AI Referrals Up 197% and Converting 2x Better Than Organic: Inside Shopify's Q2 Commerce Data

Q: Should I move budget from SEO to GEO?

Not wholesale. Organic search still delivers more sessions than every tracked AI platform combined. Put incremental budget into product-data engineering, which feeds both sides. Google's own guidance says AI Overviews shares the same core ranking systems as regular Search.

AI Did Not Eat Search. It Ate the Decision.

On 11 August 2026, Shopify published its Q2 commerce data report, "AI and organic search are doing different jobs," and on 13 August Search Engine Land carried it into the industry press. Two numbers stand out: AI-referred sessions grew 197% year over year, with orders rising roughly threefold. In spec-led categories, AI-referred visitors converted at about twice the organic rate. This is a rare first-party dataset that measures AI traffic by completed transactions rather than citation counts.

For 18 months the dominant narrative has been that AI is eating search: rising zero-click rates, falling organic clicks, publisher traffic cut in half. But those studies all measure the same thing — clicks to content sites. Shopify measured transactions. In the same quarter organic sessions grew 12%, which looks unremarkable until you account for the base: organic still delivered more sessions than every tracked AI platform combined. Shopify CTO Mikhail Parakhin named the gap on X — everyone, himself included, assumed AI was killing organic discovery, but the numbers say the pie itself got bigger. Two channels growing at rates 16x apart, at the same time, does not fit a substitution model.

That also explains why GEO studies keep contradicting each other. Tools like Profound and Bluefish measure citation rate and visibility, media research measures click loss, and Shopify measured orders. Different instruments, opposite conclusions: by visibility, AI is a zero-sum predator; by revenue, it is a new high-intent channel. And Shopify itself warns that Google AI Overviews traffic is usually bucketed as organic, so whatever AI share you see today is an undercount.

For small merchants in Taiwan, the question is not whether to "do AI optimization." It is what your product data looks like right now. The rest of this piece covers how to read the numbers, what each role can act on today, and the landing-point bias big enough to discount that 2x headline.

The Full Numbers: What Shopify Actually Measured

Combining the report with Search Engine Land's summary, the usable facts are:

AI-referred sessions grew 197% year over year (roughly 3x), with AI-referred orders growing at a similar rate.
Organic sessions grew 12% from a far larger base, and organic still delivered more sessions than all tracked AI platforms combined.
Once they reached a product detail page (PDP), AI-referred visitors converted about 80% better than organic visitors.
In spec-led categories — specifications, compatibility, reviews, trade-offs — AI visitors converted at roughly 2x the organic rate.
In taste-led categories, organic remains the primary discovery channel, but AI brought about 1.3x more brand-new first-time buyers.
Category spread is wide: apparel around 1.6x, watches around 2.4x, necklaces around 2.3x.
50% of AI-referred sessions in Q2 landed directly on a product page.
When AI consumed structured Shopify Catalog data rather than scraped content or third-party feeds, the visitors it referred converted 2x better.
Shopify disclosed neither merchant count nor transaction count, nor how an AI referral was classified.
For context, Google's official AI optimization guide states that AI Overviews and AI Mode run on the same core quality and ranking systems as regular Search.

Three Roles, Three Sets of Moves

Brand owners and SMB operators

Audit your top 20 SKUs: are specs machine-readable text fields, or baked into an image? An image means the data does not exist.
Break AI referrals out as their own revenue line instead of leaving them in "other." You cannot manage what you cannot see.
Do not cut the organic budget. It is still the larger source, and the two channels are doing different jobs.

Marketing and SEO practitioners

Build a GA4 custom channel group isolating chatgpt.com, perplexity.ai, gemini.google.com, claude.ai and copilot.microsoft.com.
AI Overviews traffic is folded into organic, so your AI share is a floor, not a ceiling. Label it that way in reporting.
Segment conversion by landing type — PDP versus homepage and category pages — to see how much of the 2x is a landing-point effect.

Developers and agencies

Convert specs from free text into structured fields with types, units and enumerated values, so they are queryable and comparable.
Complete Product, Offer and AggregateRating markup, and keep price and stock identical to what renders. Mismatches get your trust downgraded.
Push referrer classification into the data layer and backend events rather than relying on GA4 UI config.

Four Routes to Making AI Understand Your Products

Option	What it actually does	Cost tier	Best fit
Shopify Agentic Plan	Structures Catalog data and syncs it to AI platforms through an official pipe instead of leaving it to scraping	Bundled with plan, mid to high	Shopify merchants with many spec-led SKUs
DIY Product schema plus feed	You maintain JSON-LD, spec fields and the product feed yourself; platform neutral	One-off build, low upkeep	Self-hosted, WooCommerce, Laravel commerce
Third-party GEO monitoring (Profound, Bluefish)	Tracks mention rate and visibility across AI assistants; measures exposure, not orders	Monthly subscription, mid to high	Teams with brand budget and reporting duties
Google Merchant Center conversational attributes	Enriches an existing feed with granular attributes for Search and AI surfaces	Free, labour only	Merchants already running Google Shopping

Three Things Nobody Will Tell You

One: the 2x claim carries heavy landing-point bias. Half of AI sessions land straight on a product page, while organic piles onto homepages, category pages and blog posts. Comparing people already at the checkout door with people who just walked into the mall is a rigged race. Shopify also published the more conservative figure — 80% better once the visitor reaches a PDP. Both numbers coexisting is an admission that the bias exists.

Two: the sample is opaque and the publisher has skin in the game. No merchant count, no transaction count, while Shopify sells an Agentic Plan built on syncing catalogs to AI platforms. That does not make the numbers false, but it does mean they are not peer-reviewed evidence.

Three: "structured data converts 2x better" is probably selection bias. Merchants who keep a clean catalog usually also keep accurate inventory, fast support and clear return policies. You may be measuring operational discipline, not field-level magic.

A DIY Roadmap With No SaaS Subscription

You do not need a subscription to start. Ordered by return on effort, a first pass takes two to three weeks:

☐ Complete Product, Offer and AggregateRating markup, consistent with what the page displays
☐ Move specs out of images and PDFs: dimensions, material, compatible models, voltage and warranty each get a field
☐ Add compatibility, use-case, returns and FAQ blocks to the PDP
☐ Build a GA4 custom channel group for AI referrers and backfill a 90-day comparison
☐ Add review snippet markup so ratings become citable structured facts
☐ Monthly, take 10 hero SKUs and ask a buying-comparison question in ChatGPT, Gemini and Perplexity; log whether you are mentioned and whether the specs are stated correctly

FAQ

Should I move budget from SEO to GEO?

Not wholesale. Organic still delivers more sessions than every tracked AI platform combined. Put incremental budget into product-data engineering, which feeds both sides; Google's own guidance says AI Overviews shares the same core ranking systems as regular Search.

I am not on Shopify. Does this still apply?

The direction applies, the multiples may not. The mechanism is that AI assistants prefer structured, field-level, comparable product data, and that is platform-agnostic. WooCommerce, Magento or a custom Laravel store can get there — you just maintain the schema and feed yourself.

How do I separate AI referrals in GA4?

A custom channel group covering the major AI assistant domains gets you most of the way, but AI Overviews traffic counts as organic and some assistants rewrite the referrer. What you get is a floor, so trust the trend over the absolute value.

My Take

The argument everyone is having — whether AI really converts better — is the wrong argument. That number is contaminated by landing-point bias, the publisher has a commercial interest, and no third-party replication is coming soon. The real signal is elsewhere: comparison behaviour in spec-led categories is migrating off your site into chat interfaces, and half of AI sessions land straight on a product page. Users arrive with the decision already made, and your site handles only the last mile — checkout.

If that holds, over the next 12 to 18 months the value of an SMB storefront shifts from persuading to being described correctly. Copy and layout lose marginal return, the accuracy and structure of product data become decisive, and part of the marketing budget migrates from content production to data engineering. Bad news for teams that only write copy, an opening for teams that can touch the database and the frontend in the same sprint.

ScriptWalker is packaging this as a priced engagement: product data structuring, spec field modelling, deep-funnel PDP rebuilds (compatibility, use case, returns, FAQ), plus GA4 AI channel grouping with a 90-day before-and-after comparison. For a store with 300 to 500 SKUs, a first pass usually runs four to six weeks and leaves a maintainable data model behind.