Loading today’s stories
Build1 distinct publisher3 min readPublished
The doubled conversion rate in spec-led categories arrives with imperfect attribution and an undisclosed sample size, but the structured-versus-scraped comparison buried inside it is the one a single merchant can actually reproduce.
The Engineer · Build desk
Compiled by The EngineerSomething wrong?How this is made
A conversion rate is a ratio, and the denominator here is not holding still. If half of one channel's sessions begin on the page with the buy button [7], and the other channel's sessions are distributed across whatever pages happen to rank, then a per-session comparison between the two is partly a measurement of landing mix. Shopify lists landing-page mix as one of the measures merchants should compare [19], which is an admission that it varies.
That is why the Q1 figure is the more useful one. It was restricted to product detail pages on both sides and came out at about 50% higher conversion for AI-referred sessions [2]. Entry point roughly fixed, one variable moving. The Q2 spec-led doubling [1] is a category cut rather than a landing-type cut, and Shopify says attribution remains imperfect and does not disclose the dataset size [9][10]. Shopify also allows that a high rate could reflect better-qualified visits or could reflect where the shopper entered the journey and how the referral got classified [16].
The structured-versus-scraped result is different in kind. Both arms are AI referrals [8]. Same channel, same attribution errors, same population of shoppers who asked an assistant a question first. What moves is whether the assistant was reading Shopify Catalog data or a less-structured scraped or third-party feed. As experiments in a vendor blog post go, that is nearly one.
Do not stack the two doublings. The spec-led 2x is measured against organic search [1]; the catalog 2x is measured against other AI referrals [8]. Different baselines, so they do not multiply into 4x, and nothing in the analysis says which fields carry the effect [11]. Shopify's own list of candidates runs through titles, attributes, specifications, variants and availability [15], and that list is per-SKU work.
The mechanism is easy to state even if the attribution is not. An assistant answering a compatibility question needs the answer in a field it can read, not in the third paragraph of a description. A scraped feed is your catalog with the awkward parts dropped, and Shopify's numbers price that loss at roughly half the conversion [8].
On breadth: 23 of 25 categories beat organic search, an average uplift near 56% [4]. That is 92% of the categories measured [12], and two did not clear the bar [13]. Shopify does not name the two. The spec-led cut sits at about 1.8 times the cross-category average uplift [17], which is consistent with the story that decision-support data is what the channel is actually consuming, and Shopify frames the direct-to-product-page share as journey compression on the same logic [14].
For the 2x to transfer to your store, three things would have to be true: your organic baseline is measured on product-page entries only, your AI bucket is not quietly absorbing no-referrer traffic you cannot classify, and your category is one where a buyer needs an attribute value to choose. The version of this test that would settle it for one merchant is within-channel and small. Take the SKUs whose specs currently live in prose, move them into attribute fields, leave comparable SKUs alone, and watch AI-referred product-page conversion on both sets. That measures your catalog rather than Shopify's aggregate [18].
Ranked by verification strength, evidence, and original report placement.
Shopify's Q2 2026 analysis found that shoppers referred by AI converted at roughly twice the rate of organic-search visitors in spec-led categories.
Shopify's Q1 2026 data found that AI-referred sessions on product detail pages converted at about 50% higher rates than organic-search-referred sessions.
Orders from AI-referred traffic in Shopify's Q1 2026 data had roughly 14% higher average order values.
At the category level, AI-referred conversion outperformed organic search in 23 of 25 merchant categories, with an average uplift of about 56%.
AI-referred sessions grew 197% year over year in Q2 2026, according to Shopify.
AI-referred orders increased by roughly three times over the same year-over-year period.
Distinct publishers with included, body-backed reporting in this cluster.
dev.to
1 article · September 2, 2026
Follow any of these and your For You feed starts watching them — no settings page required.
product
Shopify frees engineers to spend on AI tokens without an ROI case1 distinct publisher
build
Thirty runs of one prompt put a 34-point error bar around a brand's mention rate1 distinct publisher
build
Write access turned a keyword brief into two clones of the homepage1 distinct publisher
build
Gemini 3.7 Flash goes GA on one model layer, and that is the actual news1 distinct publisher
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
One issuer's numbers, one relay
Doubled spec-led conversion, 197% session growth, 14% higher order values — all of it originates inside Shopify, and reaches us through a single dev.to write-up that neither links the underlying company analysis nor states how many stores or sessions sit behind it. Shopify is the only party positioned to see this data, which makes the reporting faithful and the finding unverified at the same time. What lifts the score above a press-release floor is that the same direction shows up in two consecutive quarters and the caveats are Shopify's own, not a reporter's inference.
Real motion, no denominator
Everything disclosed about scale is a multiple: sessions up 197%, orders up about threefold. Multiples off an unstated base tell you the channel is accelerating and nothing about whether it yet matters to a given storefront. The firmest adoption signal is behavioural rather than volumetric — half of these sessions arrive on a product page, meaning shoppers have already done their comparing elsewhere. Breadth helps too: uplift appeared in 23 of 25 categories, so this is not one narrow product segment, though it remains adoption measured only inside Shopify's own estate.
Headline outruns its sample
"Twice the conversion rate" leads; "we won't say how big the dataset is" and "attribution remains imperfect" follow several paragraphs later. Nothing is misstated, but the ordering invites merchants to treat a directional in-house reading as a benchmark, and the two categories that showed no uplift never make it into the framing. The gap is modest rather than severe because dev.to keeps Shopify's hedges intact and even repeats the classification worry — a shopper who entered the journey mid-research will convert well whoever referred them.
Both tellers sell the remedy
Follow the recommendation and you arrive at two cash registers. Shopify's finding — structured Catalog data converts twice as well as scraped feeds — is an argument for merchants to invest more deeply in Shopify's own product data layer, published by Shopify. The write-up relaying it ends by inviting readers to book an AI visibility scan from Scalevise. That alignment doesn't falsify anything; it does explain why the most actionable comparison in the story is also the most saleable one, and why nobody here is pressing on the sample size.
Internally consistent, externally unchecked
Two quarters pointing the same direction, a plausible mechanism, and a source candid about its own limits — that is enough to treat the shape of this finding as probably real. It is not enough to trust the magnitudes. One publisher, one measurer, no replication, no sample size, and an explicit admission that referral classification could be doing part of the work. Our confidence sits where a reader's should: act on the structured-versus-scraped test, don't budget against the 56%.