Welcome to July's Commerce Jam.
This month is about a question we spent the first half of this year researching: when an AI assistant answers a buying question, whose words does it actually quote? We analyzed 1,000 Shopify stores and 60 product categories to find out. The answer surprised us, and it is the featured piece this month.
In other news, Adobe measured AI-referred shoppers converting at a 40% higher rate than in any other channel during Prime Day. Google's Search Console started reporting AI visibility to everyone, and Similarweb measured what an AI mention is actually worth.
The EU started collecting a new duty on every parcel entering the bloc, and Shopify folded it into product prices ten days later. HubSpot rewrote its data terms, faced four days of open revolt, and reversed. And Shopify quietly rebuilt the collections model in a way that will break integrations that are not paying attention.
We also get into two Google updates, the deadlines coming between now and November, and the usual sites and tools from the team.
Duplicate content and AI citations in 1,000 Shopify stores
Across the 60 product categories we tested, established brands earned an AI citation 9.5% of the time. Brand product pages of any kind, from any store, accounted for 2.8% of all cited sources. The other 97% went to publishers, reviewers, and community threads.
The usual diagnosis for this is thin content, and the data does not support it. Only 15.6% of the 1,000 stores in our sample have genuinely thin product content when you read the raw HTML, which is what the assistants read. Most merchants wrote the words years ago.
What the sample shows instead is a gap between the page a human sees and the page AI parses. Among stores with structured data in place, 59% leave the AI-readable description under 50 words, so the AI assistant is working from a stub. On top of that, 20% of product descriptions in the sample appear verbatim in at least one other domain, usually a retailer or marketplace that carries the same feed.
That second number is the most important one. An AI model deciding which source to credit has no reason to pick your page when the same paragraph exists on eight domains, and every incentive to pick the publisher who wrote something original about you.
The full article covers the methodology, the category-by-category breakdown, where the syndication problem concentrates, and the checklist we now run on client catalogs before any AI visibility work.
Shopify Platform Updates
These updates came from the Shopify Changelog, the Shopify Developer Changelog, the dev community forums, and our internal team. Everything here is from the last 30 days.
Collections are being rebuilt, and older integrations will not see them
The 2026-07 API version replaces the smart-versus-custom collection split with a single model built from product group sources. A collection can now combine multiple rule sets, reference other collections, and target specific variants, which makes an Activewear parent assembled from Leggings, Tops, and Shorts a native feature rather than a workaround. Our own Mike Warwick has been tracking the developer preview and calls it the largest change to the product data model in years.
The part that needs your attention is backward compatibility. API versions before 2026-07 cannot represent the new collections and do not return errors. They filter them out silently, so a feed or app on an old version keeps running while showing an incomplete catalog.
If you run custom collection logic, product feeds, or a merchandising app, put an API version audit on the list for this quarter. Doing the check now is cheap. Discovering it through a customer asking where half a category went is not.
Shopify Scripts stopped executing on June 30
Legacy Scripts are gone. Any discount, shipping, or payment logic still running on them stopped working on July 1, and editing has been locked since April. This closes a sunset that Shopify had postponed twice since 2024.
We went through all our client websites ahead of the deadline and found one store that needed review. That matches what I would expect across the industry: well-run stores completed the Functions migration a long time ago, and the merchants at risk are those who inherited Scripts from a developer who left. If your checkout math changed on July 1 and you do not know why, start here.
The EU duty is live, and Managed Markets now folds it into your prices
The EU abolished its €150 de minimis exemption on July 1. Under Council Regulation 2026/382, every B2C parcel valued at €150 or less now carries a flat €3 customs duty per tariff heading, added by category rather than by parcel. That means that if there are three product types in one box, you incur €9. The rule runs through July 1, 2028, and it covers roughly 93% of eCommerce parcels entering via IOSS.
Shopify's response arrived on July 10, when Managed Markets began embedding duties and import fees directly into product prices. The buyer sees one number, nothing surprises them at checkout, and your payout stays domestic.
Two things to plan for. Because the charge is per tariff heading, multi-category carts absorb the steepest increase, so the cost review should happen at the cart level rather than the SKU level. Additionally, a second handling fee of roughly €2 per heading is expected around November, which means EU landed costs get recalculated twice in one year. Model both now and decide in advance what you absorb and what you pass through.
Trustpilot reviews now sync natively with Shopify stores
Shopify and Trustpilot made a rebuilt reviews app generally available on June 29, with automated collection and native display in the storefront. In their announcement, Trustpilot disclosed an important stat: click traffic arriving from AI search grew 1,490% over its last financial year.
That growth rate says something about how AI assistants weigh evidence. An AI model evaluating a brand treats third-party reviews as independent testimony in a way it will never treat your own product copy, which is the same attribution logic our featured study documents at the product page level. If reviews have been an afterthought in your program, they are now part of how LLMs decide whether to recommend you.
B2B on Shopify keeps improving gradually
Three changes were launched within two weeks of each other. B2B discounts are on by default for new B2B stores as of June 23. Draft orders now include deposit fields for partial-payment workflows on July 1. Market-Driven Shipping entered feature preview the same day, organizing rates around target markets while the older merchant-owned delivery profile APIs were deprecated.
None of these is a headline on its own, and together they continue a roadmap that started with native net terms in the Summer Edition. Shopify is methodically removing the reasons wholesale portals stay in spreadsheets and email. Shopify's own B2B data shows merchants gaining up to 33% more self-serve orders within six months of launching a proper portal, and the Gen X and Gen Z buyers driving that number increasingly expect the same experience they get as consumers.
You can now control hreflang from the admin settings
As of July 10, merchants can turn automatic hreflang generation on or off in admin settings without a support ticket. Anyone who has fought Shopify's auto-generated hreflang on an international store knows how long overdue this is.
The default behavior helps most stores and actively hurts a few, particularly where market structure and domain strategy diverge. Elsid's new international SEO guide walks through when to keep the default and when to override it, based on our experience with client migrations.
Four new staff permissions separate the money from the admin
Since July 7, payments, payouts, disputes, and tax documents exist as separate staff permissions. A bookkeeper/accountant can pull payout reports without holding full admin access.
In our audits, over-permissioned staff accounts are among the most common findings, usually because the platform never offered anything more granular. Now it does. Fifteen minutes with the permissions page is worth it.
The Shopify Tax fee
A pattern from our production team, courtesy of Chris Roach, our tax Director of Solutions and Sales Engineering. Merchants who use TaxJar, Vertex, or Avalara for filing still get billed by Shopify for tax calculations because Shopify charges whenever it performs the calculations, regardless of who files the return. Several clients found the line item this quarter and assumed it was an error.
It is not an error: calculation and filing are separate services, and the fee is based on the calculation. If you want a single bill, choose a provider that does both. If you keep Shopify Tax for calculation, budget for the fee, and ask whoever scoped your tax stack which side of that line you sit on.
Shopify moves toward a US vape ban
After pressure from 25 state attorneys general over an illicit vape market they estimate at $9 billion, Shopify is preparing to ban US vape sales on the platform.
The narrow story is vapes. The broader one, for anyone operating near a regulated category, is that platform policy moves faster than legislation and arrives with less warning. If a category ban would materially hurt you, the time to understand your dependencies and your alternatives is before the AGs' letter, not after.
POS updates: connectivity health, staff attribution, and an activity log
A selling environment health screen now shows internet, Shopify service, and hardware status with offline guidance. Staff attribution was turned on by default on July 6 for stores that had never configured it, so in-person sales are now attached to named staff. And a new POS activity log ties high-risk register actions to specific people.
Unglamorous, all three. But attribution and audit trails are the difference between noticing shrinkage in a report and reconstructing it from memory, and most multi-location retailers we work with had exactly none of this a year ago.
Smaller updates worth a minute
Flow now supports copy and paste for workflow steps with configuration intact, which turns duplicating automation logic across workflows from a chore into a shrug. Purchase orders now create transfers directly, closing a small gap in inventory operations that previously required a manual step. And a new app pixel activity log shows when an app changes what customer data it collects, which is worth a look after the HubSpot story below.
Some deadlines for your calendar
July 27, 2026: Amazon cuts product titles to 75 characters in nearly all categories, with a new 125-character Item Highlights field. Audit your best sellers before the deadline rather than trusting the automated rewrite.
July 30, 2026: Amazon's New Selection Program begins. Eligible new-to-FBA products get referral fees capped at 10% on the first 100 units and 5% on the next 100, which is worth timing Q3 launches around.
August 3, 2026: Google begins using IP addresses for ad personalization in the EEA, UK, and Switzerland.
August 5, 2026: Shopify reports Q2 earnings before market open. It is the first quarter with Catalog and UCP fully live, so watch for any disclosure on AI channel volume.
Around November 1, 2026: The EU adds the roughly €2 per tariff heading handling fee on top of the €3 duty. Second landed cost recalculation of the year.
Pending: USTR held its Section 301 hearing on July 7 on proposed tariffs of 10 to 12.5% across roughly 60 economies. The final determination can arrive at any time, and importers should already have both rates modeled.
AI, SEO & Agentic Commerce
Prime Day was the first real stress test of AI referrals, and they converted best
Adobe reported that shoppers arriving from AI chatbots were 40% more likely to complete a purchase during the four-day Prime Day event than visitors from search, email, or social. In Adobe's earlier data, the same cohort was the least likely to buy from any channel, so this is a full reversal, measured on the biggest shopping event of the summer. Their broader June analysis points the same way: AI-referred visitors across retail spend 49.9% more time on site, add to cart 33% more often, and convert 50.7% better than other sources.
The result is announced weeks after OpenAI retired its own in-chat checkout, the retreat we covered in June, and Forrester's read of that pullback holds up well against the new data. Walmart reportedly measured ChatGPT-native conversion at roughly a third of what the same shopper produced on Walmart.com. Shoppers want the agent's advice and the merchant's checkout.
Two things are true at the same time: AI agents have become excellent at delivering ready-to-buy shoppers, and they remain poor at closing the sale themselves. For a merchant, the argument is to spend on being parseable and trustworthy within the AI agent's answer, and to avoid spending on in-chat checkout until the conversion data says otherwise. That work is catalog and content, which is exactly what the featured study measures.
The measurement layer for AI search finally arrived
Three measurement releases were launched within days of each other. Google's Search Console AI performance reports are rolling out to more users, and John Mueller clarified the counting rules on June 25: an impression means a link to your page appeared in an AI Overview or AI Mode, and the same URL in the overview and the blue links counts once. Bing took this one step further on June 16, adding Citation Share to Webmaster Tools, a metric that shows your percentage of citations for a given query, with intent and topic breakdowns behind it.
The more consequential release is Similarweb's Downstream Impact of AI Visibility, built on six months of US panel data with Rand Fishkin advising. Brands recommended by an AI assistant were 2.5 times more likely to receive a site visit within the following seven days, and those visitors spent more time on a site, 12 pages and nearly 12 minutes, against 6.5 pages and 5.6 minutes for everyone else. The catch sits in the attribution: 55.9% of that downstream traffic arrives days later as branded search, where the AI recommendation gets zero credit.
AI assistants have been influencing your revenue all along, and your analytics have been crediting Google for it. Citation share belongs on the same dashboard as rank now, and between Search Console, Bing, and the free trackers in the tools section, the tooling excuse is gone.
A four-year study of 100 blogs, and what the survivors did differently
Daniel Stanica tracked 100 once successful blogs across four years of Helpful Content Updates and AI Overviews. The median organic traffic loss was 85%, and only 21 of the 100 are still growing. It is the most complete longitudinal dataset anyone has published on what AI search did to content sites, and the headline supports every "SEO is dead" post you have seen this month.
The survivors argue against that reading. Stanica found the 21 growing blogs shared four traits: firsthand experience, an owned audience beyond Google, a real product or service behind the content, and branded search demand. None of those is a content tactic, and all four are reasons a model or a person treats a site as a source rather than a summary.
Suganthan Mohanadasan spent days reading ChatGPT's raw network traffic and published how it actually picks sources: the model goes to the official page first for pricing and specifications, and falls back to third parties like G2 when your page loads those facts through JavaScript it cannot parse. His line for that failure deserves a pin: "Your facts, someone else's page, because yours wouldn't parse." Reviews and communities own the opinion citations either way, but the factual citations about your own products are yours to win or lose, and our study suggests most stores are losing them.
The agentic layer has reached general availability
Salesforce made Agentforce Commerce generally available on July 6: a Shopper Agent for storefront conversations, a Buyer Agent that meets B2B buyers in WhatsApp and SMS, and a Merchant Agent for back-office work. Native ChatGPT integration is live now, with Google AI Mode and Gemini following this summer. Salesforce says AI influenced 20% of global online sales during the 2025 holidays, worth $262 billion, though it is worth remembering that figure comes from the company selling the agents.
Shopify reached the same conclusion in a different way, turning on the Universal Commerce Protocol by default for every store in the Summer Edition. Between the two, agentic access to commerce catalogs stopped being a roadmap item this quarter and became infrastructure that simply exists.
The question we should all ask is whether a website's product data would survive an AI agent reading it cold, with no human there to compensate for a vague title or a missing attribute. Most catalogs fail that reading, and the failure usually has nothing to do with the platform.
AgentMCP is coming to Google Chrome
Chrome is rolling out WebMCP to all users, a standard that lets sites expose structured actions to AI agents directly in the browser session.
In parallel, Cloudflare's PACT protocol, which lets a site verify that a human is in the loop behind an agentic request, added Chrome, Edge, Firefox, and Shopify as signatories. Neither changes anything on your Q3 roadmap. Together, they describe where this is going: agents as first-class citizens of the web, with readability and trust standards underneath them, and structured data graduating from SEO chore to load-bearing infrastructure.
Meta's new image model feeds straight into your ad account
Meta launched Muse Image, the first image model from its Superintelligence Labs, and is wiring it into Advantage+ creative with agentic brief parsing, so the system can take a rough brief and generate the variations itself. Meta says 8 million advertisers already use its AI creative tools, double last year.
My personal view on AI ad creative has not changed since the same idea was called dynamic creative optimization. Once every advertiser can produce 200 variations, volume will no longer be an edge, and the differentiating input becomes what you feed AI: a distinct visual identity and real product truths to vary. Get the brand system tight before handing Advantage+ the keys, because declining the tooling is not a position either.
Two Google updates in six weeks
Google ran a June spam update from June 24 to 26, an unusually fast two-day rollout aimed at sites violating its spam policies. It follows the May core update that finished on June 2, during which Sistrix found reference brands, local eCommerce entities, and category-defining marketplaces benefiting, while interchangeable affiliate-style content lost ground. The same preference for editorial originality appears in the AI citation data, as reflected in classic rankings.
If your rankings changed in late June, separate the two updates before diagnosing, because recoveries from spam updates differ from losses from core updates. Second, Google tightened its domain migration requirements in mid-June. If you are replatforming or consolidating domains this year, the redirect and verification work now requires more care than the standard checklist assumes, and we are updating our own migration processes to reflect this.
eCommerce News & Industry Trends
Prime Day reached $26.4 billion, and measured which retailers let agents in
US shoppers spent $26.4 billion online across all retailers during the four-day event, per Adobe, with Amazon itself taking roughly $15.6 billion, up 7.1% year over year per eMarketer. Growth, but unremarkable growth for the tentpole event of the summer.
The detail with more information comes from J.P. Morgan's data in GeekWire's Prime Day coverage: agentic AI drives under 1% of traffic at every major online store, and Amazon's share is the lowest in the group at about 0.4%, by design. Walmart and Target opened their catalogs to outside AI assistants, while Amazon blocked the crawlers and won a preliminary injunction against Perplexity's shopping browser. Openness to AI agents is now a measurable strategic split among the biggest retailers, and a merchant with exposed, parseable product data sits on the open side of it by default.
Amazon started paying for placement in ChatGPT
Amazon bought its first ads inside ChatGPT to promote Prime Day, while continuing to block AI scrapers from its catalog. The two policies are consistent once you see the strategy: refuse to be aggregated, pay to redirect attention back to a storefront you control.
The channel around that decision is maturing quickly. OpenAI told reporters in Cannes it is "clearly in the advertising business," with more than 2,000 brands advertising and about 20% of the queries from its 900 million weekly users carrying commercial intent, while Amazon's own conversational display ads report that 70% of buyers are new to the brand. And the price of entry keeps falling: Back in April, Digiday reported CPMs compressed from roughly $60 in the early pilot to about $25, which is the window before real budgets arrive and reprice it.
HubSpot changed the rules on customer data, then reversed them in four days
On July 1, HubSpot updated its terms of service to feed customer CRM data into a shared enrichment pool for an upcoming Contact Discovery feature, with every account opted in by default. By July 5 the company had reversed course, its product chief published "We got this wrong. And we are fixing it", and co founder Dharmesh Shah committed to opt in only data changes going forward.
We use HubSpot ourselves and moved our email program onto it this spring, so I read the terms changes with some personal interest. CRM data that customers spent years building is now an asset every platform wants to pool and train on, and default opt-in is how it will be attempted, because asking first yields lower adoption rates.
Percentage discounts and dollar discounts do different jobs
Seguno analyzed millions of Shopify discount codes and found percentage off promotions drove 42% higher average order values than amount off, $95.75 against $67.35, while amount off codes redeemed 1.26 times more often. The sweet spot in the data sat at 20% off.
Read as a pair, the two formats separate cleanly: dollars off converts the hesitant buyer, percentage off grows the committed one's cart. Most promo calendars we review treat them as interchangeable, which quietly leaves one of the two jobs undone. Match the format to what the campaign is for before BFCM planning starts in earnest.
Faire opened to business buyers, and wholesale expectations moved again
Faire, the wholesale marketplace Shopify holds a stake in, opened to business use buyers: hotels, restaurants, offices, and event planners can now purchase wholesale directly, placing Faire in the lane Amazon Business and Walmart's B2B effort already occupy.
Put that next to what Shopify made native this quarter, with default B2B discounts, net terms, and ACH, and the trajectory for wholesale is set by the buying experience rather than the sales relationship. Brands that still gate wholesale behind a rep and a PDF price list are training their buyers to prefer whichever marketplace asks less of them, and early July reports putting Shopify in substantive acquisition talks with Faire only raise the stakes. The B2B guide covers what a self-serve portal needs before the marketplaces set the default.
Shero News
On the Shero Live: why most merchants run two or three CRO tests a month
Our June session brought in the team from Visually for a mid-year CRO reality check: why most Shopify merchants are stuck at two or three tests a month, and what testing velocity looks like when the bottleneck comes out. The short version is that most brands are not limited by ideas or traffic. They are limited by how long each test takes to build, which is a tooling and process problem with a known fix.
The full 62-minute recording and the written recap are on the site, alongside the rest of the Shero Live library. We also rebuilt the recap format this month, so every session now gets a proper written summary with the embedded video, and you can skim the takeaways in five minutes before deciding whether the hour is worth it.
Shero Team out in the field
July is a travel month for us. Gavin spoke at the Dotdigital Summit on July 2 about adapting to a world where AI is deciding for your client. 
On July 16, we are co-hosting a webinar with our client, Gen-Y Hitch, and Patchworks on their migration to Shopify, B2B, integration challenges, and how we overcame them together. There is still time to register.
Then a group from the team heads to Dotdev from July 18 to 23. Half the value of these events is the hallway conversations, and we reliably come back with two or three ideas that end up in client roadmaps. If you are at any of these, find us.
On the client side, June and early July brought a number of new projects across migrations to Shopify, SEO, CRO, B2B, and email, including several brands moving deeper into the AI visibility work this newsletter keeps returning to.
From the blog
A few more reads from the Shero blog this month:
International SEO for Shopify stores selling to the EU and the UK by Elsid covers hreflang, market structure, and pricing pages in the new duty era, based on our real experience with client migrations.
How to fix Shopify UTM tracking and stop losing attribution data, also by Elsid, starts from an observation we keep confirming: the attribution gaps are rarely in your ad platform. They are in your links.
7 friction points killing your Shopify conversion rate by Altin Gjoni distills hundreds of CRO audits, and pairs well with the Shero Live session above.
Sites Worth Seeing
Four sites the team passed around this month, all courtesy of Ash's radar. As usual, they are chosen for craft, not because they run on any particular platform.
dulve-in.com - A hemp milk brand built on restraint: no oils, no gums, no added sugar, and a site that argues the case through education rather than volume. The product page reads like a position paper, in a good way.
koppen.co - Premium oral care sold as a three part system rather than a shelf of SKUs. The bundle architecture quietly does the merchandising work, and it is a useful study in selling a routine instead of a product.
balmoralrunning.com - Montreal running apparel with tight, editorial product presentation. Proof that a small catalog can carry a full brand when the art direction is committed.
bathroomcity.co.uk - A UK bathroom retailer pairing a deep catalog with genuinely useful buying guides, on Shopify. This one is personal: it was Ash's final project before joining us, and it went live this month.
Tools Worth Knowing
A short list, biased as always toward things you can use this week rather than demos.
FreeSOV. A free AI share-of-voice tracker across ChatGPT, Claude, Gemini, Perplexity, and AI Overviews. The fastest zero-budget way to learn whether the assistants mention you at all, and a natural first step after reading this month's featured piece.
Microsoft Clarity Bot Analytics. Clarity now flags AI crawlers that violate your robots.txt. Useful for deciding, with evidence instead of vibes, which bots you actually want reading your catalog.
Lantern. A new agentic commerce platform that monitors how your products appear inside AI shopping results and helps close the gaps. Early, and aimed at exactly the problem our featured study documents.
Claude Tag. Anthropic's AI agent that lives inside Slack group chats. Kevin from our team set it up on our internal channels, and it has quietly replaced most of the "can someone summarize this thread" moments in a 65-person company.
Cloudinary Image to Video. Turns product stills into short videos programmatically. Worth testing against a sample of your catalog before you pay for a video shoot.
Text for Shopify. Connects WhatsApp Business to your store for conversational support and recovery flows. WhatsApp continues to grow as a commerce channel, and most US brands still have no presence there.
ImageKit Creative Automation. Generates and resizes product creative variations at scale with an AI assist. If your team spends hours cropping the same hero image for six placements, the math works out quickly.
Final thoughts
Growing up in my hometown in Albania, there were mountain trails everyone used, but nobody remembered how they came to be. They still exist today. Shepherds walk them, tourists photograph them, and the mapping apps show them. The person who first worked the path through the rocky terrain gets nothing, because a trail carries no signature.
Syndicated product content ends up the same way. Your words move through feeds, marketplaces, and resellers until they are everywhere and attributed nowhere, and the AI walking that trail credits the map, not the trail cutter. The way back is to write things that could only have come from you in a place the LLMs can read.
Original data carries a signature. That is what we keep finding, and we will keep publishing what we find.
Until next time,
Warm regards,
Gentian
