Why Most Furniture Catalogs Aren't Ready for AI Shopping Agents (2026)
We checked 21 furniture retailers' sofa pages: only two described every fabric option as a machine-readable variant. What AI shopping agents need from a furniture catalog, and a readiness checklist for brands, distributors and retailers.
A shopper can now ask an assistant for a washable corner sofa that fits a 9-foot wall, in something close to oatmeal bouclé, delivered before Christmas. The assistant compares the options and narrows the list without the shopper opening 10 product pages.
Your sofa can only enter that comparison if the assistant can reach the product and read the facts that decide it. For furniture, that's much more than a name and a "from" price. Width, depth, orientation, cover material, leg finish, lead time and the exact variant all change whether the answer is right.
Most furniture teams already hold those facts somewhere: a supplier spreadsheet, the PIM, the configurator's own code. The question is whether they reach the layer machines read.
How will AI shopping assistants change furniture retail?
AI shopping assistants move the comparison stage of a furniture purchase out of the retailer's website and into the conversation. Instead of browsing a sofa category, a shopper states constraints (size, fabric, budget, delivery date) and the assistant shortlists specific products from many retailers at once.
That shift is already visible in furniture:
- Checkout inside the assistant. Ashley launched a fully transactional shopping experience in Perplexity in late 2025, so customers can ask for recommendations, add items to cart and pay in one conversation (Retail Dive).
- Open protocols for agent purchases. Google launched the Universal Commerce Protocol (UCP) in January 2026, co-developed with Shopify, Etsy, Wayfair, Target and Walmart (Google). OpenAI and Stripe released the Agentic Commerce Protocol in September 2025 (OpenAI).
- Retailers using AI to police supplier data. Wayfair says its AI systems have validated over 11 million product specs; when confidence is high the system updates product content and notifies the supplier, while riskier changes need the supplier's confirmation (OpenAI customer story).
What changes for furniture businesses is less about checkout than about inclusion. An assistant can only recommend the products it can identify and describe with confidence, so catalog completeness starts to decide which pieces get shortlisted at all.
Why are many furniture catalogs not ready for AI shopping agents?
Many furniture catalogs aren't ready because their machine-readable product data describes one default configuration of a product that is actually sold in dozens. The page a person sees offers 30 fabrics and three sizes; the structured data an agent reads often describes one.
We wanted to see how common that is, so we checked.
What we found on 21 furniture retailers' sofa pages
On October 17, 2026, we fetched one configurable sofa page from each of 21 furniture retailers in the US and UK, and read the structured data in the HTML the server sends, as a client that doesn't run JavaScript would see it.
Ten of the 21 returned no product data to our client: nine answered with an access-denied or rate-limit response (HTTP 403 or 429), and one returned an empty JavaScript shell. That may reflect our location or network rather than a policy aimed at AI agents, so treat it as a caution, not a verdict. It's still a reminder to check what your own bot-management settings do.
The 11 pages that did serve data all used schema.org Product markup. The gaps were in what that markup described:
| What the structured data included | Pages (of 11) |
|---|---|
Product JSON-LD | 11 |
A ProductGroup with hasVariant | 3 |
| Every fabric or color option a shopper can pick, as a variant with a color attribute | 2 |
| Only one configuration (the default or selected one) | 6 |
| Price per variant | 4 |
| SKU | 10 |
| GTIN | 1 |
| Color that names a color | 5 |
| Material | 3 |
| Fiber composition (for example "100% polyester") | 0 |
| Width, depth and height | 4 (two with units) |
Two retailers, Burrow and Joybird, described each color or fabric option as a separate variant with a declared variesBy axis. That's the structure Google's variant guidance asks for.
The more common pattern is that the variant data exists, just not where machines look for it:
- John Lewis's sofa page lists 103 fabric options in its own page data; one appears in its structured data.
- Sofology's page data lists 67 SKUs with dimensions and lead times; its structured data describes one.
- Arhaus's page offers 16 fabrics in two depths and three widths; its structured data lists 83 offers without saying which option each one is.
- DFS's page data lists 12 colors across four covers; its structured data names no color.
We also saw smaller errors that make otherwise good data unreadable: an availability property with the wrong capitalization, dimensions with no unit, a price range whose upper bound was zero, and a set of variant images whose URLs were empty fragments. On one page, 84 of 91 variant images were fabric swatch thumbnails rather than photos of the sofa in that fabric.
Robots files weren't the problem. None of the 18 readable robots.txt files disallowed AI crawlers site-wide, and one explicitly allowed OpenAI's crawlers, yet its product page still returned a bot challenge to our client. Declared policy and what the bot-management layer actually does can differ.
How we tested
- Sample: 21 retailers, one configurable upholstered sofa page each (a sofa sold in several fabrics, colors or sizes), checked on October 17, 2026. This is a spot check, not a census: other pages on the same sites may be marked up differently.
- Method: a standard desktop browser user agent requested each page; we parsed every JSON-LD block in the raw HTML without running JavaScript, and checked for microdata and RDFa (none found). Option counts come from the same HTML: configurator labels, links to sibling colors and the page's embedded app data.
- Caveats: requests came from a UK connection, so some blocks may reflect geography, network reputation or our client's fingerprint rather than a decision about AI agents. We recorded blocked and empty responses as outcomes and didn't try to get around them. Google renders JavaScript and some agents may too; this test shows what a non-rendering client receives.
The four gaps behind the numbers
- Variant identity. A sofa sold in 30 fabrics is 30 sellable products. If they share one identifier, one price and one image, an agent can't tell the oatmeal bouclé from the charcoal velvet a shopper asked for.
- Furniture attributes. Dimensions without units, material as a single word and no fiber composition leave an agent guessing whether the piece fits the room or survives a dog.
- Imagery per variant. One hero image for every colorway means the picture an assistant shows may not match the option it recommends.
- Availability per variant. A made-to-order fabric with a 10-week lead time and an in-stock one look identical if availability is recorded once for the product. That's also why SKUs can seem to go missing after launch; see why furniture ranges take so long to go online.
Configurable furniture is the hardest case. Why AI shopping assistants struggle with configurable furniture goes deeper on sectionals, modular pieces and option dependencies.
Why is structured product data more important for AI search?
Structured product data matters more for AI search because an assistant comparing dozens of products can't ask a sales associate what fits; it reads fields. If the field for a variant's color, material or availability is empty, that fact doesn't exist for the comparison.
The platforms say this plainly:
- OpenAI says ChatGPT shopping uses "structured metadata from first-party and third-party providers," that merchants can apply for direct product feeds, and that Shopify merchants' product data is already integrated through Shopify Catalog (OpenAI Help Center). Its product feed asks for one row per purchasable item or variant, each with its own price, availability, URL and an image "showing this variant" (OpenAI product feed spec).
- Google lists furniture among the product types sold in variations, and recommends "putting Product structured data in the initial HTML for best results" (Google Search Central). Since May 2026, Merchant Center's
variant_optionattribute is used "to show your product together with all its available variants to consumers in conversational AI as well as traditional Search experiences" (Google Merchant Center). - Google's AI features have no special requirements beyond normal search: pages need to be indexable, important information should exist as text, and structured data should match the visible page (Google Search Central). An
llms.txtfile isn't a substitute for any of that.
Structured data doesn't replace good page copy. A feed carries standard fields and frequent price and stock changes; the product page explains what doesn't fit a schema, such as how a sofa bed opens or whether the frame comes apart for a narrow staircase. Agents read both, so both have to agree.
What does agentic commerce mean for furniture brands, retailers and distributors?
Agentic commerce means AI assistants find, compare and sometimes buy products on a shopper's behalf, using product data from merchants' pages and feeds. For furniture, each part of the supply chain owns a different piece of the data those assistants rely on:
| Audience | What changes | What you own |
|---|---|---|
| Manufacturers and brands | Retailers' systems increasingly check supplier data automatically, and assistants compare your product against everyone else's on the facts provided | The source record: variant structure, dimensions with units, materials and fiber composition, GTINs where assigned, imagery per variant, lead times by option |
| Distributors | Several suppliers' data has to become one consistent catalog before any retailer or assistant can use it | Stable identifiers, supplier-to-catalog mappings, de-duplication and a delivery format for each retailer |
| Retailers | Assistants read your storefront, structured data and feeds, and may complete checkout through protocols such as UCP or ACP | Product page structured data, Merchant Center and other feeds, bot-management settings and the decision on in-assistant checkout |
The manufacturer's record is where most of the gaps start, and it's the hardest to fix downstream. A retailer can't publish a fiber composition or a per-fabric lead time it never received. Why supplier product data slows furniture ecommerce launches covers that handoff.
How do furniture retailers make their catalog ready for AI shopping agents?
Furniture retailers make their catalog ready for AI shopping agents by giving every sellable variant a complete, consistent and reachable record: its own identifier, attributes, image and availability, published identically on the page, in structured data and in feeds. Work through this checklist on one configurable range first.
| Area | What agents need | How to check |
|---|---|---|
| Reach | Product pages that return content to crawlers and agents, with Product structured data in the initial HTML | Fetch a product page without JavaScript; review bot-management rules, not just robots.txt |
| Variant identity | One identifier per sellable variant; a ProductGroup with hasVariant and variesBy (color, size, material, pattern); a group ID shared by all variants; a URL that preselects each variant | Every fabric and size a shopper can pick appears as its own variant with its own SKU and offer |
| Identifiers | GTIN where one is assigned; MPN with brand where it exists | Never invent a value to fill a missing GTIN (OpenAI's spec says the same) |
| Furniture attributes | Width, depth and height with units; seat height and depth; materials (Merchant Center allows one primary and up to two secondary); fiber composition; color; style | Units present on every dimension, and the same unit across all three |
| Imagery | A main image showing that variant; additional views; alt text | Images match the variant, meet each channel's published specification (Google requires 500 × 500 pixels minimum from January 31, 2027), and AI-generated images carry the metadata Merchant Center requires |
| Availability and lead time | Availability per variant (in stock, out of stock, preorder, backorder); lead time for made-to-order options | Configuration availability and stock availability recorded separately |
| Consistency | The page, structured data, feeds, marketplace listings and images agree | Compare a few high-value configurable products across every surface |
| Measurement | A defined set of real shopping questions reviewed over time | Note which products appear, whether the named variant is correct, which pages are cited and whether price and availability are current |
Three practical notes:
- Channel requirements differ. OpenAI's feed, Google Merchant Center and each marketplace set their own fields and image rules. Check each destination's published specification rather than assuming one export fits all.
- Check your CDN defaults. Cloudflare says that from September 15, 2026, new domains default to blocking bots classified as Training or Agent on pages that display ads, while Search stays allowed (Cloudflare). Confirm the live setting for your own account.
- Discovery and checkout are separate decisions. Being understood by an assistant doesn't require supporting an agent-completed purchase. Delivery access, made-to-order options, swatch decisions and return costs can make a referral to your own site the better experience for some ranges. A checkout protocol can't compensate for a catalog with no dependable dimensions or variant identity, so fix discovery first.
What to fix first
- Confirm that product pages can be reached and indexed.
- Give each sellable variant a stable identity and a URL that loads it.
- Complete dimensions with units, materials, lead times, delivery and returns.
- Make visible page text, structured data and feeds agree.
- Add accurate imagery for every variant shoppers can choose.
- Measure a defined set of shopping questions and correct the gaps.
The underlying record is the subject of the furniture product data model, and the furniture ecommerce readiness checklist covers the broader go-live review.
Where Furniture Connect fits
Furniture Connect's Agentic Catalog prepares the product record that storefronts, feeds and assistants read. It doesn't publish your storefront's structured data or submit feeds to AI platforms; it makes sure the data those depend on is complete and consistent.
- Variants as real records. Products hold variants on up to three axes (fabric, size and leg finish, for example), with materials and finishes drawn from shared libraries and dimensions in structured fields.
- Imagery per variant. Images attach to the product or the variant they show, and Studio can generate a variant's imagery from an approved photo and the material record.
- A per-channel gap check. Channel completeness scoring shows what's still missing for each channel before anything is published.
- Routes out. The two-way Shopify app syncs variant SKUs and barcodes, media and furniture metafields into Shopify (products arrive as drafts, and publishing stays a separate step). CSV and XLSX exports carry the same record elsewhere.
- Assistants that work on the catalog. A remote MCP server lets assistants such as ChatGPT and Claude read and update the catalog, within the permissions a workspace admin sets.
People stay in the loop. AI-filled attributes and generated imagery need a person's review before they publish, and linking an image to an exact SKU or colorway needs someone to confirm it.
This work pays off fastest for catalogs of roughly 200 SKUs or more, or ranges sold through more than two channels, where gaps multiply faster than a team can check them by hand.
Frequently asked questions
What is agentic commerce?
Agentic commerce is shopping in which an AI assistant does part of the work for the shopper: finding products that meet stated needs, comparing them and, increasingly, completing the purchase. Open standards such as the Agentic Commerce Protocol (OpenAI and Stripe) and Google's Universal Commerce Protocol define how assistants, merchants and payment providers exchange product, cart and checkout data. For furniture, it means the comparison often happens before a shopper ever visits your site.
Why are furniture catalogs not ready?
Because the facts that decide a furniture purchase rarely reach the machine-readable layer. In our October 2026 check of 21 furniture retailers' sofa pages, 11 served structured product data to a non-JavaScript client, and only two of those described every fabric option as a variant. None included fiber composition. The variant data often exists in the page's own code or in the PIM, just not in the fields shopping agents and search engines read.
What product data do AI agents need?
A stable identifier for every sellable variant, with its own price, availability, URL and image; a product group that links the variants and says how they differ (color, size, material); dimensions with units; materials and fiber composition; and lead time. OpenAI's product feed asks for one row per purchasable item or variant, and Google recommends putting Product structured data in the page's initial HTML.
How do I make my catalog agent-ready?
Start with one configurable range. Give each sellable variant its own identifier and image, complete dimensions with units, materials and availability, then check that the product page, structured data, feeds and imagery agree. Confirm crawlers and agents can reach the page, decide separately whether to support checkout inside an assistant, and have a person review AI-filled data and generated imagery before it publishes.
If AI shopping is on your roadmap, start with one difficult range. Book an Agentic Catalog demo and we'll work through the variant data and imagery it needs before any assistant can represent it properly.
Sources
- Furniture Connect structured-data spot check of 21 furniture retailers' sofa pages, October 17, 2026 (method and caveats in "How we tested" above).
- OpenAI Help Center: Shopping with ChatGPT Search, where ChatGPT's product data comes from, direct feeds and Shopify Catalog.
- OpenAI: Product feed specification, one row per purchasable item or variant, variant grouping, dimensions and identifiers.
- OpenAI: Buy it in ChatGPT, Instant Checkout and the Agentic Commerce Protocol, September 29, 2025.
- OpenAI customer story: Wayfair, AI validation of product specs and the supplier confirmation loop.
- Google: New tech and tools for retailers to succeed in an agentic shopping era, the Universal Commerce Protocol launch, January 11, 2026.
- Google Search Central: Product variant structured data,
ProductGroup,hasVariant,variesByand variant URLs. - Google Search Central: AI features and your website, requirements for AI Overviews and AI Mode.
- Google Merchant Center: Product data specification, item group, color, material, size, dimensions and image requirements.
- Google Merchant Center: Variant option, the variant attribute for conversational AI.
- schema.org: ProductGroup, the vocabulary for a product sold in variants.
- Retail Dive: Ashley launches purchases through Perplexity, December 4, 2025.
- Cloudflare: Block AI bots, defaults for new domains from September 15, 2026.
