AEO (AI Engine Optimization) — The New E-Commerce Strategy for Getting Chosen by AI

Published in April 2026 and updated in August to reflect what changed since: a controlled experiment on structured data, Google's official position, AI crawler behavior, and the shift in the purchase route.

How AEO (AI Engine Optimization) is changing e-commerce marketing

Consumers now get an answer before they ever see a results page. The shift is visible in the numbers.

Pew Research Center reported in July 2025 that when an AI Overview appeared, organic results were clicked in 8% of searches, against 15% when none was present — roughly half. Ahrefs measured the drop in click-through rate for the top result at 58% when an AI Overview is shown.

Japan shows the same pattern. Browser log analysis found the share of Google searches leading to a site visit fell to 41.1% in 2025. Generative AI usage reached 51% as of February 2026, nearly double the 27% a year earlier.

When a consumer asks "what are the best running shoes", AI presents three to five options. Brands outside that list are treated as though they do not exist.

This is where AEO (AI Engine Optimization) comes in. Where SEO aims to rank high in search engine results, AEO aims to be cited and recommended inside AI-generated answers. It is also called LLMO or GEO. Our overview of LLMO sorts out the naming and the full picture.

AEO optimizes for entering the answer, not for position

Here is the structural difference from traditional SEO.

The user behavior row is the one to watch. A search results page returns a list of links, so ranking fifth still left room for a click. An AI answer narrows to roughly three candidates, so missing that set means the opportunity never occurs.

The content authority built through traditional SEO still matters, but it alone does not get you into the recommendation set. In the age of agentic commerce, product data itself becomes the front line of marketing.

Can AI agents actually read your product pages?

Before structured data, there is something to check: whether your product information reaches AI at all.

The major AI crawlers — GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot — do not execute JavaScript and read only the HTML that comes back. Google's line is the exception, since it builds on the Google Search index and can use rendered information.

Which means on a store where price, stock and specifications are drawn by JavaScript alone, product information is not reaching AI before any question of schema arises. This happens on pages that rank well in Google. Checking takes seconds: disable JavaScript and open a product page.

We measured how common this is. In our audit of nine major electronics retail brands, Product and Offer microdata was confirmable on product pages for only one of the nine. In the provisional distribution, six brands were stuck before the comprehension stage — the largest group. That is the state of the market leaders.

Product data AI chooses is built from three elements

With the technical route open, the data comes next: structured data, semantic summaries, and reviews and trust signals.

Structured data: a common language with AI agents

AI agents do not see web pages the way people do. They parse structured data written in JSON-LD and understand product attributes mechanically.

This part needs stating precisely.

Three reasons to implement it anyway: rich results in search, verifying consistency with your Merchant Center feed, and preventing misreading. The third is where the AI connection lives. Handed only body text, AI infers whether a price includes tax and what stock means; definite values remove that room. Implementation is covered in our structured data guide.

Three schemas are the minimum for an online store. Product defines basic attributes such as name, brand, GTIN, images and description. Offer makes pricing, stock, shipping terms and return policy machine-readable. AggregateRating and Review structure review counts and rating scores.

What gets overlooked is the importance of identifiers. GTIN and MPN are the keys by which an AI agent identifies the same product across platforms. Missing them can leave you outside the comparison set entirely.

More important still is data consistency. Since Google's UCP launch, discrepancies between Merchant Center feed data and on-site structured data translate directly into lower trust. When price or stock disagree across channels, AI has no way to judge which is correct. Feed work is covered in our product feed guide.

Semantic summaries: turning specs into context

Structured data alone does not give AI enough to reason with. In AI-driven product discovery, responding to constraints rather than keywords is what decides the outcome.

Take a concrete case. A traditionally SEO-optimized description reads "waterproof lightweight outdoor jacket men's". But when a consumer asks "I'm traveling to Europe in April — is there a jacket that works in rain and fits in carry-on?", what AI uses is not a list of attributes but context tied to a use case.

"Handles a light-rain commute but is not designed for heavy downpours." "Folds to 30×20cm and fits the side pocket of a standard carry-on." That is the essence of a semantic summary. Product data needs organizing by the problems it solves rather than by category.

Three implementation points. State who the product is for. Describe usage in concrete scenarios. And state honestly who and what it is not suitable for. AI does not recommend universal products; it selects what fits a specific context, so well-chosen exclusions raise recommendation accuracy rather than lowering it.

Reviews and trust signals: the evidence AI needs

The last piece determining whether AI recommends a product is reviews and trust signals.

The review quality AI weighs differs from human assessment. Volume and recency dominate: five reviews from the past month carry more signal than twenty from six months ago.

The second factor is attribute-specific feedback. "Good product" does far less than "runs slightly small — if you usually wear M, take L. The material is soft and did not shrink after three washes", which becomes evidence when AI answers a specific question.

Automating review collection, and using post-purchase requests and smart prompts to elicit attribute-specific feedback, converts directly into advantage. Beyond that, placement on third-party surfaces such as comparison articles and review sites matters. There is a region you cannot reach by improving your own site.

Share of Model: measuring brand visibility inside AI

With implementation underway, how do you measure it? This is where Share of Model comes in.

Traditional digital marketing tracked Share of Voice — a brand's proportion of media exposure. Share of Model is its successor, measuring how often your brand is mentioned when AI answers questions about a category. The formula is your brand mentions divided by total category mentions, times 100.

In practice: design 20 to 50 prompts your prospective customers might use, run them across ChatGPT, Gemini, Claude and Perplexity, then record mention frequency, position, context and citation type, and compare against competitors.

Do not measure on one AI and generalize. In our washing machine study, the retailer named first split by service: ChatGPT and Gemini named one, Claude another, Perplexity and Copilot a third. Each is stable internally while disagreeing with the others.

Official instrumentation has also arrived. Google announced generative AI performance reports in Search Console on 3 June 2026 — the first official data separating AI Overviews and AI Mode impressions from ordinary organic search. Only impressions are available for now, with data beginning 18 May 2026.

In our nine-brand audit, no published information about AI-referred traffic or product feed integration was confirmable for any of the nine. It may simply be invisible from outside, but few companies are measuring this yet.

AEO implementation roadmap

Organized by execution priority.

Treat this as gradual reallocation from existing SEO and SEM spending rather than new budget. The high-priority items all fall within what internal resources can handle.

As a first step, audit your current product data. Structured data implementation can be verified through the rich results reports in Search Console, and Merchant Center alignment through feed diagnostics. Checking JavaScript dependency requires nothing more than disabling JavaScript in your browser.

Purchase has moved back to your own site

The purchase route shifted over the past year, so it is worth stating where things stand.

OpenAI launched Instant Checkout on 29 September 2025, allowing a purchase to complete inside ChatGPT. It ended in March 2026, with the focus returning to discovery and recommendation. Today AI recommends products and sends the purchase to the merchant's storefront or app.

Two implications. The first is that the division is now discovery through AI, purchase on your own site. You need both to enter the candidate set through AEO and to build an experience that sells once someone arrives. A recommendation landing on a confusing page is where the funnel breaks.

The second is that this is not the final shape. As protocols such as UCP mature, the purchase experience will move again. Which is why product data work now is not wasted: whatever the route, the quality demanded of the data is the same.

The difference from GEO, and AEO's limits

Treating AEO as a silver bullet is dangerous. As our analysis of GEO risks shows, AI model responses are unstable, with errors frequent in financial and governance-related areas.

In our study putting 100 question patterns about returns and warranties to five services, 60.0% of answers matched the official information. Source handling was worse, with 62 answers citing URLs that did not exist or belonged to another brand. Before increasing exposure, confirm that your official information can be copied correctly.

AEO is a way of giving AI accurate data and raising the probability of entering the recommendation set. Controlling AI answers is impossible in principle, and structured data is a necessary condition, not a sufficient one. Our article on how LLMO, GEO, AIO and AEO relate sorts out the terms.

Summary

As the main arena of search moves from ten blue links to an answer window, what e-commerce businesses need is a change of framing. Not abandoning SEO, but using SEO assets as the precondition.

And the order matters. Before implementing structured data, confirm that product information is reaching AI at all. Of nine major brands we measured, one had product data structured. Even at that scale the ground is open.

Our overview of LLMO gives the full map, and the ten steps for AI search give the order of work. We offer a free AI visibility assessment for retail and e-commerce businesses.