Key takeaways
- Google does not require a separate technical optimization layer for AI Overviews or AI Mode: pages still need to be indexed and eligible to appear in Search.
- JSON-LD remains a strong technical SEO practice and AI-cited pages are almost three times more likely to contain it, but current evidence does not show that structured data itself causes higher citation rates.
- Organic rankings still matter, but their overlap with AI Overview citations has weakened considerably, and position-one CTR can fall sharply when an AI Overview appears.
- Getting discovered by AI still rewards familiar SEO and product-data fundamentals. Getting bought by an AI agent increasingly requires feeds, APIs, authenticated pricing, inventory, and checkout infrastructure.
There's a tidy story going around the ecommerce world right now: AI search (what the industry now calls Answer Engine Optimization, or AEO, and Generative Engine Optimization, or GEO) changed everything, and then, on closer inspection, it didn't change much at all. Old-school technical SEO is back in fashion, structured data is suddenly cool again, and the brands that spent the last decade doing the fundamentals well are the ones showing up inside ChatGPT and Google's AI Overviews.
It's a good story. It's also only half true.
The fundamentals of discovery really have held their ground. But a second, quieter shift is happening alongside them, one that has nothing to do with content or keywords at all. As AI moves from answering questions to completing purchases, brands are being asked to expose live pricing, inventory, and checkout logic through APIs and machine-readable feeds, a discipline that looks nothing like traditional SEO. Calling all of this "old SEO in a new coat" undersells how much engineering work now sits behind a simple product recommendation.
Here's the fair version of the story: some things are genuinely resurging. Others are brand new. And the line between the two runs right through the difference between getting discovered and getting bought.
What's actually holding true
To appear as a supporting link inside AI Overviews or AI Mode, a page simply needs to be indexed and eligible to show up in standard search results with a snippet. There's no separate technical bar to clear. The old prerequisites, crawlability, indexability, and a clean site structure, are still the price of entry.
Google has also pushed back against the idea that generative search requires an entirely new technical playbook, and explicitly says publishers do not need to rewrite pages specifically for AI systems. The message is consistent: established search fundamentals and useful, human-readable content still matter.
JSON-LD (JavaScript Object Notation for Linked Data) is now by far the most common structured-data format on the web: HTTP Archive’s 2025 data found it on 43% of home pages, compared with 16–17% for Microdata and about 1% for RDFa (Resource Description Framework in Attributes). AI-cited pages also over-index strongly for JSON-LD—in a 2026 Ahrefs analysis of 6 million URLs, cited pages were almost three times more likely to contain JSON-LD than non-cited pages. However, current evidence does not show that JSON-LD is itself a direct driver of AI citations. Even so, maintaining accurate JSON-LD remains a strong technical SEO practice because it gives search engines a standardized, machine-readable representation of page entities and content and supports eligible search features.
Shopify merchants are seeing an even more literal version of this pattern play out. Keeping product variants organized under coherent product structures has long been a familiar ecommerce and SEO practice. That same structure now also helps preserve clear relationships between products, variants, and their attributes when AI shopping systems interpret a catalog. Fragmenting variants across separate listings can also split reviews, ratings, and product information across multiple pages. A familiar ecommerce architecture is taking on a new role in 2026.
And then there’s the product feed. Once treated mainly as Google Shopping infrastructure, Merchant Center feeds now also help products surface in Google’s AI responses and other Search experiences. Keeping those feeds complete and current gives Google cleaner product data to work with, while structured, complete product information is becoming part of the intelligence layer AI systems use to understand and represent a brand. The role of the feed has expanded well beyond paid-media plumbing.
So yes, in a real sense, everything old really is new again. The technical bones of SEO, indexability, structured data, and clean feed hygiene, aren't just still relevant. They remain an important machine-readable foundation across traditional and AI-driven search experiences.
| SEO consideration | Old playbook | What's true now |
|---|---|---|
| Indexability and crawlability | Baseline requirement to rank | Still the baseline. Google confirms there's no separate technical bar for AI Overviews or AI Mode |
| Structured data (JSON-LD) | Nice-to-have for rich snippets | Present on 43% of home pages, and AI-cited pages are almost three times more likely to include it, though the evidence shows correlation, not proven causation |
| Organic ranking position | Strong predictor of visibility | Overlap with AI Overview citations has fallen to roughly 17%, so a page-one ranking no longer reliably predicts an AI citation |
| Position-one CTR | Reliable click driver | Can fall 58% when an AI Overview appears above the results |
| Product feeds (Merchant Center) | Paid Shopping infrastructure | Now also feeds AI Overviews and other AI search experiences directly, not just paid placements |
| Product and variant structure | Ecommerce and SEO organizational best practice | Also determines how AI shopping agents parse relationships between products, variants, and attributes |
| Transaction readiness (feeds, APIs, authenticated pricing, inventory, checkout eligibility) | Not part of SEO at all | New territory, governed by commerce protocols like OpenAI's product-feed spec and Anthropic's Model Context Protocol. An engineering and operations problem, not a content one |
Where the old playbook runs out
That's the reassuring half of the story. The uncomfortable half starts with a simple fact: the relationship between ranking well and getting cited by an AI system has weakened considerably.
In 2024, AI Overview citations showed much stronger overlap with pages already ranking in the organic top ten. By early 2026, one analysis cited by Omnibound put that overlap at roughly 17%. Different studies produce different overlap rates, but the direction is clear: ranking on page one is a less reliable proxy for AI visibility than it once was. AI citation selection is not simply an extension of the traditional organic ranking order.
The traffic math has shifted too. When an AI Overview appears above the results, even the top-ranked page can lose more than half its clicks: Ahrefs’ 2026 analysis of 300,000 keywords found position-one CTR was 58% lower when an AI Overview appeared. Seer found the same broader pattern across 5.47 million tracked queries and 2.43 billion organic impressions, with AI Overviews suppressing organic CTR particularly for informational searches. Ranking first still matters, but its value is increasingly split between earning the click and earning visibility inside the AI-generated answer.
All of this means the foundations aren't the finish line anymore. Crawlability, indexation, and good content may get a page into the candidate set, but they do not guarantee that a generative system will choose it as a source.
Discovery is old news, transaction is not.
The more consequential shift is happening below the content layer: in the systems that let an AI agent determine whether a product can be found, priced, and purchased.
ChatGPT
ChatGPT’s commerce infrastructure makes that explicit. OpenAI’s product-feed specification includes is_eligible_search and is_eligible_checkout controls governing whether a product can surface in ChatGPT search or support direct purchase. The same specification accepts optional signals such as product popularity and return rate, which OpenAI says can help improve ranking and highlight high-performing products. That is fundamentally different from optimizing a page for crawlability: product eligibility and commerce data now live partly in the feed and integration layer.
Perplexity
Perplexity is building toward the same outcome through different infrastructure. Its Instant Buy with PayPal experience lets eligible U.S. shoppers purchase from participating merchants without leaving the conversational flow. The optimization problem is no longer just whether an agent understands the product; it is whether the merchant’s systems can complete the transaction.
Anthropic
Anthropic’s Model Context Protocol creates a standard way for AI systems to connect directly to external data and tools. Shopify now exposes Storefront MCP capabilities for product discovery, carts, and checkout, alongside authenticated customer-account capabilities. Domaine has previously covered how Shopify’s MCP and UCP direction is pushing commerce beyond the storefront and toward standardized agent-to-commerce interactions.
That becomes especially important in B2B, where the correct product or price can depend on who is asking. Shopify’s B2B infrastructure supports company-specific catalogs and pricing. An agent serving those buyers may therefore need access to protected commerce systems, not simply a publicly crawlable product page.
None of this makes the storefront obsolete. ChatGPT still supports both merchant-site links and Instant Checkout for eligible products. What is changing is the technical surface brands have to optimize: discovery still matters, but increasingly so do feeds, APIs, authentication, pricing logic, inventory, and checkout infrastructure.
The honest verdict
Everything old is new again, but only for discovery. Crawlability, structured data, and product feeds are still what get a brand found by AI systems, and the mechanics haven't changed, only the stakes have. Getting bought is a different problem entirely: exposing live pricing, inventory, and checkout to an autonomous agent is a systems integration challenge that sits with engineering and operations, not content, and no amount of clean metadata substitutes for a working API.
There’s also a real caution worth naming. In EXL’s 2026 Enterprise AI Study, 70% of executives cited data as their biggest barrier to AI adoption, while Salesforce’s 2026 CIO research found data security, privacy, and lack of trusted data remain the biggest obstacles to scaling agents.
