What AEO, GEO and ACO actually measure, and what it means for brands

Key takeaways

• AEO and GEO are one discipline. Both measure whether an LLM such as ChatGPT, Gemini or Perplexity mentions and cites a brand. Every brand needs that visibility, and it sits at the top of the funnel.

• Agents recommend products, so commerce brands need ACO. An AI shopping agent picks a specific SKU, in a specific size and variant, and links to buy it. ACO is built for that decision.

• ACO covers the agents where products are bought. That means answer engines, retailer shopping assistants such as Alexa for Shopping and Walmart Sparky, and personal AI agents such as Meta Muse. Success is measured in SKU-level discoverability, revenue and conversion.

• Every agent reads the web differently. In September 2026, Alexa for Shopping took 87.6% of its citations from editorial and affiliate publishers, while Google AI Overviews leaned most on retailer sites. One strategy cannot cover both.

• The 5Cs are how ACO gets done. Completeness, Context, Citations, Correctness and Customer Acquisition, worked SKU by SKU and agent by agent.

• Azoma optimizes products for AI shopping agents for enterprise consumer brands across Alexa for Shopping, Walmart Sparky, ChatGPT, Gemini and Meta Muse. See how it works.

Answer Engine Optimization (AEO) or Generative Engine Optimization (GEO) both do one job well: they get a brand mentioned in an AI answer. For a brand, that is only where the work starts. The sale happens further down.

What do AEO, GEO and ACO each measure?

AEO and GEO measure brand visibility in large language models (LLMs). In practice they are the same discipline: same LLMs, same tactics, same metrics of mentions, citations and share of voice. Wikipedia records that the terms are frequently used interchangeably, and Azoma treats them as one.

Agentic Commerce Optimisation (ACO) is the practice of making a brand's products discoverable, understandable and recommendable by AI shopping agents across ChatGPT, Gemini, Amazon Rufus, Alexa for Shopping, Walmart Sparky and Perplexity.

What ACO and AEO/GEO each measure.

Why do brands need ACO, and why isn't GEO/AEO enough?

Commerce brands need ACO because the agent chooses the product, and GEO/AEO stops at the brand. A brand can be named in the answer while the agent recommends a competitor's stock-keeping unit (SKU) and links the shopper straight to it.

GEO misses the product. GEO tests whether the brand is mentioned. The agent is asking something more specific: which product, in which size, variant and pack. It answers from structured product data, and a single SKU on a global retailer can need more than 600 attribute values to count as complete, according to our 5Cs of Agentic Commerce guide. Brand coverage cannot fill those fields.

GEO misses the agents where brands sell. AI shopping agents come in three types:

• Answer engines such as ChatGPT, Google Gemini and Google AI Overviews, which answer shopping questions from across the open web.

• Retailer shopping assistants such as Alexa for Shopping and Walmart Sparky, which answer inside a retailer's own store.

• Personal AI agents such as Meta Muse, which act on a shopper's behalf, researching, comparing and completing tasks for them.

GEO was built for the first type, where shoppers research. ACO covers all three, including the two where the purchase actually happens.

GEO misses the sale. Mentions and citations are where GEO reporting ends. ACO keeps going, into the metrics a commerce business already runs on:

• SKU-level discoverability: whether each product appears when shoppers ask an agent category questions, and how highly it ranks against competitors' products, a measure Azoma calls Product AI Rank.

• Revenue: incremental sales linked to AI visibility, measured by comparing optimized SKUs with a control group left untouched.

• Conversion: whether shoppers who meet a product through an agent go on to buy it, from glance views to purchase on Amazon, item views to purchase on Walmart, and AI referral traffic to transactions on brand.com.

Azoma is an agentic commerce optimization platform, and those are the three numbers it is built to move.

How does Agentic Commerce Optimisation (ACO) work?

Agentic Commerce Optimisation (ACO) works on what an AI shopping agent reads before it recommends anything: the product data behind each listing, and the outside sources it trusts to back that data up. Every agent reads a different mix, so the first job is knowing where each one looks.

Where do AI shopping agents get their answers?

Answer engines and retailer shopping assistants build their answers from very different sources. Optimize for answer engines alone and the point of sale goes uncovered. Azoma classified the websites behind more than 38 million citations across five agents to see exactly where.

Largest source category for each AI shopping agent from the Azoma dashboard, September 2026 (Walmart Sparky: August 2026)

Answer engines spread their trust widely. Gemini leans on editorial publishers, Google AI Overviews on retailer sites, and ChatGPT on institutional sources, with brand websites taking over a fifth of ChatGPT's citations (21.5%).

Retailer shopping assistants play by different rules. Walmart Sparky cites retailer pages and brand websites most (34.9% and 21.6%). Alexa for Shopping, previously branded Rufus, looks almost entirely outside retail, Amazon included, and gives brand websites under 1% of its citations.

And the mix keeps moving. Social and community content fell from roughly a fifth of ChatGPT's citations to 0.5% in the three months to September 2026.

Visibility is won agent by agent and product by product, and a strategy built for one type of agent leaves the others to chance.

How do brands improve ACO?

Brands improve ACO with the 5Cs of Agentic Commerce, the framework Azoma co-authored with the Digital Shelf Institute. The five work as a flywheel: Completeness makes a product eligible, Context makes it relevant, Citations earn it the sources AI trusts, Correctness keeps the answer accurate, and together they drive Customer Acquisition.

Each C has to be worked agent by agent, because every agent takes in product data through its own systems, fields and diagnostics. The example below follows one shopper question through all five: "What's the best mineral sunscreen for sensitive skin?"

1. Completeness: Is the product eligible to appear?

Agents decide which products qualify for an answer from structured product data: what a product is, who it is for, what it is made of and how it is packed. A single SKU on a global retailer can need more than 600 attribute values to count as complete, most of them backend fields shoppers never see. Copy and imagery decide which eligible product wins; attributes decide which products are eligible at all.

For the sunscreen, that means sun protection factor (SPF), active minerals such as zinc oxide, fragrance-free status, skin type, water resistance, size and variant structure, filled in as attributes on every surface the brand sells through.

On each agent:

• Alexa for Shopping: The Amazon Standard Identification Number (ASIN) catalog data behind the product page, including product type, browse node, Search Terms, unit count, ingredients, claims and variation structure. Amazon's catalog quality diagnostics in Vendor Central or Seller Central show where fields are missing.

• Walmart Sparky: Walmart item setup and attribution, including category, product type, ingredients, claims, use case and variant group. Walmart's Listing Quality Score flags the content and discoverability gaps.

• Gemini: The Google Merchant Center feed, the main input to Google's Shopping Graph, plus brand.com product pages that are crawlable and marked up with product schema.

• ChatGPT: Product feeds, retailer and brand.com product pages, and pages that are open to OpenAI's crawlers, such as GPTBot and OAI-SearchBot.

• Meta Muse: Meta's personal AI agent, launched on September 8, 2026, can buy through connected merchants such as Shopify stores, while Amazon has blocked it from buying on Amazon.com. Complete product data on brand.com and Shopify storefronts is the starting point.

Completeness is never finished. Retailer requirements change, and every reformulation, new pack or new claim reopens it.

2. Context: Is the product the relevant answer?

Agents read listings for meaning, checking whether a listing answers what shoppers actually want to know in the category. Context is the work of finding every question, comparison and consideration a shopper might raise, and answering each one explicitly with verified product information.

It is the one C that lifts AI visibility and on-page conversion at the same time. Agents organize shopper intent into relations such as who a product is used by, what it is used for and where it is used, so for the sunscreen that means plain statements such as "suitable for sensitive and reactive skin," "wears well under makeup" and "water resistant for swimming."

On each agent, the shopper questions come from different places:

• Alexa for Shopping: The questions come from the suggested questions shown on the product page, plus Amazon Brand Analytics and Search Query Performance. Each one gets answered somewhere in the listing: the title and bullets for the top questions, A+ Content, images and video for benefits and use cases, Q&A for recurring questions Alexa for Shopping answers incompletely, backend attributes for audience and use case, and review themes for what buyers care about after purchase. If shoppers keep asking whether the sunscreen leaves a white cast, the answer belongs in a bullet and in Q&A, and because answers are personalized to each shopper's profile, the listing has to qualify for every segment asking.

• Walmart Sparky: Sparky shows no suggested questions on product pages today, so the signals are Q&A, reviews, Walmart's AI-generated review summaries and Listing Quality tools. The answers go into the item name and key features, the product description, rich media, Q&A, item attributes and variant setup.

• Gemini: Search Console queries filtered for who, what, why and how, People Also Ask boxes, and Reddit and Quora threads. The answers go into brand.com product pages, FAQs and product schema, and into the Google Merchant Center feed, where the title, description, product type and category-relevant attributes should answer the top questions directly. For the sunscreen, that means a feed title and description that say "mineral," "sensitive skin" and "fragrance-free" in plain words.

• ChatGPT: The same signals as Gemini, plus Bing Webmaster Tools, since OpenAI's web layer has historically drawn on Bing's index. The answers go into retailer product pages and Q&A, brand.com content and earned media. Google Shopping data also appears to shape ChatGPT's product answers, so the Merchant Center feed matters here too.

• Meta Muse: Muse works from each shopper's own preferences and compares products across services, so its questions are personal and specific. Listings need to say plainly who a product suits and why.

3. Citations: Do trusted sources back the product up?

Citations are the sources an agent reads to build its answer: earned media, retailer pages, user-generated content such as Reddit, and brand.com. The mix shifts by agent and by category, so a citation plan rarely travels

On each agent:

• Alexa for Shopping: Editorial and affiliate publishers supplied 87.6% of citations in September 2026, so a place in "best mineral sunscreen" roundups matters most here.

• Walmart Sparky and Google AI Overviews: Retailer pages lead.

• Gemini: Editorial publishers lead, at 51.9% of citations in September 2026.

• ChatGPT: Brand.com earns over a fifth of citations, and institutional sources such as dermatology bodies carry weight. When ChatGPT's shopping module appears, which Azoma's data put at about 14% of queries (June 2026), its buy links become the highest-intent citation of all.

• Meta Muse: Muse launched with no paid placements or brand controls, so what it recommends rests on organic sources.

That puts public relations (PR) and communications teams at the center of ACO. They already hold the publisher relationships that earn citations.

4. Correctness: Does AI describe the product accurately?

Correctness is the work of catching and fixing what agents get wrong. Agents predict their answers, and when a product is new, reformulated or thinly covered online, they fill the gap with whatever sounds plausible. It happens even when every listing is accurate, and the only way to catch it is to query agents systematically against a verified source of truth.

For the sunscreen, an agent might state the wrong SPF, list a chemical filter the formula does not contain, or call it unsuitable for sensitive skin. The fixes work at the source:

• Retailer Q&A: Answer or correct Amazon and retailer Q&A that may be feeding the error.

• Review responses: Respond to Walmart review themes that cause confusion about ingredients or suitability.

• Authoritative content: Publish the accurate claim across brand.com, FAQs and earned media, so the correct version is harder for the model to ignore.

• Evidence-path cleanup: Update or counter outdated third-party pages and listicles.

• Platform feedback: Flag high-risk errors, such as safety or ingredient claims, then re-run the prompts to confirm the fix.

Health, safety and regulatory claims come first. Each agent draws differently on trained knowledge and live web content, so the same product can be described correctly by one agent and wrongly by another. Meta Muse can complete a purchase without the shopper ever opening a product page, so an error there reaches the sale with nothing on the listing to correct it.

5. Customer Acquisition: Is visibility turning into sales?

Customer Acquisition is where the first four Cs show up in the numbers. Two signals come from the agents themselves, Share of Voice and Product AI Rank, and both belong alongside traffic, conversion and incremental revenue.

On each agent, the commercial data sits in different systems:

• Alexa for Shopping: Vendor or Seller Central sales and glance views, Brand Analytics and Search Query Performance, Amazon Ads and Amazon Marketing Cloud.

• Walmart Sparky: Scintilla, Retail Link, Seller Center and Walmart Connect for item views, conversion and sales.

• ChatGPT and Gemini: Google Analytics 4 (GA4) AI referral traffic, Google Search Console, Merchant Center and retailer click-outs.

• Meta Muse and other personal AI agents: Muse can complete a transaction without the shopper visiting a product page, so the recommendation inside the agent is the sale. Checkout protocols such as Google's Universal Commerce Protocol, OpenAI's Agentic Commerce Protocol and Google's Agent Payments Protocol let agents buy, and each depends on complete, accurate product data behind it.

For the sunscreen brand, the practical starting point is a control group: optimize half the sun care range, leave the rest, and compare Share of Voice, Product AI Rank, traffic and conversion across both. Logging every change with its go-live date lets the team link movement in sales back to specific actions.

Run together, the 5Cs move all three ACO metrics: Completeness and Context make each SKU discoverable, Citations and Correctness make agents recommend it, and Customer Acquisition ties that to sales and conversion. Azoma runs this work on every agent a brand sells through, so teams can see exactly which C is holding each product back.

What should commerce brands look for in an ACO platform?

Look for a platform that tracks every SKU on every AI shopping agent, shows why each product is or isn't recommended, and ties that to sales. Five checks:

1. All three types of agent are covered. That means retailer shopping assistants such as Alexa for Shopping and Walmart Sparky, answer engines such as ChatGPT and Gemini, and personal AI agents such as Meta Muse. A platform built for answer engines alone misses the agents where products are bought.

2. Results are tracked per SKU and per agent. Brand-level share of voice hides which products are winning and which are losing.

3. Every answer shows its sources. The citation list tells a brand exactly where to act.

4. Visibility connects to sales and conversion. Product AI Rank and discoverability should sit alongside retailer sales, traffic and conversion data.

5. The platform helps improve results. Measurement shows the gap, and work across the 5Cs closes it.

Azoma is built to meet all five. It tracks every SKU on each AI shopping agent separately, shows the citations behind every answer, and works with brands through each of the 5Cs to turn recommendations into sales. 

For a board asking for an AI strategy, the first step is a diagnosis. Run real shopping questions through the agents customers buy from, and find which of the 5Cs is holding the products back.

Summing Up

AEO and GEO earn a brand its place in the conversation. On their own, they stop short of the decision that matters: which product an AI shopping agent recommends, and whether the shopper buys it.

That decision is made product by product and agent by agent, from structured data, trusted sources and listings that answer real questions. ACO helps achieve this across the retailer shopping assistants where products are sold, the answer engines where shoppers research and the personal agents that buy on their behalf. Its measure is what moves a business: SKU-level discoverability, sales and conversion.

Azoma is built for exactly this. We track how AI shopping agents describe and recommend products across Alexa for Shopping, Walmart Sparky, ChatGPT, Gemini and Meta Muse, re-running questions daily and recording the mentions, citations, rank, sentiment, buy links and carousel placements in every answer. Then we fix the product data, content and sources those answers are built from. 

Get in touch to see where your products stand.