early access. TailorTalk is opening it in stages, starting with brands already selling through a TailorTalk agent. Endpoints and documentation will be published as access opens; agent builders can contact TailorTalk to join early.
What data does TailorTalk Shop share with shopping agents?
Only what a shopper could see on a product page: products, photos, prices, sizes, stock, and delivery, cash-on-delivery and exchange policies. Customer conversations, contact details, order history and sales numbers are never shared, and each brand opts in.
Is TailorTalk Shop the same as TailorTalk's MCP server?
No. TailorTalk's MCP server lets a brand's own team work in their TailorTalk account from Claude or ChatGPT, signed in as themselves. TailorTalk Shop is for shopping agents that want to read fashion brands' catalogues and ask their agents questions.
Sources
- OpenAI, Products feed specification and file upload overview (developers.openai.com, read 2026-10-23).
- Shopify, Catalog for agents (shopify.dev, read 2026-10-23).
- Google Merchant Center Help, Conversational attributes (read 2026-10-23).
- schema.org, SizeSpecification (V30.1).
- Agentic Commerce Protocol (agenticcommerce.dev) and Universal Commerce Protocol (ucp.dev).
- McKinsey, The agentic commerce opportunity (October 2025).
- Coresight Research, "Shifting the Size and Fit Paradigm" (May 2026, sponsored by Alvanon), as reported by FashionUnited.