Search "generative engine optimization ecommerce" and the first page is a list of schema types. Product schema, Offer schema, FAQPage schema, AggregateRating schema, add them all and AI systems will supposedly start citing your store. Several of those guides cite a specific multiplier for how much more often a schema-marked page gets picked up.
Google's own documentation on AI features says something different. There is no special schema.org markup required to appear in AI Overviews or AI Mode. None. The requirement is that a page be indexed and eligible to show with a snippet in regular Search, the same bar that has applied to ranking for years.
That gap between what the platforms document and what gets sold as generative engine optimization advice is the actual story. Not every tactic in a GEO guide is wrong, but a lot of the ones presented as mandatory are closer to a traditional SEO checklist with a new name attached, and a few of the real requirements get skipped entirely because they are less exciting to write about than a schema shopping list.
- Google's AI features documentation requires indexing and snippet eligibility, not special markup, for AI Overviews and AI Mode.
- OpenAI runs two separate crawlers with two separate jobs, GPTBot for model training and OAI-SearchBot for ChatGPT search citations, controlled independently through robots.txt.
- The most common reason an otherwise complete product catalog does not get cited is a crawlability gap, usually JavaScript-rendered content a crawler cannot read, not a missing schema tag.
- Structured data is still worth doing. It just is not the gate most generative engine optimization content claims it is.
What generative engine optimization actually changes for a catalog
Traditional SEO for ecommerce optimizes a product or category page to rank on a results page a shopper then clicks through from. Generative engine optimization optimizes for a different outcome: being the source an AI system pulls into a generated answer, a comparison, or a direct recommendation, sometimes with a citation link and sometimes without one at all.
That distinction matters because the two outcomes get measured differently and, more importantly, fixed differently. A ranking problem is usually a content or backlink problem. A citation problem is, more often than most GEO advice admits, an indexing and rendering problem that sits underneath content quality entirely. We've covered the measurement side of this before: once a shopper does click through from an AI system, GA4 can show that session, but it cannot tell you why your competitor got cited and you did not. That answer lives upstream, in what the crawler could actually read.
What the platforms' own documentation requires
Three AI systems, three different and specific requirements, none of which is "add more schema."
Google AI Overviews and AI Mode: indexed and snippet-eligible, nothing special
Google's documentation on AI features and your website states it directly: "You don't need to create new machine readable files, AI text files, or markup to appear in these features. There's also no special schema.org structured data that you need to add." The actual requirement is that "a page must be indexed and eligible to be shown in Google Search with a snippet," which means the same technical SEO fundamentals that have always mattered, crawlability, a clean robots.txt, real textual content, acceptable page experience, are what AI Overviews and AI Mode inherit, not a new and separate checklist.
OpenAI's two bots do two different jobs
OpenAI's documentation on its crawlers splits the work into two agents that most "allow all AI bots" robots.txt advice treats as one. GPTBot is "used to crawl content that may be used in training our generative AI foundation models." OAI-SearchBot is the separate crawler that "is used to surface websites in search results in ChatGPT's search features." The documentation is explicit that these settings are independent: a store can disallow GPTBot to opt out of training use while still allowing OAI-SearchBot so ChatGPT keeps citing and linking to its product pages in search answers. Block both under one blanket "AI" rule and a store loses ChatGPT citations it never meant to give up.
Perplexity crawls for citation, not training
Perplexity's documentation on its crawlers describes PerplexityBot as the agent responsible for discovering and indexing pages that Perplexity then cites with a source link in its answers, a different purpose from a training crawler. Perplexity's own guidance is to allow the bot specifically when the goal is to make product pages, documentation, or reference content discoverable and linkable inside its answers, which is the one AI system among the three that consistently attaches a visible source link back to the originating page.
The checklist that follows from what's actually required
Once the requirement is "indexed, crawlable, and snippet-eligible" rather than "add five schema types," the checklist looks more like a technical SEO review than a GEO strategy deck.
- Check what the crawlers see, not what the browser renders. Load a product page with JavaScript disabled, or view its rendered HTML through Google Search Console's URL Inspection tool. If price, availability, or review count only appear after client-side scripts run, a crawler that does not fully execute them sees an incomplete page, which blocks the snippet eligibility Google's documentation requires regardless of how good the content is.
- Set robots.txt per bot, not per vendor. Treat GPTBot, OAI-SearchBot, and PerplexityBot as three separate lines with three separate intents:
User-agent: GPTBot
Disallow: /
User-agent: OAI-SearchBot
Allow: /
User-agent: PerplexityBot
Allow: /
That configuration opts a catalog out of AI model training while keeping it eligible for citation in ChatGPT search and Perplexity answers, a distinction a single blanket rule cannot make. 3. Keep structured data for the job it is actually good at. Product, Offer, and AggregateRating schema still earn traditional rich results and give any system, human-built or AI, an unambiguous read of price and availability instead of a scraped guess. Add it for that reason, not because a guide claimed it is a citation requirement Google's own documentation does not list. 4. Write the content an AI model would actually quote. A direct, specific answer to a real buyer question, sized to standalone as a sentence or two, is what a generative system extracts and restates. A paragraph that requires three sentences of context before it makes sense rarely survives that extraction intact, independent of any markup around it. 5. Confirm the discovery file reality on your platform before paying for one. We've written about why llms.txt specifically is one of the more oversold items on this list: Shopify generates an equivalent file automatically, and the evidence that a standalone llms.txt file changes citation rates on its own is thin. Spend the budget on the four checks above first.
Where this breaks down at catalog scale
A single flagship product page is easy to check by hand. A catalog with thousands of SKUs across Shopify apps, a headless frontend, and a handful of legacy redirects is a different problem: the same JavaScript-rendering gap that blocks one product can be silently present across an entire category, and nobody finds out until a competitor starts showing up in AI answers that should have included both stores.
If your catalog fits on one platform with a handful of templates, the five checks above are something a developer can run through in an afternoon using Search Console and a browser's dev tools. If it spans multiple platforms, inconsistent templates, or a product feed that several different teams touch, that is the point where a dedicated AI readiness audit earns its cost faster than checking each template by hand, because the failure mode that matters is the one no single page view will show you: which parts of the catalog a crawler simply cannot read, at scale, right now. The same fixed-scope discipline behind Anlyto's analytics audits applies here: a short, evidence-based check against what the platforms actually document, not a strategy deck built around what sounds impressive in a sales call.
