Welcome to August's Commerce Jam.
This month Shopify switched agentic tools on across every storefront and posted a huge Q2. ChatGPT started selling ads to anyone who signs up, and a study of 2 million listings measured how few products Google's AI Mode actually shows. All of it points the same way: selling is turning into a conversation.
The featured piece takes that literally. I built an AI agent that sells on WhatsApp for a Shopify store without using any external app. It works, there is a recording of it working, and the code is yours for free.
Let's dive in!
Selling on WhatsApp with Shopify, code included
Selling on WhatsApp is now a real option for Shopify stores.
Think about how people treat their two inboxes. Nobody reads marketing emails the way they read WhatsApp messages. The numbers back that up: open rates are around 79% compared to email's 20%. Your customer's inbox holds every brand they ever bought from. Their WhatsApp holds family and friends, and you would be nearly alone in there.
If you sell into Europe, WhatsApp is already how your customers talk to everyone they know. In the US, it is where the younger buyer lives. So when Shopify quietly added WhatsApp opt-ins to Forms and WhatsApp campaigns to Messaging this summer, two of the three pieces you need showed up natively.
The last piece is the conversation where someone asks for a product and buys it. I wanted to know what that piece actually takes, so I built it. An AI agent sits between the store and the WhatsApp number, answers with real products from the catalog, and sends a checkout link. There is a screen recording of the whole conversation in the article.
The code is free on GitHub under an MIT license, and the repo includes a Claude skill that does the setup for you, from the Meta app to the first paid order. Read the guide, grab the code, and start collecting opt-ins now, because everything else can easily be set up, while the list cannot.
Shopify Platform Updates
These updates came from the Shopify Changelog, the developer changelog, and our own team. Everything here is from the last 30 days.
Q2 earnings: 30%+ growth on every line
Shopify reported Q2 on August 5: $115.6B in GMV, up 32%, revenue of $3.6B, up 34%, and the fifth consecutive quarter of 30%+ GMV growth. B2B GMV grew 76%, international 37%, and guidance points at low-30s revenue growth again for Q3.
The agentic commerce numbers came out on the earnings call. Catalog, Shopify's product feed for AI assistants, now holds over 1 billion products and converts at twice the rate of general search, and AI search traffic converts about 80% higher than traditional organic.
Shopify's President, Harley Finkelstein, also confirmed agentic transactions carry the same economics as online store orders, with no new fees. Sidekick's numbers give a sense of how fast merchants are adopting AI tooling: daily active merchants up 3.6x, and 36,000 custom apps built with it, up from 12,000 in Q1. Shop Pay GMV grew 53%, and payments penetration reached 68% of GMV. Those AI conversion numbers match what Beth heard at dot dev Toronto, covered in Shero News below.
Every storefront can now talk to AI agents
As of August 5, WebMCP went live on all Liquid storefronts, with Hydrogen in developer preview. All of it is ready to use out of the box and needs no setup from merchants. An AI agent in the browser can now call 9 structured tools on your store:
- search_catalog
- browse_store
- get_product
- show_variant for the catalog
- get_cart, update_cart
- cancel_cart for the cart
- proceed_to_checkout
- manage_orders for buying
- search_shop_policies_and_faqs for everything else.
Agent cart actions use standard storefront mechanics, so your theme's cart and analytics still work.
The browser side is the only limit at the moment. Agent support is limited to a Chromium origin trial for now, and Shopify is working on the spec with Google and Microsoft. Every store got this by default. We now test this on client stores as part of AI readiness work.
The old storefront MCP cart retires August 31
The cart tools on the storefront MCP endpoint at /api/mcp are deprecated and stop working August 31. The successor is the UCP cart endpoint at /api/ucp/mcp, and migrating takes more than swapping the URL: every request carries a ucp-agent profile in a meta object, cancel_cart requires an idempotency key, and update_cart now expects the full line items array rather than partial updates.
I verified the deprecation firsthand against the live API while working on the WhatsApp agent, and the open source Worker in my repo already uses the new endpoint. If you built anything on /api/mcp cart tools, you have under three weeks.
Managed Markets drops DDU on August 24
Announced August 11, effective August 24: Managed Markets stops supporting delivered duty unpaid everywhere it supports DDP. Markets on DDU move to DDP automatically, so buyers pay duties and taxes at checkout instead of at the door.
That is under two weeks' notice for a change your buyers will see in their checkout total, and the only way out is to turn off Managed Markets entirely before the date.
For most merchants, the forced move is the right outcome, since surprise courier fees at delivery are a top reason for refused parcels. Model the checkout price change on your top international markets before the 24th rather than discovering it in conversion data.
Microsoft Advertising joined Campaign Autopilot
Since August 4, Campaign Autopilot can create or connect a Microsoft Advertising account and start Performance Max campaigns across Bing, Copilot, Edge, and Outlook, with budget allocated automatically across channels by performance. Microsoft bills the spend directly.
This is the first way to buy presence in Copilot from the Shopify admin. Ad inventory inside AI assistants is arriving through the ad platforms merchants already use, so testing it is a straightforward settings change, with no new platform to learn.
The payout currency cap is gone
Since July 27, the eight-currency limit on payout bank accounts is removed. You can now add one bank account per supported payout currency and get paid in every currency your region and plan allow.
Fees still apply on non-domestic payouts. For merchants paying suppliers or staff in multiple currencies, holding balances in those currencies cuts FX exposure without a treasury product. If you sell in ten markets, you no longer have to choose which eight currencies need a bank account.
Shop Campaigns data opened up
As of August 10, Shop Campaigns performance is queryable through ShopifyQL under a new shop_campaign_insights schema: spend, sales, orders, ROAS, AOV, and acquisition cost, by campaign, segment, and time.
The existing read_reports scope covers it, and Triple Whale, Northbeam, and Elevar already integrate with them. Shop Campaigns spend has been a reporting blind spot since launch, so getting it into the same tools as the rest of your attribution is overdue and welcome. Pull campaign-level ROAS into the same dashboard as your paid social before you set BFCM budgets.
Managed Markets opened to Canada and the UK
Canadian and UK merchants can use Managed Markets with Global-e as the merchant of record, which provides: duties-inclusive local pricing, automatic tax registration and remittance, discounted DHL and FedEx labels from the admin, and daily payouts in CAD or GBP.
Cross-border compliance is the part of international selling that stalls mid-market teams, and merchant-of-record programs move that burden off the merchant entirely.
Sidekick moved in-house, 96% cheaper
Shopify Engineering published the details on August 5: the frontier model behind Sidekick's store analytics questions was replaced with a small in-house model that retrains every day, cutting serving costs from about $27M a year to about $1M, with 38% faster end-to-end responses.
The retraining loop grades its own worst conversations nightly, repairs them with a panel of reasoning models, and feeds the fixes back as training data. The fine-tuned model now scores better than the frontier model it replaced.
The same math applies outside Shopify: a small model trained on your own domain data can beat a frontier model at a narrow job, at a fraction of the cost. Support bots and product Q&A at merchant scale are the obvious places to use it.
Smaller updates worth a minute
The POS printed receipt editor was redesigned on August 11 with a single section covering returns and exchanges, per-location overrides, and proper VAT, GST, and ABN fields. DHL Express became available natively for merchants in the UK, Germany, France, Italy, and Spain on August 13. UPS return labels also arrived in Shopify Shipping on July 29, closing a gap that pushed many US merchants to third-party label apps for returns alone.
Some deadlines for your calendar
August 24, 2026: Managed Markets ends DDU support. Affected markets move to DDP automatically.
August 29, 2026: One year since the US suspended the sub-$800 de minimis exemption. A good prompt to re-check the duty math in your US inbound pricing.
August 31, 2026: Storefront MCP cart tools at /api/mcp stop working. Migrate to /api/ucp/mcp.
September 15, 2026: Cloudflare's new AI crawler defaults take effect for newly onboarded domains: training and agent crawlers blocked by default on pages showing ads, search crawlers allowed. Existing domains keep their current settings.
September 30, 2026: Amazon requires a 90% business hour delivery rate on seller-fulfilled orders to Amazon Business customers in the US, UK, and Germany. Offers can be deactivated for business buyers by October 30 if you stay under it.
AI, SEO & Agentic Commerce
ChatGPT ads went self-serve
OpenAI opened its "Advertise in ChatGPT" Ads Manager to any advertiser on July 22, live in the US, UK, Canada, Australia, New Zealand, Japan, and Korea.
Ads now appear as labeled sponsored suggestions alongside responses, targeted on conversational context rather than keywords, with both CPM and CPC pricing. Self-serve is the last step of a rollout that added CPC bidding, a measurement pixel, and a Conversions API back in May.
Best Buy, Lowe's, and VistaPrint were among the early advertisers, and two days later OpenAI added a conversions objective, geo exclusions, and bulk campaign tools.
My personal view is that this deserves a small test budget, with two conditions. Install the measurement pixel and Conversions API before spending a dollar, because I've seen early advertisers calling the channel "quite expensive with not much return" after two weeks, and without conversion data you cannot tell whether that is the channel or their setup.
Treat the conversational targeting seriously too: the AI model decides relevance from what it knows about your brand, which is one more reason to do the AI visibility work covered in the rest of this section.
AI Mode shows 95% fewer products than search
Productrise ran the same shopping queries through standard Google search and AI Mode, same day, for 21 days in July: 100,000+ result sets and over 2 million product listings across the US and UK.
Standard search returned products on 88% of shopping queries. AI Mode returned them on 23%, and with roughly 4.3 listings per answer against 22.5 per results page. Put together, that is about 95% fewer product listings. The overlap is the most important number: only 0.8% of products visible in standard search also appeared in AI Mode for the same query on the same day.
Ranking well in classic search buys you almost nothing in AI Mode, according to this data. Classic search and AI Mode pick products differently, so measure them separately and treat AI Mode visibility as its own project. If Google makes AI Mode the default and you only optimized classic search, your products will be missing from most answers.
AI Mode reached France, and measurement got messier
Google switched on AI Overviews and AI Mode in France on July 22, closing its last major EU gap. The German precedent explains why French publishers moved fast: with AI Overviews present, position-one CTR there dropped from 27% to 11%.
Google published a domain-level opt-out toggle two days before the launch, with page-level controls not scheduled until March 2027. By August 11, an association of roughly 300 French newspapers had referred Google to the French competition authority, alleging traffic losses up to 38%.
Search Console counts AI Mode impressions inside your overall Web Search totals, and the newer Generative AI report shows AI impressions without clicks while those same impressions also sit in the totals, so the reporting needs more work and clarity.
Impression growth across an AI Mode rollout can be the rollout, and clicks are the honest metric. Annotate July 22 in GA4 if you have French traffic, and read impressions with suspicion for a few months.
Google Maps now orders your dinner
Ask Maps, the Gemini-powered conversational layer in Google Maps, gained agentic features on August 6: food ordering through Square and Toast with Uber Eats coming, hotel booking based on your budget and criteria, and event tickets.
The demo query is worth reading as a spec: "order spicy pad kee mao from that Thai place I like, for pickup on my way home." Maps picks the restaurant from your saved places, fills the cart, and hands off payment to the partner.
An optional Personal Intelligence setting, off by default, lets it read Gmail and Calendar for flights and reservations. These agentic features are US first.
Agentic checkout arriving inside the app people already navigate with is a bigger distribution story than any AI browser. For local retail and food, the merchant side of this is your POS platform: Square and Toast sellers are in the first wave without lifting a finger, and more agentic distribution will arrive the same way, through the platforms that already host your catalog and your checkout.
Classic SEO still decides your AI visibility
Contrary to the Productrise research findings I mentioned above, Cyrus Shepard published a useful corrective on August 4: the SEO strategies that drive 90% of your AI visibility are adapted SEO fundamentals. His claim rests on the citation ranking factors study he assembled from 54 studies, experiments, and patents, where the top-scored factors read like a classic SEO checklist: URL accessibility, search rank, ranking across the fan-out queries AI search sends out behind your prompt.
The layer that is genuinely new sits after retrieval, where reranker models pick the handful of cited sources from the 40 to 65 the system fetched, a mechanism Metehan Yesilyurt has documented in detail: short, answer-first, factually specific passages get picked, and a page can rank in the retrieval set on every sub-query and still get skipped at the citation stage.
Our own duplicate-content study from last month reached a similar conclusion on the data side, where stores losing AI citations mostly have classic, fixable problems. Before buying anything with GEO in the name, do the boring audit.
Cloudflare now measures your AI visibility
Cloudflare added an AEO Visibility dashboard to its answer engine optimization suite on August 6. It takes a set of buyer prompts for your industry, puts them to AI models like GPT and Claude, and reports citation rate, mention rate, prominence, and share of voice, with category benchmarking that shows which brands the AI models group together and which competitors are gaining visibility.
Cloudflare also has one data source the SEO tools cannot get: it sits in front of your site, so it pairs those answer-level numbers with the actual AI crawler activity and referral traffic it sees directly.
We set it up on our website last week, and the whole thing took under five minutes. If your site is behind Cloudflare, the steps are: log in to the dashboard, open your domain, then click Agent Readiness in the left menu.
The Diagnostics scan is free on any plan and grades your site across three levels, from house rules for AI agents (robots.txt, sitemap, content signals, AI-optimized Markdown) up to whether agents can log in and complete tasks, plus an optional commerce check covering inventory and purchases. The visibility dashboard itself is early access: press Register your interest on the same screen, and you are in the queue.
Our own scan came back "Almost ready". Do the scan before your next AI visibility conversation, because it gives you a graded checklist to work from.
The Agent Readiness scan for sherocommerce.com, with the AEO Visibility panel on early access. The AEO percentages shown are from Cloudflare's sample data, and the real numbers are available with access.
The court sided with the shopping agent
The Ninth Circuit vacated Amazon's injunction against Perplexity on August 4, unanimously. Amazon had blocked the Comet browser and its shopping assistant under the Computer Fraud and Abuse Act; the court held the CFAA "remains principally an anti-hacking statute" and that when a person shops through Comet, the person is the one accessing Amazon.
The AI assistant is a tool under the user's direction, like a browser extension, and the court explicitly left the door open to different outcomes for more autonomous systems. The separate Dow Jones copyright case against Perplexity is still in pretrial and unaffected.
Blocking AI shopping agents with hacking statutes just got much harder, so the fight moves to terms of service and contract law. Merchants should read this as one more reason the agent-access question is a business decision now, and one you should make deliberately: decide what agents may do on your store, publish that, and optimize for the agents you want.
eCommerce News & Industry Trends
Amazon's AI is rewriting product titles
Enforcement of the 75-character title limit began July 27, and Amazon is now auto-generating AI title recommendations for non-compliant listings, built from the seller's own product data.
Amazon says over 984 million titles have been updated since the June announcement. Brand Registry sellers get a 14-day window to review or modify the AI version before it goes live; sellers without Brand Registry get no window at all, though anyone can override a change afterward through Manage All Inventory.
The new 125-character Item Highlights field started displaying under titles on August 10, and Amazon says the full rollout continues through 2026, with non-updated listings staying active and searchable in the meantime.
Audit your best sellers now rather than trusting the rewrite, because title edits at this scale move CTR and Sponsored Products performance, and you want those changes to be yours.
Google's ads now use IP addresses in Europe
As of August 3, Google uses IP addresses for ad measurement and personalization in the EEA, UK, and Switzerland, jurisdictions where it previously did not, wrapped in on-device processing and trusted execution environments.
The change reaches AdSense, Ad Manager, Display & Video 360, Campaign Manager, and Search Ads 360, so it touches most display and video buying in the region. The compliance detail is in your consent setup: Google is registered for TCF Feature 3, so your consent platform must disclose it to Google, or ad processing may be limited.
If you sell into Europe, this is a one-question email to whoever manages your consent banner, and worth sending this week.
Amazon's quarter came with a $53.4B Anthropic gain
Amazon reported Q2 on July 30: net sales of $200.6B, up 20%, with AWS at $42.2B growing 37%, its fastest in 18 quarters. Net income of $62.6B includes a $53.4B non-operating gain, primarily from revaluing its Anthropic investment, and on the earnings call capital spending for 2026 was raised to about $220B. Advertising grew 26%, and Amazon Now, the quick commerce experiment, grew gross sales 80% quarter over quarter.
Read those numbers together and the retail business is becoming the distribution arm of an AI infrastructure company. For merchants, the near-term signal is the ads line: Amazon advertising keeps growing faster than its retail business, which means the marketplace gets more pay-to-play every quarter.
TikTok Shop did $50.3B in half a year
Momentum Works measured TikTok Shop's H1 2026 at $50.3B in GMV, up 92%, with the US at $11.8B, up 103% and now TikTok Shop's largest market. The mix detail changes how you should sell there: the Shop tab drives 51.4% of attributed GMV against 8.2% for live streams, so in the US most buying happens from the Shop tab, and live selling is the minority channel.
Seller depth is growing fast too: more than 5,700 stores passed $1M in GMV, 8x the count a year earlier. Average prices fell in 21 of 27 categories, which is worth putting into your margin plans before going deeper.
Shero News
On the Shero Live: takeaways from dot dev Toronto
Beth's new episode, "Agentic commerce, Vercel's Eve AI Framework, and Magento migrations: takeaways from dot dev Toronto," recaps the week. Shopify shared figures at the event showing AI traffic converting well above organic search, with about half of AI sessions arriving directly on the product page.
She also covers the Vercel and Shopify partnership around the open-source Eve agent framework, what's actually hard in Magento migrations, and the five Shopify updates she would prioritize before Black Friday. The written recap sits alongside the video, so you can skim it in five minutes.
The Shero team in the field
More than a dozen of us were at dot dev Toronto in late July. The team came back with a takeaways document that has already changed two client roadmaps. Agentic commerce ran through nearly every hallway conversation at the event, and the conversion figures Shopify shared there are the same ones that showed up on the Q2 earnings call two weeks later.
I've spent the last month in Greece and Albania. Part of our Finnish team paid a visit to our Albania office this month, resulting in a rare photo of both teams in one place.
At the Tirana office. From left: Petteri, Gentian, Mikko, Kozeta, Beth, and Olli.
In July we also co-hosted a webinar with our client Gen-Y Hitch and Patchworks on their migration to Shopify and the integration work behind it. You can find it on the webinar page.
People news
Cande Rybecky stepped up to Chief Marketing Officer, after leading the marketing team in everything but title for a while now. Her team is behind the results our clients are seeing in their digital marketing.
Hemi joined us as Performance Marketing Manager, adding paid media depth to that same team. We also signed our first Amazon marketplace client, a brand we have done Shopify work for, which makes the marketplace practice official.
Site launches
A few new launches that went live last month. One of them is Island Watch, a migration from BigCommerce to Shopify Plus. They sell watches people actually wear, with a catalog deep enough to make merchandising the hard part. 
Another one is, Graf & Sons brings a large ammunition and reloading catalog onto Shopify, a vertical where platform rules make eCommerce genuinely difficult.
From the blog
Server side tracking for Shopify stores explains why ad platform numbers and Shopify orders stop matching, and when moving tracking from the browser to a server is worth it.
Shopify BFCM pixel and tracking health check is a twenty-minute check covering Meta, Google Ads, GA4, and TikTok, written so you can do it before the season without a developer.
Google Search Console now tracks your social and video content covers the new platform properties and how to see which searches send people to your Instagram, TikTok, and YouTube content.
Semantic triples in eCommerce SEO breaks down how Google and AI models pull facts from your product pages, and how to write pages they can quote.
Sites Worth Seeing
Six picks this month, chosen for craft rather than platform.
notom.co Adventure gear, on Shopify. A full-bleed video hero and a mono-spaced type system over a dark, editorial catalog. Outdoor gear sites mostly look the same, and this one does not.
realeactives.com Skincare, on Shopify. Campaign photography carries the whole homepage, with a serif and sans pairing that keeps it from reading like a template.
giellygreen.co.uk London salon and product line, on Shopify. Booking and product cart live in one header, which is harder to do cleanly than it sounds, and the sculptural product photography earns the screen space.
cotypefoundry.com Type foundry, digital goods. The specimen grid is the store. Our one non-Shopify pick this month, included because selling a digital product this cleanly is rare on any platform.
rains.com Danish rainwear, on Shopify. One full-screen campaign film does the selling, and the catalog underneath stays fast and simple.
phoebephilo.com Luxury fashion, on Shopify. No search bar, no filtering, and structured data that gives AI models almost nothing, and all of it looks deliberate. A luxury house can make that trade. Enjoy the photography, and copy none of the architecture unless you can also afford to ignore Google.
Tools Worth Knowing
A short list, biased as always toward things you can use this week rather than demos.
prompt-master. A free Claude skill that turns a rough request into a precise, full-context prompt before the AI sees it. Past 11,000 GitHub stars, installs in a minute, and the difference shows up most on complex tasks like data analysis and content briefs.
Profit Bid. POAS bidding for Shopify: it computes real per-order profit after COGS, fees, and VAT, then feeds profit-weighted conversions back to Google, Meta, and TikTok so the algorithms optimize for margin instead of revenue. From $14.99 a month.
ShieldFlow. Shopify fraud protection that blocks card testing and bot orders before checkout completes rather than flagging them after. Free for the first 100 checkouts a month, so you can test it on live traffic before paying anything.
NotientAI. Social proof notifications, live visitor counts, and low-stock urgency from one script tag, localized to the visitor's region. From $9 a month, and works on Shopify without an app install.
Proofling. An AI agent that asks customers for testimonials and video reviews after purchase, handles the follow-up and consent, and publishes to an embeddable wall. Five free asks to test it. No native Shopify integration yet, so it fits best if you already use Stripe somewhere in the stack.
Anthropic Academy. Anthropic's free self-paced courses with certificates, covering prompt engineering, agents, MCP, and Claude Code. The fastest way to train the person on your team who has become the unofficial AI lead.
Final Thoughts
Two books taught me more about selling than the rest of my shelf combined: How to Win Friends and Influence People, from 1936, and Good to Great, from 2001.
Carnegie's lesson fits in one line: people buy from people who take a genuine interest in them, and the best salesperson in the room is usually the best listener.
Collins spent five years studying why some companies pull ahead and stay ahead, and his least-quoted finding is the one I use most: the great ones never chase technology for its own sake. They used it to accelerate something they already did well.
Judge every tool in this edition by those two lessons. Does it help you listen to your customers, and does it help you act on what they tell you? If it does neither, skip it.
Until next time,
Warm regards,
Gentian