GEO for fashion: the short version
In 60 words: Fashion is the category where fit decides the sale, so AI search resolves fit before brand. To get recommended by ChatGPT, Perplexity and Google AI, a Shopify apparel store needs a machine-readable size guide, fabric composition and care, per-variant size and color availability, a clear return policy in structured data, and substantiated sustainability claims. This guide covers each lever with sources.
On 24 March 2026, Gap Inc. announced AI-powered fit guidance built on Bold Metrics’ Agent Sizing Protocol, plus support for Google’s Universal Commerce Protocol so shoppers can buy from Gap inside Google’s AI Mode and the Gemini app (Gap Inc., March 2026). CNBC called it an AI first from a major fashion company (CNBC, March 2026). The detail that matters for the whole category: Gap’s headline AI feature is not a chatbot, it is a way for agents to answer “will this fit me?” from structured body and garment data. When the largest apparel retailers reorganise around machine-readable fit, that is the signal for everyone else.
This is Generative Engine Optimization (GEO) applied to fashion and apparel. New to the term? Start with what GEO is, then how AEO, GEO and SEO differ and the 9 factors of a GEO readiness score. From here on, this guide is only about what changes for clothing on Shopify.
Why fashion is a distinct GEO problem
Three forces make apparel different from every other vertical, and one keeps the case honest.
AI-driven discovery is climbing fast. The State of Fashion 2026 report from Business of Fashion and McKinsey found that shopping-related searches on generative AI platforms grew roughly 4,700% between 2024 and 2025, and that executives now name AI as the single biggest opportunity for the industry; McKinsey’s broader work estimates AI agents could mediate $3 to $5 trillion of consumer goods commerce by 2030 (BoF / McKinsey, 2026). The same report is candid that traditional search still drives around 80% of search traffic today, so the case for GEO in fashion is the trajectory and the low competition, not today’s volume.
Fit is the structural difference. In beauty, AI routes through ingredients; in apparel, it routes through fit and availability before brand. A shopper does not ask for “a black dress,” they ask for “a black midi dress that runs true to size for a UK 12,” and the agent has to resolve size, measurements and stock before it can recommend anything. That is why Gap’s first big AI move was a sizing protocol, not a style quiz.
And returns hang over every recommendation. Apparel is the most-returned category online: clothing and footwear top the return-rate tables, with apparel commonly running 20 to 40% returns versus 5 to 9% in physical retail, driven mostly by size and fit, and made worse by “bracketing” (ordering several sizes to send most back) (Richpanel, 2026; Statista, 2026). Because returns are this expensive, both shoppers and the agents acting for them weigh return terms heavily, which makes a clear, structured return policy a ranking factor, not a footnote.
The honest counterweight: AI shopping is still early in absolute terms. A study of 1.5 million ChatGPT conversations found only 2.1% involve purchasable products (NBER, September 2025). The window is open precisely because few apparel brands have structured this data yet, fewer than 12% of marketing teams have a documented GEO strategy.
How AI actually recommends a fashion product
The retrieval pattern in apparel is consistent: the model resolves a fit-and-availability question first, then layers brand, price, returns and trust on top. Practically, that means the brand whose page connects a body query to a confident answer wins the citation. Here is the mapping AI assistants most often draw:
| Shopper question shape | What your page must expose to match it |
|---|---|
| ”Does it run true to size / small / large?” | Fit type (slim, regular, relaxed, oversized) and a model-wears note, in text |
| ”What are the actual measurements?” | Garment dimensions (chest, waist, hip, length, inseam) as text, ideally per size |
| ”Is it available in my size and colour?” | Per-variant size and color with stock status, marked up as ProductGroup variants |
| ”What is it made of, and how do I wash it?” | Fibre composition (% per fibre) and care instructions, as structured properties |
| ”Can I return it if it does not fit?” | Return window, fees and method, in prose and MerchantReturnPolicy structured data |
| ”Is it actually sustainable?” | Named certification (GOTS, OEKO-TEX) and quantified material claims, with proof |
Build for this from the start: the engines rarely name the same garment, so each one is its own surface. A dress that ChatGPT puts forward for a “true to size” query can be missing from Perplexity entirely, which is why you run the same prompt across at least ChatGPT, Perplexity, Gemini and Claude rather than tuning for one. And as in every vertical, a large share of what AI cites in fashion comes from third parties (editorial roundups, marketplace reviews, community threads), so off-site signals matter alongside your product page.
The 7 on-page levers for fashion
Seven moves on your own Shopify product pages, ranked from the one that wins the most citations to the one that polishes the edges. Every one is a content or structured-data edit (sizing, variants, returns, claims), so nothing here means rebuilding your theme.
1. Put your size guide in text and structured data, not in an image
This is the single highest-leverage apparel fix, and the most commonly broken one. A size chart saved as a JPEG or PNG is invisible to the AI crawlers that do not run OCR, so the most decisive piece of fashion data on your page can be unreadable to the model. Expose sizing three ways: as a real HTML table of garment measurements, as a stated fit type in the description (“Relaxed fit, model is 1.78m wearing size M”), and as structured size data on the variants.
There is rich schema for this. Google supports WearableSizeGroup values and SizeSpecification with sizeSystem, sizeGroup, suggestedGender and suggestedAge, and you can attach measurements through additionalProperty on the Product:
"additionalProperty": [
{ "@type": "PropertyValue", "name": "Fit", "value": "Regular" },
{ "@type": "PropertyValue", "name": "Size System", "value": "EU" },
{ "@type": "PropertyValue", "name": "Chest (size M)", "value": "100 cm" },
{ "@type": "PropertyValue", "name": "Length (size M)", "value": "70 cm" },
{ "@type": "PropertyValue", "name": "Model", "value": "1.78m, wears size M" }
]This is exactly what Verity Score checks for the fashion vertical: whether your sizing is present as text and structured data, not trapped in an image an AI cannot read.
2. Mark up size and color variants so availability is machine-readable
Apparel lives or dies on the size-color grid, and an agent will not recommend a product it cannot confirm is in stock in the shopper’s size. Use Shopify’s category metafields to link size, color and material to structured data, then mark up the product with ProductGroup schema, variesBy set to https://schema.org/size and https://schema.org/color, and one hasVariant per SKU with its own offers and availability (Google Search Central, 2026). Use full schema.org URIs for color and size values, plain text strings will not trigger Google Shopping swatches.
Shopify raised the variant limit to 2,048 in October 2025, so a full grid (for example 8 band sizes by 12 cup sizes by 10 colors) can now live on a single page rather than being split across listings, which is the consolidated structure AI agents read most reliably. The failure mode to avoid is variant incoherence: a color shown in the gallery but missing from the variant data, or a size listed as available in the picker but out of stock in the feed, which makes the model distrust the whole record.
3. Expose fabric composition and care as structured facts
“What is it made of?” and “how do I wash it?” are among the most common apparel questions, and the answer should be data, not a paragraph. State fibre composition as percentages (“80% organic cotton, 20% recycled polyester”) and care instructions explicitly, in the description and in additionalProperty (Material, Fabric Composition, Care). Composition is also load-bearing for the next lever, because a recycled-content claim is only credible when the percentage is stated.
4. Make the return policy clear, and machine-readable
Because apparel returns run so high, the return policy is a recommendation factor, not fine print. State the window, the fees and the method in plain prose near the buy box, then expose it as structured data. Google’s MerchantReturnPolicy is normally nested on the Organization via hasMerchantReturnPolicy, and you can override it per product on the Offer; the key properties are applicableCountry, returnPolicyCategory, merchantReturnDays, returnFees and returnMethod (Google Search Central, 2026). Two rules: the structured data must repeat the exact terms from your human-readable returns page or you risk warnings, and a free-and-easy returns story is worth stating explicitly because it is precisely the kind of merchant-comparison signal an AI shopping surface reads. See our shipping and returns guide for the full field list.
5. Substantiate sustainability claims, because greenwashing is now a legal and AI risk
This is the lever where compliance and AI visibility converge, and the regulatory ground shifted in 2026.
Replace vague green adjectives with named, verifiable claims: not “eco-friendly” or “conscious,” but “GOTS-certified organic cotton” or “made with 60% recycled polyester (GRS certified).” GOTS (Global Organic Textile Standard) requires a minimum 70% certified organic fibres plus social and chemical criteria across the supply chain, while OEKO-TEX Standard 100 certifies that the finished textile is tested free of harmful substances, two different claims that should not be used interchangeably. AI assistants favour claims that are specific, qualified and corroborated, and tend to refuse to repeat vague environmental superlatives.
The regulators now require the same discipline:
- In the EU, the Empowering Consumers for the Green Transition Directive (EU 2024/825) must be transposed into national law by 27 March 2026 and applies from 27 September 2026. It bans generic environmental claims such as “eco-friendly” without proof, and sustainability labels that are not based on a recognised certification scheme; penalties can reach up to 4% of turnover in the Member State concerned (EUR-Lex). Note that the separate, broader Green Claims Directive is not in force: the Commission announced its intention to withdraw the proposal in June 2025 and negotiations were suspended, so as of 2026 the binding framework for apparel is the Empowering Consumers Directive, not the Green Claims Directive (Hogan Lovells, June 2025).
- In the US, the FTC Green Guides require recycled-content and recyclability claims to be qualified and substantiated (for example “made from 30% recycled material”), and warn against broad, unqualified claims like “green” or “eco-friendly” (FTC). A revision has been pending since 2022 and is expected in 2026; the guides were last updated in 2012.
The same care with wording keeps a sustainability claim both compliant and AI-recommendable:
| Defensible (with proof held) | Risky / non-compliant |
|---|---|
| ”GOTS-certified organic cotton” (certificate linked) | “natural”, “eco-friendly”, “conscious” with no proof |
| ”made with 60% recycled polyester, GRS certified" | "made with recycled materials” (no percentage, no scheme) |
| “OEKO-TEX Standard 100 certified, tested for harmful substances" | "non-toxic”, “chemical-free”, “safe” (unqualified) |
| “designed to last, 2-year repair guarantee" | "sustainable”, “planet-friendly” as a collection name |
Verity flags sustainability and material claims that have no backing data an AI could verify, which is the same gap a regulator would catch. See our claims and proof guide for the verification loop.
6. Use an answer-first title and description formula
Lead with the answer, then layer detail. A workable apparel title formula: brand + product type + key material + fit + primary use + color + size range. For descriptions, open with an identity block (what it is, who it is for, how it fits, in 50 to 75 words), then specs (composition, measurements, care), then styling and “who should size up or down,” then the return note. A large share of AI citations come from the first portion of the page, so fit and material cannot be at the bottom.
Weak: “The Luxe Midi Dress. Effortlessly elegant, this stunning piece elevates any wardrobe. One size flatters all.”
Strong: “Organic Cotton Midi Dress, GOTS-certified, relaxed fit, for everyday and workwear, in Forest Green, sizes XS to XL. A breathable 100% organic cotton midi in a relaxed fit; model is 1.78m wearing size M. Runs true to size. Size up if you prefer an oversized look or are between sizes. Free 30-day returns.”
7. Answer the real questions in FAQPage schema
Add six to eight Q&As per product, wrapped in FAQPage structured data, answering what apparel shoppers actually ask AI: “Does this run true to size?”, “What are the exact measurements?”, “What is it made of?”, “How do I wash it?”, “Is it available in my size?”, “What is your return policy?”, “Is the fabric certified?”. Each answer should carry a specific data point, not generic reassurance. This is the same pattern described in our conversational content guide.
One prerequisite holds all seven together: in fashion the most valuable review is the one that says “ordered my usual M, fits true”, because that is the line AI lifts to answer a sizing question, and it is worthless to the model if it lives only in a JavaScript widget. Most AI crawlers never run that JavaScript, so server-render the reviews, and keep the star rating on the Product as AggregateRating rather than the Organization, since Google reads a site-wide self-rating as self-serving and leaves it out of rich results. Encourage structured fit feedback, and let Verity confirm the reviews are crawlable and the AggregateRating sits where Google accepts it. See reviews and AI.
The technical layer: feed, crawlers, schema
The content levers above do most of the work. The technical points below are fewer, but they are where apparel feeds pick up stale advice, so here is what the official documentation actually says in 2026.
ChatGPT Shopping feed. On Shopify, your catalog is already wired into ChatGPT through Shopify’s integration, so no separate feed is needed, per OpenAI’s merchant documentation. For apparel the spec is specific: size and size_system are recommended, color is optional, and group_id (with listing_has_variations) is how you bind a size-color grid into a single listing; gtin is optional (OpenAI Commerce docs, 2026). Whatever fit and measurement data is in the feed has to agree with the live page, because the model checks one against the other before it recommends. Worth noting too: in early 2026 OpenAI began moving away from in-ChatGPT instant checkout toward merchant-owned checkout, which makes discovery, not checkout, the durable bet here. See our walkthrough on selling on ChatGPT for Shopify.
Perplexity Merchant Program. No cost to join: it draws on the same Shopify integration for stores shipping to the US, and the apparel cards it returns are organic placements, not ads (Perplexity Merchant ToS, May 2025). More on Perplexity Shopping.
robots.txt. At minimum, allow OAI-SearchBot, ChatGPT-User, PerplexityBot and Googlebot. The myth that trips up a lot of brands is that blocking GPTBot takes you out of ChatGPT; in fact GPTBot and Google-Extended only control training data, while your presence in ChatGPT search is decided by OAI-SearchBot. Verity probes each crawler tier (search, user, training) against your robots.txt so a one-line block does not quietly hide your catalog. See robots.txt for AI crawlers.
Schema.org. Use ProductGroup with hasVariant for the size-color grid, MerchantReturnPolicy for returns, SizeSpecification and WearableSizeGroup for sizing, and additionalProperty for measurements, composition and certifications, alongside the standard brand, gtin, offers and shippingDetails. Full detail in our schema.org for Shopify guide.
Off-site: where fashion AI authority is really built
Because a large share of cited sources are not your own site, off-site presence is part of GEO, not separate from it.
Editorial roundups carry weight because they are hard to fake. “Best sustainable basics” and “best jeans for [body type]” lists from outlets that run real testing are treated as high-authority by AI. Getting included means submitting real product data and surviving comparison, which is why those lists get cited.
Marketplace and platform reviews feed fit-matched recommendations. Reviews that mention the reviewer’s usual size and how the item fit (“I am normally a 10 and the 10 fit perfectly”) are the ones AI extracts to answer “does it run true to size?”. Encourage structured fit feedback across your own server-rendered reviews and any marketplace presence.
Communities influence AI, mostly through training data. Threads on fit, quality and sizing in fashion communities shape what models associate with your brand, but the legitimate path is genuine participation, not astroturfing, which violates platform policy and carries disclosure risk.
Your 30/60/90 plan
- Days 1 to 30, foundation. Move every size guide out of images into HTML tables and structured size data. Mark up variants with ProductGroup (size and color) and confirm per-variant availability. Add fibre composition and care as structured properties. Check robots.txt allows OAI-SearchBot, ChatGPT-User, PerplexityBot and Googlebot.
- Days 31 to 60, content and returns. Add MerchantReturnPolicy structured data that matches your returns page exactly. Rewrite top product descriptions answer-first with fit and material up top. Add six to eight FAQs per hero product in FAQPage schema. Audit every sustainability claim against the substantiation table and name the certification or remove the claim.
- Days 61 to 90, authority and measurement. Expose certifications (GOTS, OEKO-TEX) as structured facts with links to public certificates. Pursue two or three editorial roundups. Test your category queries monthly across ChatGPT, Perplexity, Gemini and Claude, tracking whether you appear, in what position, and whether the fit and return details are accurate. Since June 2026, Google Search Console also has a generative AI performance report for a free first-party view of where you surface in AI answers (rolling out to a subset of sites first).
How Verity Score fits in
Verity Score reads a Shopify store the way an agent resolving a fit question would, and the fashion vertical is built in. It checks whether your size guide and measurements are present as text and structured data (not trapped in an image), validates that size and color variants are coherent and have per-variant availability, confirms fibre composition and care are exposed, reads your MerchantReturnPolicy against your returns page, flags sustainability and material claims with no backing data, detects reviews that load only via JavaScript, validates AggregateRating against Google’s self-serving rule, and probes which AI crawlers your robots.txt allows. Each finding comes with the fix.
Fashion is the category where fit decides the sale, returns decide the margin, and sustainability claims now carry real legal weight. The brands structuring that data now are the ones AI will name when a shopper asks for a dress that will actually fit and that they can return if it does not.
Curious what an AI makes of your size guide and variants? Run a free GEO audit in 60 seconds.