Shopify's Q2 2026 earnings call produced a large first-party dataset on what AI search does to traffic, and most of the coverage stopped at the reassuring part. TechCrunch led with AI search not replacing Google. PYMNTS led with shoppers skipping the search bar. The more useful finding came further down the call: half of all AI-referred sessions land directly on a product description page, and 75% of AI-attributed purchases came from outside Shopify's top 100 categories.
Those are not claims about AI search in general. They are claims about which kind of page survives it.
The same TechCrunch report that carried Shopify's good news noted that AI summaries have produced a measurable drop in click-through rates for online publishing, cutting traffic and the ad revenue that depends on it. Same technology, same window, opposite outcomes. The variable is not the platform. It is whether your page is closer to a database row or closer to an essay.
A note on sourcing. The financial figures below come from Shopify's press release and the performance table filed with it. Every Finkelstein quotation here is as reported by TechCrunch: the full earnings-call transcript was not publicly retrievable at the time of writing. PYMNTS, covering the same call separately, carries the same figures: tripled AI traffic and orders, the product-page landing rate and its 2.5x multiple, traditional search at 1.3x over two years holding roughly a third of storefront traffic, and the car seat example below. Where this post corrects TechCrunch on a number, it is checking arithmetic against the filing, not disputing the reporting.
The financials, taken from the filing rather than the coverage
Shopify's own August 5 press release reports 34% revenue growth (33% in constant currency) and an 18% free cash flow margin for the quarter ended June 30, 2026. The selected business performance table gives the raw figures: revenue of $3,583 million against $2,680 million a year earlier, GMV of $115,567 million against $87,837 million, gross profit of $1,708 million against $1,302 million, operating income of $488 million, and free cash flow of $654 million.
One correction worth making, because it is being repeated. TechCrunch reported revenue "rising 36% to $3.6 billion." Shopify's release says 34%, and $3,583m divided by $2,680m is 33.7%. PYMNTS reported 34% from the same call. The gross profit figure in TechCrunch's same paragraph, up 31% to $1.71 billion, matches the filing exactly. Use 34%.
A second point matters more than the arithmetic. None of the AI statistics appear in the press release. The tripling of AI-driven traffic and orders, the product-page landing rate, the top-100-categories number: all of it came from management commentary on the call. Note that the financial statements are not audited either. Shopify's release says its Form 10-Q for the quarter includes "unaudited Condensed Consolidated Financial Statements." So the distinction is not audit status. It is that the financials are filed with a regulator under defined accounting rules while the AI metrics are spoken aloud, self-defined, by a company whose stock was moving that morning. That does not make them false. It does mean they are evidence, not proof.
Half of AI sessions skip your homepage entirely
Here is the quote that should have led the coverage. "Buyers' shopping journeys are being compressed as half of all AI-referred sessions are landing directly on a product description page," Shopify President Harley Finkelstein told analysts, as reported by TechCrunch. "That is 2.5 times more than what we see with traditional search."
Read what that implies about the pages that are not getting the visit. The category page, the buying guide, the comparison listicle, the merchandised homepage: those exist to narrow a broad intent down to one item. When an AI referral arrives on a product page, that narrowing already happened inside the model. The intermediate layer was consumed, not clicked.
For a store, this is good news, because the destination page is the one that takes money. For a site whose entire business is the intermediate layer, it is the same mechanism pointed the other way. Pew Research's browsing-data study of Google AI Overviews found users clicked a traditional search result on 8% of visits where an AI summary appeared, against 15% where none did. Clicks on a link inside the summary itself happened on 1% of visits. Sessions ended outright on 26% of pages with a summary versus 16% without.
The compression Shopify is celebrating and the compression publishers are complaining about are the same event. Side by side:
| Shopify storefronts | Publisher pages in Google results | |
|---|---|---|
| Source | Q2 2026 earnings call, management commentary | Pew Research Center browsing-data study |
| Period and population | Quarter ended June 30, 2026, global storefront sessions | March 2025, 900 US adults sharing browsing activity |
| Effect on traffic | AI-driven traffic and orders tripled year over year | Result clicks on 8% of visits with an AI summary, 15% without |
| Where visitors land | Half of AI-referred sessions arrive on a product page, 2.5x the traditional-search rate | 1% of visits produce a click on a source cited in the summary |
| Effect on the session | Journey compressed toward purchase | Session ended on 26% of pages with a summary vs 16% without |
| Traditional search | Up 1.3x over two years, roughly a third of storefront sessions | Not measured over time, the study is a single-snapshot comparison |
| Page type | Structured records with queryable attributes | Prose an LLM can summarize in full |
| Verification status | First-party, self-defined attribution, not in the filing | Independent, published methodology |
The contrast is directional, not controlled. Different periods, different populations, different methods: one side reports year-over-year change, the other compares two kinds of page at a single moment. It is the best evidence available, and it is still two studies pointed at each other rather than one experiment.
Why a catalog wins where an article loses
Finkelstein gave the mechanism, and it is more precise than most vendor commentary on this subject: "While search engines rank by popularity against a handful of keywords, AI agents make multiple calls into Shopify's catalog, working with richer structured data to match products with the buyer's specific intent, rather than just keywords."
Two words do the work there. Multiple calls.
Keyword ranking is one query producing one ordered list, and clicks distribute along a power law from position one. Agent retrieval is many queries against a structured store, each carrying narrow constraints, with the result set assembled by filtering rather than ranking. His example: a buyer asks for the best car seat that fits three across a sedan. Keyword search sees "car seat." An agent sees the dimensions, the vehicle type, and the count, and searches across all of those constraints at once.
A product record with forty attributes can satisfy a five-constraint query. A 2,000-word essay cannot be queried on constraints at all. It can only be read and summarized, and a summary does not require the reader to click through. That is the asymmetry in one sentence: a catalog gives the model something it cannot generate, and a prose explainer gives it something it can.
This also explains why the connector list matters. Shopify has built integrations into Claude, ChatGPT, Perplexity, Manus, Replit and Vercel, plus vibe-coding platforms like Lovable. Merchants on the platform get queryable-by-agent as a default property of being on Shopify. An independent site has to build that surface itself.
AI search inverts how keyword SEO distributes traffic
The 75% figure is the most testable claim on the call, and also the thinnest-sourced. TechCrunch reports that 75% of AI-attributed purchases in Q2 happened outside Shopify's top 100 categories. That number is not in the press release, and the full transcript was not publicly retrievable, so it rests on call reporting rather than a document you can open. Weigh it accordingly.
TechCrunch's summary of Finkelstein's remarks is that AI has particularly benefited the long tail of e-commerce, including the smaller merchants that make up most of Shopify's customer base. The words he is directly quoted saying are narrower: AI has become a "complement to search, rather than a substitute for it."
Keyword SEO does the opposite of what that 75% implies. Volume follows a power law, the head terms carry the competition, and position one takes a disproportionate share of whatever clicks remain. If Shopify's distribution generalizes beyond commerce, it produces a falsifiable prediction about your own site: AI referrals should concentrate on your most specific, lowest-volume URLs, while Google organic stays concentrated on your broadest ones.
That is directly testable by anyone with analytics or server logs. Segment referrals by AI source, then compare the landing-page distribution against Google organic for the same period. A flatter, longer tail on the AI side supports the pattern. The same head pages on both sides refutes it, at least for you. Run it per source rather than as one lumped "AI" bucket, because the behavioural differences between Perplexity and ChatGPT show up in referral data.
We have not run that test at publishable scale on our own logs. Treat the prediction as a hypothesis with one strong data point behind it, not a finding.
What this data does not prove
Five limitations, in order of how much they should change your reading.
"AI-attributed" is never defined. Attribution for assistant referrals is genuinely difficult because some tools strip referrers, some route through their own redirectors, and some surface a brand without any link at all. Any AI traffic number is an undercount, an overcount, or both depending on method, and Shopify did not publish its method.
"Tripled year over year" starts from an undisclosed base. Tripling a small number is not hard, and the absolute contribution is unstated.
Shopify has privileged distribution. Being queryable by six or more major agents is a platform feature its merchants receive for free. That is not evidence that an ordinary site can earn the same visibility by adding schema markup.
The reporting party has an obvious interest. This was an earnings call, and the narrative was flattering.
And critically, search did not shrink. Finkelstein, as quoted by TechCrunch: "Traditional search sessions are up 1.3x over the past two years, holding roughly a third of all storefront sessions." Whatever AI added here was additive. Nobody has shown a substitution effect running in Shopify's favour.
What to change, and how to check whether it worked
The actionable reading of this data is a content-type verdict, so the actions are about content type.
Make your specifics machine-readable. Product, Offer, FAQPage and HowTo markup are not ranking tricks here. They are the surface an agent's constraint query can actually hit. A page whose facts exist only inside paragraphs is a page an agent can summarize but cannot filter on.
Put something on the page that a model cannot generate. Prices, dimensions, versions, dates, availability, measured results from tests you ran. A general explainer of a concept is the exact shape of content the summary layer absorbs. Our rundown of AI SEO tools for small sites covers the tooling side of this, and content optimizers like NeuronWriter can help with structure, but no tool substitutes for having original numbers.
Stop measuring top-of-funnel explainers on impressions. If the Pew pattern holds, impressions on that content type will keep rising while clicks fall, and the gap is not a title-tag problem.
The defensible summary is narrow: in one quarter, at one company, structured transactional inventory gained from AI search while prose in the same market lost clicks. If you run a site, the question is not whether AI search is good or bad. It is which of those two things your pages are.
FAQ
Does AI search actually send traffic to websites? It sends traffic to some kinds of pages. Shopify reported AI-driven traffic and orders to its merchants tripled year over year in Q2 2026, with half of AI-referred sessions landing on a product page. Pew Research found the opposite for publishers: people clicked a result on 8% of visits where a Google AI summary appeared, against 15% of visits where none did. Both can be true because the page types differ.
Why did Shopify gain from AI search while publishers lost? Because a product catalog answers a constrained query and an article answers a general one. Finkelstein described agents making multiple calls into structured data to match specific intent, rather than ranking against a handful of keywords. An assistant can fully summarize an explainer, removing the need to click. It cannot invent a real product with real dimensions, price and stock, so it hands the user off.
What does "half of AI-referred sessions land on a product description page" mean in practice? It means the browsing and shortlisting work happened inside the model instead of on the site. Shopify said this rate is 2.5 times what traditional search produces. The pages being skipped are the intermediate ones: category pages, buying guides, comparison listicles. For a store this compresses the path to purchase. For a site whose product is the buying guide, it removes the visit.
Is AI search killing SEO? The Shopify data argues against a blanket yes. Traditional search sessions to its merchants were up 1.3x over two years and still hold roughly a third of storefront sessions. What is changing is which pages get rewarded. Content that exists to summarize or explain is being absorbed by the summary layer. Content carrying specific, structured, verifiable detail is still getting the click.
How do I optimize a site for AI agents? Start by making facts queryable rather than only readable: schema markup for products, offers, FAQs and how-tos, plus explicit values for prices, dimensions, versions and dates in the page body. Then give the page something a model cannot generate, such as your own measurements or first-party data. This is a hypothesis worth testing on your own analytics, not a proven playbook.
Can I trust Shopify's AI numbers? Partly. The financial figures are filed with a regulator under defined accounting rules, though Shopify labels that filing's condensed statements unaudited as well. The AI statistics are neither filed nor defined. They came from management commentary on an earnings call, "AI-attributed" was never defined, and the base for "tripled" was not disclosed. Shopify also has connectors into major assistants, giving its merchants agent visibility that an independent site does not get by default.
Does this apply to sites that do not sell products? The mechanism should, though nobody has published data confirming it. The dividing line Shopify's numbers draw is between structured records an agent can filter and prose an agent can summarize. A directory, a specification database, a pricing table or a dataset sits on the winning side of that line. A general explainer article sits on the losing side regardless of industry.
How do I measure AI search traffic in my analytics? Create a channel group or segment for known assistant referrer domains and check it against Google organic for the same period, comparing which URLs each one lands on. Expect undercounting, because some assistants strip or rewrite referrers and some mention a brand without linking at all. Read the resulting number as a floor and a trend line, not an exact count.
Sources
- Shopify says AI search is driving more traffic and sales, not replacing Google (TechCrunch, August 5, 2026)
- Shopify's AI Traffic Triples as Shoppers Skip the Search Bar (PYMNTS, August 5, 2026)
- Shopify Delivers Big: 30%+ Growth Across GMV, Revenue, Gross Profit, and Free Cash Flow (Shopify press release, August 5, 2026)
- Full Q2 2026 press release with financial statements (GlobeNewswire)
- Google users are less likely to click on links when an AI summary appears in the results (Pew Research Center, July 22, 2025)