跳到正文
DEV Community· Daniel Wishnia·· 4 小時前AI 評分60

AI 購物數據對電商商品資料的啟示:先完善商品資料,再考慮結帳整合

Agentic Commerce for Builders: What the AI Shopping Data Means for Your Product Feed

AI 導讀

AI 購物時代,電商團隊應先完善機器可讀、完整一致的商品資料,而非優先整合結帳。文中列出的不同口徑數據顯示,美國樣本中 6.9% 的 ChatGPT 對話有購買意圖,Adobe 2025 年美國假日數據則顯示 AI 引薦流量的轉化率高 31%,兩者不可直接比較。

正文

If you build or maintain an online store, you've probably been asked some version of "should we integrate checkout into ChatGPT?" this year. The honest engineering answer, based on what the platforms themselves now say, is: the integration that matters first is not checkout. It's your data.

This post walks through the numbers, the protocols, and the parts of your stack that decide whether an AI assistant can cite your products at all.

A quick note on where I'm coming from. I've worked in SEO since the beginning, in the AltaVista days, and I've watched visibility algorithms change over and over. My work has covered eCommerce retail, real estate, hospitality and startups, and I mentor founders and teams on digital exposure and their digital footprint. I try to think one step ahead and keep common sense in the loop. I'm an Event Ambassador at E-commerce Berlin Expo, where I've also spoken, and I've sat on the jury of the E-commerce Germany Awards. In February 2027 I'll be on that jury again in Germany.

TL;DR

  • Shopping is a small slice of ChatGPT use: 6.9% of conversations in a U.S. panel showed purchase intent. [1] [2]
  • Traffic that does arrive from AI converts better: 31% more often than other traffic in Adobe's 2025 U.S. holiday data. [5]
  • OpenAI is moving away from standalone in-chat checkout and toward discovery driven by product feeds. [8]
  • Your job: make product data machine-readable, complete and consistent, then measure AI referrals as their own channel.

The data, as a dataset

Treat these as separate measurements with different methods. They are not directly comparable.

Source Measured Result Ref
SparkToro / Verso U.S. opt-in panel, Jan to Oct 2026 ChatGPT conversations with purchase intent 6.9% total: 4.9% weighing options, 2.1% about a specific brand; 93.1% none (figures rounded) [1] [2]
Same panel ChatGPT responses naming no brand, product or business 87.9% [2]
Profound, June 2025 ChatGPT prompts by intent 9.5% commercial, 6.1% transactional [3]
Profound, Sep 2025 to Jan 2026 Shopping card trigger rate About 9% overall; 12.1% open-ended vs 3.1% brand-named [4]
Adobe Analytics, 2025 U.S. holiday season AI referrals to U.S. retail sites Traffic up 693% YoY; converted 31% more often; 45% more time on site; revenue per visit up 254% [5]
Salesforce, 2025 holiday season AI search referrals vs social Converted 9 times more often [6]

The engineering takeaway from row four: open-ended prompts ("lightweight running shoes for flat feet") trigger shopping about four times as often as brand-named ones. [4] The model picks products when the user doesn't. That selection runs on whatever data it can parse.

The protocol layer: ACP and UCP

Two open protocols are worth knowing.

Agentic Commerce Protocol (ACP). OpenAI and Stripe released it on 29 September 2025 alongside Instant Checkout, starting with U.S. Etsy sellers. [9] The merchant stays merchant of record; orders, payments and fulfillment run through the merchant's existing systems. [9] OpenAI's merchant page now describes ACP as supporting "product discovery and more accurate, personalized recommendations through shared product data", and says it is "moving away from a standalone Instant Checkout experience", with purchases completed on merchant websites or apps. [8]

Universal Commerce Protocol (UCP). Google announced it on 11 January 2026, co-developed with Shopify, Etsy, Wayfair, Target and Walmart. It is meant to power checkout on eligible listings in AI Mode in Search and the Gemini app, with the retailer remaining seller of record. [10]

So the pattern in late 2026: the agent protocols exist, but in ChatGPT the live value is discovery from shared product data, and the transaction lands on your site. If you're prioritizing work, feed quality comes before checkout plumbing.

What "AI-readable" means for your catalog

The platforms spell this out. OpenAI invites merchants to share product feeds and says richer, up-to-date data helps products surface with more accurate information. It also says a feed isn't required if ChatGPT already crawls your site, but feeds give more control. [8] Shopify generates a catalog for AI channels automatically and lists what helps: clear titles, full-sentence descriptions, technical details, sizing and care information, public return policies, and reviews. [7]

A practical checklist for the catalog:

  • Every product has a specific title and a full set of attributes: material, size, variant, price, availability.
  • Delivery times, shipping costs and return terms are published and match across your site, feed and marketplaces.
  • Descriptions are written in sentences that answer "who is this for" and "how does it compare", not keyword lists.
  • Common pre-purchase questions are answered on the product page itself (Q&A content).
  • Reviews are collected and visible, and your facts are consistent on third-party sites.

I call the resulting dynamic the Citation Economy, a term I coined: search made us compete for clicks, AI makes us compete for citations. The path is cited, compared, chosen, bought. Generative Engine Optimization (GEO), making pages and product data easy for AI assistants to read, trust and cite, is just SEO with a new reader: the model.

Example: a jewelry store's product page

A hypothetical German online jewelry shop gets this kind of request through an assistant: a minimalist gold necklace, hypoallergenic, within a budget, delivered before a birthday. No brand named.

A product record that only carries title: "Necklace gold" gives the model nothing to match. One that states metal, plating, allergy information, delivery time and return policy can be matched to every constraint in the prompt, and to the follow-up ("which is better for sensitive skin?") if that answer is on the page. Same product, very different odds of being cited.

Measuring it

Shopify recommends tracking AI referrals separately from organic search, using custom filters or reports in GA4. [7] A simple starting point is a custom channel group or exploration filter on session source matching the assistants' referrer domains, for example chatgpt.com, perplexity.ai and gemini.google.com. Then compare conversion rate and revenue per session against organic search.

Add one manual step: once a month, ask the assistants your key buyer questions by hand and log whether your products are cited. ChatGPT's shopping features are U.S.-only for now [8], so test in the markets you actually sell in.

What about ads?

OpenAI began testing ads in ChatGPT in the U.S. on 9 February 2026, matched to conversation topic, and says ads don't influence answers. [11] It added CPC bidding and a self-serve Ads Manager in May 2026 [12], and in September 2026 launched a ChatGPT Ads app for U.S. Shopify merchants plus a Sponsored Agents test. [13] Ads sit next to the answer. Getting cited is still how you get into it.

Wrapping up

For developers the useful reframe is this: AI shopping in 2026 is less an integration project and more a data-quality project. Ship clean, complete, consistent product data, publish the answers shoppers ask for, and instrument AI referrals as a first-class channel. Checkout can follow when the platforms and your numbers say it's worth it.

Sources

  1. SparkToro, "We Analyzed 118,000 ChatGPT Prompts from 12,000+ Users to Answer: How Do Americans Use AI Tools in 2026?", 17 Nov 2026.
  2. Search Engine Journal, "6.9% Of ChatGPT Conversations Showed Purchase Intent In U.S. Panel", 17 Nov 2026.
  3. Profound, "AI Search intent study: What 50M+ ChatGPT prompts reveal", 25 Jun 2025.
  4. Profound, "We tracked 2 Million ChatGPT prompts. Shopping showed up less than 10% of the time.", 3 Mar 2026.
  5. Digital Commerce 360, "Generative AI shifts online holiday shopping traffic in 2025" (Adobe Analytics data), 13 Jan 2026.
  6. Salesforce, "Holiday Season Rakes in Record $1.29T for Retailers, Salesforce Data Shows", 8 Jan 2026.
  7. Shopify, "How ChatGPT Shopping Works for Ecommerce Stores", 15 Sep 2026.
  8. OpenAI, "Power product discovery in ChatGPT" (merchants page), accessed 21 Nov 2026.
  9. OpenAI, "Buy it in ChatGPT: Instant Checkout and the Agentic Commerce Protocol", 29 Sep 2025.
  10. Google, "New tech and tools for retailers to succeed in an agentic shopping era", 11 Jan 2026.
  11. OpenAI, "Testing ads in ChatGPT", 9 Feb 2026, updated 11 Aug 2026.
  12. OpenAI, "New ways to buy ChatGPT ads", 5 May 2026.
  13. OpenAI, "Reimagining advertising with AI", 16 Sep 2026.

Daniel Wishnia - WishOL. More on AI visibility and the Citation Economy at wishol.com.

來源:DEV Community · dev.to