What is Agentic Commerce Optimization (ACO)?
Agentic Commerce Optimization (ACO) is the practice of making product catalogs found, understood and recommended by AI shopping agents.
Agentic Commerce Optimization, defined
Agentic Commerce Optimization (ACO) is the practice of making every product in a catalog found, understood and recommended by AI shopping agents such as ChatGPT, Google Gemini and AI Mode, Perplexity, Claude, Microsoft Copilot, Alexa for Shopping, Muse by Meta and Instinct. It works on the product data itself: product detail pages (PDPs), category pages (PLPs) and product feeds.
ACO in 30 seconds
- 01
Shoppers increasingly ask an AI agent what to buy instead of scrolling a results page: 68% of US consumers used an AI tool for shopping in the last three months (McKinsey and ICSC, 2026). The agent answers with a handful of products, not ten blue links.
- 02
Agents choose products by reading product data: attributes, descriptions, specs, use cases, reviews and structured feeds. Missing or vague data means the product is skipped.
- 03
SEO optimizes pages to rank. GEO optimizes content to be cited. ACO optimizes the product catalog to be recommended and bought.
- 04
ACO has three jobs: measure how ready each SKU is for AI agents, fix what is missing, and publish the fixed data everywhere agents read it.
- 05
The brands that do it now become the default answers agents learn to give. The brands that wait become invisible in the channel that is growing fastest.
Why now: the numbers
AI shopping went from experiment to the fastest-growing, best-converting traffic source in eCommerce in under two years. The window to become the product agents recommend by default is open now.
in US B2C retail revenue orchestrated by AI agents by 2030; $3T to $5T globally
A channel the size of today's eCommerce, not a nicheSource: McKinseyof US consumers used at least one AI tool for shopping in the last three months
The shopper is already thereSource: McKinsey and ICSC, 2026AI traffic to US retail sites in holiday 2025; +393% in Q1 2026
Exponential growth, year after yearSource: Adobe Digital Insightsbetter conversion for AI traffic in March 2026, after converting 38% worse a year earlier
AI visitors arrive ready to buySource: Adobe Digital Insightshigher conversion for AI-referred retail visitors than non-AI traffic in July 2026
The gap keeps wideningSource: Adobe AI Traffic Trends, Aug 2026shopping questions asked to ChatGPT every week in the US, up from under 1% to over 8% of Amazon's search volume in less than a year
AI assistants are becoming a shopping destinationSource: StacklineThree shifts behind the numbers
From results pages to answers. A search engine shows ten links and lets the shopper compare. An AI agent compares for the shopper and answers with three to five products. Being on page one is no longer enough; you have to be in the answer.
The platforms are building the rails. OpenAI and Stripe launched the Agentic Commerce Protocol (ACP). Google launched the Universal Commerce Protocol (UCP) and, in May 2026, rolled out conversational attributes in Merchant Center globally, built for how people ask questions in AI Mode (Lengow).
Discovery is the battleground. In March 2026 OpenAI scaled back native checkout in ChatGPT to focus on product discovery (TechWyse). Whoever wins the recommendation wins the sale, and recommendations are decided by product data.
How AI shopping agents choose products
An AI agent does not look at your product photos or your design. It reads your product data, and if a detail is not written down, for the agent it does not exist.
- 1
The shopper asks in natural language
"A warm cashmere sweater for the office, relaxed fit, under €300." One request carries a product type, a material, a use case, a fit and a budget.
- 2
The agent fans the question out
It splits the request into many sub-queries (cashmere quality, office-appropriate styles, relaxed fits, price ranges) and runs them in parallel. Google calls this query fan-out in AI Mode.
- 3
It retrieves candidates
From product feeds (Google Merchant Center, OpenAI's ACP feeds), from the product pages it can crawl, from marketplaces and from reviews and third-party sites.
- 4
It filters on attributes
Products whose data does not confirm a constraint are dropped. If your page never says "100% cashmere" or "relaxed fit", the agent cannot match it.
- 5
It ranks and justifies
The agent picks the products it can explain best: clear use cases, benefits, answers to likely questions. A vague description gives it nothing to say.
- 6
It shows three to five product cards
The shopper usually picks from these. Everything that was not selected is invisible.
ChatGPTWhy products get skipped
| Gap | Example |
|---|---|
| Missing attributes | Composition, fit, size range or compatibility not stated in text or feed |
| Missing keywords and intents | The page says "beige sweater", shoppers ask for "women's cashmere crew neck" |
| No use cases | Nothing says who the product is for or when to use it |
| Inconsistent or wrong data | Different specs on the PDP, the feed and marketplaces |
| Weak metadata | Generic title, H1 and meta description |
ACO fixes these gaps at the source: the product catalog. That is why it is a catalog discipline, not a content marketing tactic.
ACO vs SEO vs GEO
SEO gets pages ranked, GEO gets content cited, ACO gets products recommended and bought. They build on each other, but only ACO is designed for how AI agents shop.
| SEO | GEO (also AEO, LLMO) | ACO | |
|---|---|---|---|
| Goal | Rank pages in search results | Be cited in AI-generated answers | Be the product AI agents recommend and sell |
| Who reads it | Search engine crawlers, then humans | Language models answering questions | AI shopping agents comparing products for a buyer |
| Unit of work | The web page | The article, brand mention or source | The SKU: every product in the catalog |
| What gets optimized | Keywords, links, technical health | Content, authority, citations | Attributes, descriptions, use cases, intents, feeds, PDPs and PLPs |
| Where it lives | The website | Website, PR, third-party content | Product catalog, PIM, store, product feeds |
| Success metric | Rankings, clicks | Mentions, share of voice | AI Readiness of every SKU, recommendations, AI-referred revenue |
| Scale | Hundreds of pages | Dozens of topics | Thousands of SKUs in every language and market |
What this means in practice
- GEO tools can tell a brand that it is invisible in AI answers. For an eCommerce company the fix almost always lives in the product catalog, which GEO does not touch.
- ACO is built for physical products: it works SKU by SKU, attribute by attribute, across every language and market a brand sells in.
- Good ACO also lifts SEO and Google Shopping, because complete, intent-rich product data is what every discovery channel rewards.
The ACO framework
ACO runs as a closed loop of five pillars: analyze the catalog, create what is missing, optimize product pages, optimize category pages, then publish and re-measure. Skipping a pillar leaves a gap that agents will find.
Analyze
Which products are ready for AI agents, and which matter most?
AI Readiness Score per SKU, market potential and search demand, prioritization
Create
Do all my products have a complete page in every market?
New pages from brand technical data, or from supplier data with gaps filled from other sources
Optimize product pages
Can an agent match, understand and justify each product?
Attributes, titles, descriptions, bullets, FAQ, use cases, intents, metadata
Optimize category pages
Do I appear when shoppers ask for a category?
Category pages (PLPs) built around real category demand, at scale
Publish and re-score
Is the fixed data live everywhere agents read it, and did it work?
Store, PIM, Google Merchant Center and marketplaces, then re-run the score
Re-score and repeat every quarter
The four dimensions of an agent-ready product page
Every product page is judged on four dimensions. They are the same four that make up the AI Readiness Score in the AndromedAI platform.
Product Data Completeness
Every attribute an agent needs to match the product to a request: material, size, fit, dimensions, compatibility, specs, category, tags. Aligned with the attribute requirements of OpenAI's ACP and Google's UCP.
Keyword Coverage
The words and phrases shoppers actually use, in search engines and in AI prompts, present in the title, description, bullet points and FAQ.
Customer Intent Match
The questions behind the purchase: who the product is for, when to use it, what problem it solves, how it compares. This is what lets an agent justify a recommendation.
Shopping Metadata
Page title, H1 and meta description written for both search engines and AI surfaces.
Brands and retailers start in different places
Brands
Brands own their data but usually write it for humans: beautiful copy, missing specs. Their priority is completeness and intent coverage in every market.
Retailers
Retailers sell thousands of products from many suppliers, with thin and inconsistent data. Their priority is creating complete pages at scale and filling gaps from external sources.
Measuring ACO: the AI Readiness Score
Visibility tools measure the symptom: whether a product showed up in an AI answer today. The AI Readiness Score measures the cause: whether each product's data gives an agent what it needs to recommend it. You can act on the cause.
How the score is built
| Component | What it reads | What it checks |
|---|---|---|
| Product Data Completeness | Structured attributes and metafields, category, tags, product details | Are all the attributes an agent needs present, measured against ACP and UCP attribute rules? |
| Keyword Coverage | Description, short description, bullet points, FAQ | Does the page use the terms shoppers search and ask with? |
| Customer Intent Match | Description, short description, bullet points, FAQ | Does the page answer the questions behind the purchase, based on AI query fan-out? |
| Shopping Metadata Score | Page title, H1, meta description | Are titles and metadata specific and intent-rich? |
Example: one product page, scored
page_title
Cashmere Sweater Beige
Women's Cashmere Crew Neck Sweater in Oatmeal, Relaxed Fit
AI Readiness
31/100
from 31
- Keyword coverage
28%94% - Attributes present
2 of 88 of 8 - Use cases covered
0 of 44 of 4 - Shopper intents covered
2 of 66 of 6
Illustrative example based on AndromedAI's demo product page.
Four ways to track ACO results
The AI Readiness Score tells you what to fix. These four reports, all available today, show what changes once the fixed pages are live. Check them before and after every optimization.
Your products' share of voice in Google's AI searches, split by shopping stage (discovery, evaluation, ready to buy) and by the search intents behind them, compared with your competitors.
How to set it upMerchant Center, Analytics, Products, AI performance. Filter by category and country, then look for the intents where few of your products show up.
AI performance insights for Knitwear
Share of voice by shopping stage
Top search intents
The impressions your pages earn in Google's AI features, such as AI Overviews and AI Mode. Filtered on product page URLs, it shows which PDPs AI actually surfaces.
How to set it upSearch Console, Performance, Generative AI. Add a page filter that contains /products/ (or your PDP path) and compare the weeks before and after optimization.
Sales and visits from the AI channels connected to your Shopify catalog, such as ChatGPT, Microsoft Copilot, Meta AI and Shop, and the recent searches where your products appeared.
How to set it upShopify admin, Agentic. Check revenue by channel and test queries against the Shopify Catalog to see how your products rank.
Users, sessions, key events and revenue that AI assistants send to your store. ChatGPT tags its links with utm_source=chatgpt.com; Gemini, Perplexity and Copilot arrive as referrals.
How to set it upAcquisition, User acquisition, dimension First user source / medium, filtered with a regex that matches chatgpt, gemini, perplexity and copilot. Save it as a comparison to track it every month.
AI performance insights for Knitwear
Share of voice by shopping stage
Top search intents
The ACO playbook: 7 steps
Any brand or retailer can start ACO this quarter. These are the steps, in order.
- 1
Baseline your catalog
Score every SKU, or a representative sample, with the AI Readiness Score. Most teams discover that the majority of their products miss basic attributes.
- 2
Prioritize by market potential
Start with the products and categories that carry the most search and AI demand, not with the whole catalog at once.
- 3
Complete the product data
Fill every missing attribute in the store and in the feed: material, size, fit, dimensions, compatibility, specs. Make the PDP, the feed and your marketplaces say the same thing.
- 4
Rewrite for intents, not just keywords
Titles, descriptions, bullet points and FAQ should answer who the product is for, when to use it and how it compares, using the words shoppers actually use.
- 5
Create the pages you are missing
New products, new markets, new languages: launch them complete and AI-ready from day one, from brand technical data or from supplier data.
- 6
Cover category demand
Build category pages (PLPs) for the category questions shoppers ask ("best cashmere sweaters for the office"), not only for your internal taxonomy.
- 7
Publish everywhere, then re-score
Push the improved data to your store, PIM, Google Merchant Center and marketplaces. Re-run the score every quarter and after every major catalog update.
ACO with AndromedAI: platform and results
AndromedAI is the Agentic Commerce Optimization platform that does not stop at measuring: it fixes every product page and publishes it back to your store, in every language.
| ACO pillar | AndromedAI | What it does |
|---|---|---|
| Analyze | AI Readiness Score | Scores every SKU, shows what is missing and ranks products by market potential |
| Create | Creator | Generates complete product pages from brand technical data, or from supplier data with missing information retrieved |
| Optimize product pages | Optimizer | Agents rewrite titles, descriptions, bullet points and FAQ, extract attributes and add use cases and intents |
| Optimize category pages | PLPs | Builds category pages at scale around real category demand |
| Publish and re-score | Integrations | Publishes to Shopify, Salesforce Commerce Cloud, Adobe Commerce, Shopware, WooCommerce, Akeneo, Plytix and Google Merchant Center, then re-scores |
Results from brands and retailers doing ACO
clicks from AI chats (ChatGPT, Gemini, AI Mode)
Bomboogiesales and +160% organic traffic, 483 hours saved
Semprefarmaciato the first sales from ChatGPT, and -95% catalog generation costs
Matassaadd-to-cart in two months, same marketing budget
Altaforma Milanobounce rate on product pages
AusiliumFAQ
ACO is the practice of making every product in a catalog found, understood and recommended by AI shopping agents such as ChatGPT, Gemini, Google AI Mode, Perplexity, Claude, Copilot, Alexa for Shopping, Muse by Meta and Instinct, by optimizing the product data itself.
SEO optimizes web pages to rank in search results. ACO optimizes the product catalog, SKU by SKU, so AI agents can match products to a shopper's request and recommend them.
GEO focuses on getting content and brands cited in AI answers. ACO focuses on products: attributes, intents, use cases and feeds, so agents recommend and sell them.
All agents that read product data: ChatGPT, Google Gemini and AI Mode, Perplexity, Claude, Microsoft Copilot, Alexa for Shopping, Muse by Meta and Instinct, plus Google Shopping and the new shopping agents that keep launching.
Yes. SEO pages are written to rank, not to answer a shopper's specific constraints. Agents need complete attributes and explicit use cases that most SEO content does not include.
Optimized pages can be live within days. Results depend on how fast agents and search engines re-read the pages; customers have seen first AI-driven sales within a week.
Yes. Retailers can create complete product pages from supplier data, with missing information retrieved from other sources.
Score your product pages. AndromedAI's free AI Readiness Audit scores 3 of your product pages, shows what is missing and fixes them.
Glossary
- Agentic commerce
- Shopping in which AI agents search, compare and sometimes buy on a shopper's behalf
- AI shopping agent
- An AI assistant that recommends or buys products, such as ChatGPT, Gemini or Perplexity
- ACP
- Agentic Commerce Protocol, the open standard by OpenAI and Stripe for agent-led shopping
- UCP
- Universal Commerce Protocol, Google's standard for shopping inside AI Mode and Gemini
- Query fan-out
- The technique by which an agent splits one request into many sub-queries
- Conversational attributes
- Product attributes in Google Merchant Center built for how people ask questions in AI Mode
- PDP
- Product detail page
- PLP
- Product listing page, or category page
- AI Readiness Score
- AndromedAI's score of how ready a product page is to be recommended by AI agents
- Product Data Completeness
- Share of the attributes an agent needs that are present for a product
- Customer Intent Match
- How well a page answers the questions behind a purchase
Featured in
