Fashion Product Titles for ChatGPT Recommendations: How to Order Attribute Keywords and Match Feed Fields

Type "Recommend a women's wool coat for fall office wear" into ChatGPT and check whether your product shows up. For most fashion brands, it doesn't. The photos are polished and the copy has a mood, but if the title reads something like "Daily Mood Half Coat," the AI has nothing to tell it the coat is wool, made for women, or meant for fall. Getting fashion product titles into ChatGPT recommendations comes down to which attribute keywords you write, in what order, and where. This page covers the title formula, the platform specs, bad and good examples, an audit sequence, and how to measure results.

30-second definitions: attribute keywords and AI recommendation visibility

• Attribute keywords are words that state a product's objective traits, such as material, fit, season, and audience.
• A product title is the one-line name that identifies a product in the product feed and on the product detail page.
• A variant is an individual item of the same product that differs only in a selectable value, such as color or size.
• An attribute-based query is a recommendation request with attributes attached as conditions, such as "women's wool coat for fall office wear."
• AI recommendation visibility is the state in which a product or brand appears in the recommendation list of an AI answer, such as one from ChatGPT.

According to OpenAI, ChatGPT shows products it judges to match the user's intent, weighing structured metadata from the merchant and third-party providers (price and product descriptions, for example) along with third-party content. OpenAI states that product results are not ads and are not influenced by affiliate relationships (OpenAI Help Center, 2026)^[1]. The title is the first field in that metadata.

The fashion title formula: what order should attribute keywords follow?

The order comes straight from the two platform specs. Google Merchant Center limits titles to 1 to 150 characters, notes that shoppers usually see only the first 70 characters or fewer depending on screen size, and tells merchants to put the most important details first. It says to include the brand when the brand sets the product apart, and to put variant-distinguishing details (size, color) both in the title and in the matching attribute fields (Google Merchant Center Help, 2026)^[3]. Google's own example, "Sleeveless Polka Dot Skater Dress, Small, Black," runs attributes, product type, size, then color^[3].

The OpenAI product feed spec recommends a title of up to 150 characters and asks merchants to include the selected variant where it applies, as in "Trail running shoes, black, size 10." The description field takes plain text with a recommended maximum of 5,000 characters, material holds the primary material, and gender accepts male, female, or unisex. For optional fields with unknown values, the spec says to leave them out entirely rather than enter placeholders like null or n/a (OpenAI Product Feed Spec, 2026)^[2].

Sources: Google Merchant Center Help (2026)^[3], OpenAI Product Feed Spec (2026)^[2]. Both specs can change, so check the originals again when you submit. The feed submission process itself is covered in the ChatGPT shopping feed setup guide.

Evidence: what happened when titles lacked attributes

GEO agency GeoRank published a six-month project with fashion brand K. At the start, Brand K's name did not come up even once for the query "recommend daily minimal outerwear," even though the brand was releasing 13 to 18 new products every month. GeoRank traced the problem to two gaps: product titles were missing core attributes such as thickness, season, and body type, and the product detail pages leaned on mood copy (GeoRank, 2026)^[4].

The fix was to fill in all 16 feed fields and rewrite the detail copy in a subject, verb, object structure. GeoRank reports that Brand K began appearing in ChatGPT recommendations from month four, and that the conversion rate of traffic from AI channels over the six months was 2.3 times that of its existing ad channels. The agency also cited its own analysis showing that brands with consistent titles and attributes across channels appeared in AI Overviews about twice as often (GeoRank, 2026)^[4]. Both figures come from a single agency's own case and internal analysis. Treat them as directional evidence, not as a general effect size.

Shopper queries are moving toward attributes as well. When Musinsa joined Kakao's ChatGPT for Kakao as a fashion partner in March 2026, it explained that users could ask questions describing time, place, occasion, weather, and brand preference without knowing the exact product name. Its sample query: "Recommend an outfit that works for the office tomorrow" (Herald Economy, 2026)^[6]. To be a candidate for questions like that, a product's data has to spell out attributes such as "office wear" or "fall." Mobiinside likewise named AI search and zero-click, alongside algorithmic feeds and social search, as places where fashion brands now compete to be discovered, and urged brands to shift to AEO so they get mentioned in AI answers (Mobiinside, 2026)^[7].

Appearing is not the same as being mentioned first: the AIBIX fashion study

Getting your name into a recommendation list and being named first are different outcomes. AIBIX Lab's AIBIX fashion study asked ChatGPT, Gemini, Perplexity, and Claude about 193 brands across 7 categories. Asics appeared in 97% of running shoe answers but was mentioned first in only 22.5%. Omega appeared in 100% of luxury watch answers, with a first-mention share of 5%. Of the 193 brands, 108 (56%) were never mentioned first (Wikitree, 2026)^[5].

The same study broke down the sources behind those answers. In golfwear, personal blogs made up 51.8% and official brand sites 2.2%. In luxury watches, personal blogs accounted for 39.9% and official sites 3.4% (Wikitree, 2026)^[5]. Fixing titles on your own store is not enough. Titles and attributes on marketplaces, reviews, and blogs need the same values so the AI reads the same product with the same attributes across sources. Category rankings are in AIBIX fashion brand AI recommendation rankings.

Bad vs. good: product titles compared by attribute keyword

Every example below is a fictional title made up for illustration. None of it is real brand or sales data.

Three mistakes come up again and again. The first is filling the first 70 characters with adjectives that almost never appear as conditions in a query, such as "mood," "aesthetic," or "daily." The second is a mismatch: the title says "Black" while the color field says "BK." The third is listing the same product several times, once per color, with only slight changes to the title. GeoRank warns that creating fake variant products draws a penalty (GeoRank, 2026)^[4]. Variants belong inside a single product group, distinguished by their color and size values.

Six-step audit for fashion title attribute keywords

1. Build an attribute-based query list: For each product type, write 20 to 30 queries that include the conditions shoppers ask about (audience, material, season, use, body type). Examples: "women's wool coat for fall office wear," "wide-leg slacks for short men."
2. Log current visibility: Run those queries in ChatGPT, Perplexity, Gemini, and other engines, and record whether your products and brand appear and whether they are mentioned first. A single run fluctuates, so repeat each query several times.
3. Rewrite titles: Write brand (if distinctive), audience, core attributes, product type, color, and size in that order, with core attributes in the first 70 characters. Drop promotional text.
4. Match feed fields: Fill gender, material, color, and size with exactly the same wording as the title. Leave unknown values empty, without placeholders (OpenAI Product Feed Spec, 2026)^[2].
5. Align detail copy and external channels: Spell out attributes in full sentences on the product detail page, such as "This coat is made of a 70% wool blend," and give marketplace listings on Naver Smart Store, Coupang, and similar channels the same attribute order. For Product schema, see the structured data guide for AEO.
6. Remeasure: Use the same query set from step 2 and log the changes in each engine again. In the GeoRank case, first visibility took four months, so track monthly (GeoRank, 2026)^[4].

Comparing tools that measure attribute keyword impact

Done by hand, the repeated measurement in steps 2 and 6 grows with every query and every engine you add. AI visibility tools automate that repetition. The table below is based on public information, and features and pricing may change.

When choosing a tool, look past price. Ask how accurately it reproduces attribute-based queries and whether its diagnosis translates directly into product data fixes (titles, feed fields, schema). For a Korean fashion brand, the first test is whether the tool handles Korean queries and domestic channel data. BOIDA's operator, Designovel, started out in fashion AI. For a wider comparison of companies, see GEO recommended companies.

Summary

Whether a fashion product shows up in ChatGPT recommendations depends on attribute keywords, not mood copy. ChatGPT weighs structured metadata such as price and product descriptions^[1], and the Google and OpenAI specs both tell merchants to put the most important details first within 150 characters and to place variant details in both the title and the fields^[2][3]. Put audience, material, fit, season, and use within the first 70 characters, and write the same values in the feed fields, the detail copy, and external channels. Then measure appearance and first mention by engine with attribute-based queries, again and again. For the full picture of fashion GEO, read fashion commerce GEO and what is GEO. Product recommendations in general are covered in ecommerce product recommendation GEO.