Will AI Assistants Recommend Your Magento Store? A Practical Guide to AEO

More buyers are starting with a question to ChatGPT, Perplexity or Google’s AI overview instead of a list of blue links. Those tools do not show ten results, they give one answer. Answer engine optimisation, AEO, is the work of making sure your Magento store is in that answer. It builds on SEO, it does not replace it.

Most advice on this topic is guesswork dressed up as strategy. So this guide sticks to what the vendors actually document and what independent testing actually found, including the parts that are inconvenient.

First, the thing most AEO advice gets wrong

The common claim is that AI assistants read your schema.org markup. No vendor documents that, and one vendor documents the opposite.

Google states plainly in its guidance on AI features that you do not need to create new machine-readable files or markup to appear in AI Overviews and AI Mode, and that there is no special schema.org structured data you need to add. That is Google, about Google’s own AI surfaces.

OpenAI’s crawler documentation does not mention structured data, schema.org or JSON-LD anywhere. Neither does Perplexity’s. Neither does Anthropic’s. All three document one thing only: which crawlers exist and how to allow or block them.

OpenAI’s shopping path is not a crawl at all. It is a product feed, uploaded over SFTP or API, currently open to approved partners who apply through their merchant programme. There is no markup you can add to your Magento store that gets you into it.

The two platforms that do document reading your on-page markup for commerce are Google and Bing, both of which are search engines with shopping surfaces. So structured data is not an AI ranking lever. It is still worth doing, for reasons that are better than the myth.

What structured data is actually for

Product and Offer markup makes you eligible for rich results in Search, and it feeds Google Merchant Center. Those are real, documented, measurable returns. Five things to get right on a Magento store:

• Know which Product markup you need. Product snippets are for pages where people cannot buy directly. Merchant listings are for pages where they can, which is every product page on your store. Merchant listings require an Offer rather than an AggregateOffer, and they require a price greater than zero.
• Feed automatic item updates. Google documents that price, priceCurrency, availability and condition in your markup are required for automatic item updates, which is what stops Merchant Center disapproving items when your feed drifts out of sync with your site. This is the single highest-value piece of markup on a Magento store and it is routinely half-implemented.
• Do not duplicate shipping and returns. Google applies a precedence order, and settings in Merchant Center or Search Console beat your on-page markup. If those settings exist, marking up shipping on forty thousand product pages is page weight for nothing. Google’s own recommendation is a single organisation-level policy unless individual products genuinely differ.
• Get reviews right. You may mark up customer reviews of the products you sell. You may not mark up reviews of your own business under Organization or LocalBusiness and expect stars, because Google’s self-serving review policy makes that ineligible. The distinction trips up a lot of stores running a reviews widget site-wide.
• Delete your FAQ and HowTo markup. Both rich results are gone. HowTo was removed in 2023. Google deprecated the FAQ rich result in May 2026 and removed the documentation entirely a month later. The markup now does nothing. The FAQ content on your page is still useful, the schema wrapper around it is not.

The thing that actually decides whether you get read

It is not markup. It is whether your content exists in the HTML before JavaScript runs.

Independent testing has been consistent on this for over eighteen months, across different methods. A December 2024 analysis of server logs across a large hosting network found that none of the major AI crawlers execute JavaScript, covering OpenAI, Anthropic, Perplexity, Meta and ByteDance. They fetch JavaScript files. They do not run them.

A controlled test published in June 2026 went further. Twelve assistants were each given a page where the raw HTML contained a decoy value and the real value could only be obtained by executing a script. ChatGPT, Perplexity and the other major western assistants all reported the decoy. One fetched the script through a third-party service and still never executed it. The assistants’ own claims about their capabilities contradicted the server logs, which is worth remembering the next time you ask one what it can do.

The freshest finding is that the split is now geographic. In that same test, Mistral and the China-based assistants did execute the JavaScript and reported the real value. Every major American assistant read raw HTML only.

Google is the exception, and for a structural reason. AI Overviews and AI Mode are built on the Search index, and Googlebot renders pages with an evergreen version of Chromium. To be eligible as a supporting link in an AI Overview, Google says a page must be indexed and eligible to appear in Search with a snippet. So Google’s AI surfaces inherit rendering. The standalone AI crawlers do not.

What that means for a Magento store

Mostly good news. Magento renders product pages on the server, so your title, description, specifications and body copy are in the HTML before any script runs. A Hyvä storefront is better still, because there is far less JavaScript in the way and the markup is lighter to parse.

The risk is everything your store fills in afterwards. Prices that update when a customer picks a variant. Stock status pulled from a third-party availability widget. Reviews loaded from a remote script. Specification tabs that fetch their contents on click. Anything delivered through Magento’s private content mechanism. To an AI crawler, all of that is missing.

The test takes two minutes and you can run it yourself. Fetch one of your product pages with curl, or open it in a browser with JavaScript disabled, and read what comes back. If the price, the availability and the key specifications are not in that raw HTML, then as far as every major AI assistant is concerned, your product does not have a price.

That is also the honest case for a fast, server-rendered frontend. It is the same argument as migrating to Hyvä: less client-side machinery between your content and whatever is reading it.

Check your robots.txt before you do anything else

This is the one control every vendor does document, and it is where stores accidentally opt out.

OpenAI states that sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers. Perplexity recommends allowing PerplexityBot for search visibility. Both are separate from the training crawlers, GPTBot and ClaudeBot, which you may well want to treat differently. Anthropic’s three crawlers are independently controllable, so blocking the training one does not block the search one.

Google-Extended is a different thing again. It governs whether your content may be used for training and grounding in Gemini, and Google is explicit that it does not affect inclusion in Search and is not a ranking signal.

A surprising number of stores added a blanket AI block during the 2024 panic, never revisited it, and are now invisible in exactly the answers they want to appear in. Read the file. Decide deliberately which crawlers you want, rather than inheriting a decision somebody made in a hurry.

About llms.txt

You will be told to add one. Google has stated that these files are not needed for Search and will neither help nor hurt your visibility or rankings, and that it is fine to keep one for other systems that use them. No vendor documents consuming it for ranking.

It costs nothing and it is harmless. It is not a strategy, and anyone selling it as one is selling you the easy part.

Write content that answers the actual question

The FAQ markup is dead. The reason it existed is not. Answer engines pull from pages that state facts plainly: what a product is for, what it costs, who it suits, how it compares, what its limits are.

Buried marketing copy does not survive summarisation. Specifications written as a table survive. A direct comparison with the obvious alternative survives. An honest statement of who a product is not for survives, and does disproportionately well, because it is the kind of sentence nobody else writes.

Put all of it in the HTML body. A brilliant answer inside a tab that loads on click is an answer no assistant will ever see.

The trust signals the models weigh

These systems lean on the same credibility cues Google formalised as experience, expertise, authority and trust. Named authors with real credentials. A real company behind the store, with an address and a registration number. Details that match across your site, your Business Profile and the places you are mentioned. Genuine reviews rather than a widget of five-star quotes with no names.

Google’s Organization markup has quietly expanded to support exactly this, including company identifiers, VAT numbers and tax identifiers, specifically so it can tell you apart from other organisations. Thin, anonymous stores are what an assistant hedges away from when it makes a recommendation.

Set your expectations honestly

AEO is about being in the answer. It is not a traffic channel that replaces search, and the published numbers make that clear. Network-wide measurements from 2025 put the ratio of pages crawled to visitors referred at roughly fifty thousand to one for Anthropic, around nine hundred to one for OpenAI, and about a hundred to one for Perplexity.

So the volume is not there yet, and anyone promising you otherwise is guessing. What is there is the recommendation itself: being the store an assistant names when someone asks which one to buy from. That is worth having, and the work to get it is the same work that makes you legible to Google, accurate in Merchant Center and fast for customers.

Our own site already surfaces for evaluation-style queries about e-commerce platforms, the kind of prompt a buyer types into an assistant. The stores that structure their data and content now are the ones those answers will keep citing.

Book a free strategy call

Want to know whether an AI assistant can read and recommend your store today? We will check what is in your raw HTML before JavaScript runs, audit your Product and Offer markup against what Google actually requires, and read your robots.txt to see which assistants you have accidentally locked out. Book a free strategy call.