In partnership with
Happy Monday! Amazon brands walk into Prime Day with more inventory, better conversion, and a familiar leak: the SKUs that matter most still disappear.
News
🛒 Prime Day exposes the empty shelf
🧭 Pinterest makes discovery agent-readable
📈 Similarweb catches AI demand leakage
Insights
🛒 ChatGPT's shelf keeps moving
🛍️ TikTok Shop sells personal change
🧭 Agents need cleaner catalogs
LET'S ENTER THE MAZE!
The Maze: Amazon brands enter Prime Day with better ROAS, higher conversion, and more inventory. The leak is top ASINs going out of stock.
CommerceIQ data shows revenue lost to out-of-stock events rose about 24% year over year, even as on-hand inventory increased about 12%.
Amazon Sponsored Products ads appear only when advertised items are in stock, so unavailable hero products can waste high-intent media momentum.
Prime Day 2026 runs June 23-26 as Amazon pushes groceries, essentials, and repeat orders while rival retailers chase the same shopping window.
Why it matters: Total inventory is the wrong comfort metric. Prime Day execution now depends on keeping the few ASINs that carry the revenue in stock.
The Maze: Pinterest is turning visual discovery into an AI workflow layer for advertisers, partner copilots, and shoppers before product choice moves elsewhere.
Pinterest introduced Business Assistant, Pinterest MCP, an updated Performance+ creative model, and the experimental Ask Pinterest app ahead of Cannes Lions 2026.
The MCP server uses a standard connection layer so partner tools can query campaign, analytics, keyword, and taste-signal data without bespoke exports.
Performance+ creative is globally available, while Ask Pinterest is a limited US rollout; the model lifted clicks 7.5% versus the prior single-variant approach.
Why it matters: If Pinterest makes taste agent-readable, merchants may optimize for the AI systems choosing product shortlists, not only for feeds and search pages.
The Maze: Similarweb traced ChatGPT brand recommendations to later visits. The catch: analytics often credits the search box, not the AI answer.
Recommended brands were 2.5x likelier to get a site visit within seven days than non-recommended rivals.
Search captured 55.9% of AI-influenced visits, while visible AI referrals were only 8.8%.
GA4 AI chatbot channels can catch referrers, but delayed branded search still slips into the wrong bucket.
Why it matters: AI discovery is becoming a demand signal. Commerce teams need to defend branded search and measure recommendation share before budgets follow bad attribution.
The Maze: ChatGPT search does not simply update shopping answers. In a 20,000-response test, it changed 80.2% of recommended products.
Visibility Labs ran 1,000 product prompts 10 times with search on and 10 times off, then canonicalized product names to compare overlap.
The broad overlap sat near 20%, but products recommended 100% of the time without search overlapped only 15.8% when search was enabled.
That makes cited pages, reviews, category lists, and product explainers a new shelf, because model memory alone does not guarantee placement.
Why it matters: AI shopping visibility is not just SEO with a chatbot wrapper. Brands need to win the sources assistants retrieve, not only the model's memory.
The Maze: TikTok Shop's U.S. winners are body-facing: products shoppers wear, apply, or consume, then buy because a creator can prove the promise quickly.
Among the top 1,000 U.S. products, more than two-thirds of tracked revenue comes from body-facing products, led by Beauty at about 30%, Health near 20%, and Apparel around 17%.
The price band is built for feed-native impulse: the median top product is about $40, and roughly two-thirds are under $50, low enough to buy before search takes over.
The same concentration logic appears in sellers: the top 1% drive 60% of tracked U.S. GMV, so democratized discovery still rewards operators who can repeat content, affiliate, and inventory execution.
Why it matters: TikTok Shop is not Amazon with better video. It is a demand engine for demonstrable personal change. Product fit beats channel enthusiasm.
The Maze: Agentic commerce gives brands four routes into the answer. The cheapest route is not an app. It is product data clean enough for agents to trust.
Native app lanes favor Walmart, Target, Sephora, and ASOS-style builds, but they need inventory, fulfillment, and category experiences inside the agent surface.
Discovery-feed lanes make catalog quality the shelf: OpenAI's commerce docs say ChatGPT ingests structured data to understand inventory and surface products in context.
Spec-driven queries reward attributes like price, variants, dimensions, and availability; browse queries reward richer descriptions, taxonomy, and use-case metadata.
Why it matters: Retailers optimized pages for humans. Agents read feeds. Bad product data is no longer a filter problem. It is invisibility.
Everything else you should know about
📊 Google and Microsoft redraw measurement
Google is routing Shopify purchase data into GA4, which gives merchants cleaner platform reporting while leaving attribution gaps for real shopping journeys.
Google is folding standalone Display campaigns into Demand Gen, moving old display budgets deeper into AI-led creative and audience systems.
Microsoft unveiled Web IQ, Clarity citation reporting, and an MCP server, making AI workflows more measurable for advertisers and commerce teams.
🛍️ Retail media keeps standardizing
Tesco, B&Q, John Lewis, and Waitrose joined Dunnhumby retail-media pilots, pushing UK retailers toward shared measurement and advertiser access.
🤖 Agentic commerce wants open rails
A Universal Commerce Protocol proposal outlined open rails for agentic shopping, shifting some control from storefronts to protocols and assistants.
🌏 Marketplaces stretch the shelf
THAT’S IT FOR TODAY!
You’re the reason our team spends hundreds of hours every week researching and writing this email.
What do you think of this issue?
See you next time in the maze!
MarketMaze team
