Shopify Turned UCP On For You: Now Manage Exclusions and Field Quality
Shopify Editions Summer ‘26 landed on June 17, 2026 with more than 150 updates. The one that changes your day-to-day is not a visible feature at all. It is a default: Universal Commerce Protocol is now switched on for every store, and eligible products syndicate to ChatGPT, Microsoft Copilot, Google AI Mode, Gemini and the Shop app without anyone touching a setting.
That retires the question most merchants spent the last two quarters debating. You are not deciding whether to enter agentic commerce anymore. You are already in it, and the work in front of you runs the other direction: deciding what to pull back out, and making sure whatever stays in is described well enough for a model to sell it.
What “on by default” actually changed
Getting into an AI shopping surface used to be an application process. You built a feed, submitted it, waited for review, sometimes signed a separate agreement. That process is gone, replaced by an automatic eligibility check that runs against your catalog whether you look at it or not.
A product generally counts as eligible when several things are true at once: it is published to the online sales channel, it has a price, it has sellable inventory, shipping covers the market the surface serves, and it does not fall into a restricted category. Drafts, archived items, out-of-stock variants, customer-group-gated products, and regulated categories such as alcohol, weapons, adult goods and anything requiring a prescription or age check get filtered on the platform side. You do not need to build that logic yourself.
So the first move is not a settings change, it is an inventory. Open admin, count how many SKUs are currently syndicating, and compare that against your own mental list of products you are happy to sell anywhere. Most stores find a gap of a few dozen items sitting in the channel that nobody put there on purpose.
What actually shifted is the direction of the mistake. Under the old model, forgetting a SKU cost you some impressions. Under the new one, forgetting a SKU can cost you a sale you never wanted: margin swallowed by referral fees, an order shipped into a market where you lack certification, or a pair of shoes that was always going to come back.
One more thing from the same release, since it shares your product data: Summer ‘26 also shipped native AI merchandising, with collections that sort by real-time conversion probability and predictive cross-sell blocks on product pages. Those are on-site features and technically a separate track from UCP, but they read the same catalog. Thin fields degrade both at once.
Products you probably do not want an agent selling
Start with thin-margin items, because they are the easiest to miss. Agentic orders typically carry a referral or commission cost on top of a return rate that already runs higher than your own site. Anything under roughly 15 percent gross margin can lose money per order. On your own storefront those SKUs earn their place as traffic drivers that pay off through attach rate and repeat purchase, and an agent does not run your attach-rate playbook.
Region-restricted products are the second group. Distribution agreements covering only certain countries, certifications valid in one market but not another, lithium battery shipping limits. You know these constraints, but if the only place they appear is a line of copy in your product description, no machine reads it, and the assistant will happily recommend the item to a buyer in a market you cannot serve.
Third: apparel and footwear that require size selection. On your site people open the size chart and read the review that says it runs small. In a chat interface they order after two exchanges, and returns climb accordingly. Fourth: bundles and multi-packs, where a three-item total price gets compared against single units and looks absurd. Then there is made-to-order, preorder, and anything with a lead time past three weeks, since product cards rarely communicate lead time at all.
| Product type | Recommendation | Reason |
|---|---|---|
| Core products above 40 percent margin | Keep syndicated | Survives referral fees and return losses; this is the channel’s backbone |
| Traffic drivers under 15 percent margin | Exclude | Agents do not run your attach-rate strategy, so the loss-leader logic breaks |
| Region-restricted or certification-limited | Split visibility by market | Text constraints in descriptions are invisible to machines |
| Apparel and footwear needing size choice | Fix sizing fields first, then release | High returns here signal a data gap, not a bad channel |
| Bundles and multi-packs | Exclude | Unit-price comparison makes the total look wrong |
| Custom, preorder, lead time over 3 weeks | Exclude | Cards do not show lead time, so expectations break on arrival |
| Items needing installation or extra parts | Add compatibility fields before release | Buyers who did not know they needed a bracket leave one-star reviews |
| Perishable, cold chain, short shelf life | Exclude | Delivery windows and temperature handling do not fit in a product card |
| SKUs with historical return rate above 25 percent | Hold, fix fields, re-release | Use returns as a diagnostic for missing attributes |
Adjust these thresholds to your own cost structure, but keep one habit: treat high returns as a data problem before you treat them as a channel problem. Most of the time the expectation mismatch traces back to an attribute the model never received.
Doing the exclusions in admin
Exclusion works at three levels of granularity: the whole store, the sales channel, or the individual product. Turning UCP off store-wide is almost never the right call, since it forfeits a surface that is still growing. The useful work happens at the other two levels.
The common mistake at product level is unpublishing. That removes the item from your own storefront too, which is a self-inflicted wound. Use sales channel visibility instead so you can drop a product from the AI surface while it stays live and sellable on your site.
For bulk work, manage exclusions with tags plus an automated collection. Tag every excluded SKU with something consistent such as no-agent-channel, create an automated collection rule that gathers everything carrying it, and adjust channel visibility against the collection instead of clicking through products one by one. Then add a single checkpoint to your product launch process: margin, expected return rate, region restriction. Hit any one of them and the tag goes on before the product goes live.
Verify the exclusions rather than assuming they took effect. Two checks work: confirm the product count on that channel dropped by the number you expected, and a day or two later ask an assistant a brand-plus-category question and see whether the removed items still surface. Syndication and caching both lag, so same-day results prove nothing, but an excluded product still appearing a week later means you changed the wrong level.
Region restrictions belong in Shopify Markets product visibility, controlled market by market. Do not skip this and do not substitute description copy for it. UCP syndicates structured data, and the disclaimer you wrote in the body text is not part of that payload.
Field quality is now the only variable
When every store is switched on, being switched on stops differentiating anyone. The remaining variable is whether a model can answer a shopper’s actual question from your data. Products that produce a shrug get skipped in the recommendation.
Four attribute types carry most of the weight. Materials and composition need specifications rather than adjectives, so 100% merino wool, 18.5 micron earns its place while premium wool blend says nothing. Dimensions and weight should include measured values and a fit range, not just S/M/L. Use case needs concrete conditions: water temperature tolerance, indoor or outdoor, which devices or fittings it works with. Return policy has to exist in machine-readable form, covering the return window in days, who pays return shipping, and whether cross-border returns are accepted. A policy page in your footer does not count.
A fast self-check beats auditing a field checklist. Take the data one product syndicates, hand it to ChatGPT, and ask the three questions your buyers actually ask: can it go in the washing machine, what size for someone 175cm and 70kg, and what happens if it does not fit. Any vague answer points straight at the missing attribute type, because you just simulated exactly how the channel reads your catalog.
Multi-market stores need one extra pass over translations. Specifications lose precision in localization constantly, so 18.5 micron in the English source becomes a generic adjective in the German version, and German shoppers get syndicated data carrying no information. Spot-check each language for numbers, units and return windows against the original.
Prioritize by sales volume. Fix your top 20 SKUs completely before touching the long tail. Default syndication does not spread impressions evenly, and the products already earning AI visibility are where better fields convert into revenue first.
The attribution gap after the sale
Orders arrive and you often cannot tell where they came from. Checkout completed on the agent side never touches your landing page, UTM parameters do not survive the trip, referrer is frequently empty, and several AI surfaces tend to collapse into one bucket in reporting. This is not a tracking implementation you got wrong. It is a structural gap in how the channel works.
What you can capture is order-level identification. Check whether orders arriving through UCP carry a source value that separates them from on-site orders, and if they do, pull them into their own report tracking three numbers: order count, average order value, and return rate. Return rate is the diagnostic one. When it runs meaningfully above your on-site baseline, you are almost certainly looking at an attribute gap creating an expectation mismatch, and the fix is better data, not a closed channel.
Expect this report to disagree with GA4, and expect that to be correct behavior. GA4 counts sessions that reached your site, and an agent-side checkout never created one. When the two conflict, trust the order backend and skip the reconciliation project, because you are chasing a data source that structurally cannot see these orders.
The harder gap is customer ownership. Buyers who convert on the agent side may never grant marketing consent, which keeps them out of your email flows and leaves repeat purchase entirely dependent on whether the platform surfaces you again. The practical countermeasure is a clear subscribe path inside transactional messages such as order confirmation, converting a one-time order into a contactable customer.
A reasonable sequence: this week, build a SKU sheet flagging exclusion candidates by margin and return rate. Next week, implement the exclusions with tags and an automated collection, and configure region restrictions in Markets while you are there. The week after, run the three-question test against your top 10 products and close the field gaps it exposes. The default is not yours to change, but the exclusion tag and the return-window field both are, and those two decide most of what happens next.
Related Articles
AI Mode Shows 95% Fewer Products Than Standard Search
A July 2026 Productrise study tracked 21 days of US and UK search results across 100,000 SERPs and 2 million listings. Product visibility rate dropped from 88% to 23%, listings per page fell from 22.5 to 4.3, and overlap between the two interfaces was 0.8%. Here is what actually moves the needle on those remaining four slots.
Amazon's 75-Character Title Limit: Restructure Your Listings Before AI Does It for You
Amazon begins enforcing a 75-character title limit on July 27, 2026, and over-limit titles reportedly get rewritten by AI. Here is the title formula that works, what to move into the new Item Highlights field, and a five-step compliance audit.
Bloomreach Sidekick for Loomi on Shopify: Diagnose Why Products Rank Where They Do in Site Search
Bloomreach launched the Sidekick extension for Loomi for Shopify on June 17, 2026. It gives merchants query-level ranking diagnostics inside Shopify admin: for any search term, you can see which boosts, merchandising rules, and spelling corrections shaped the results. Here is a practical 10-query workflow for using it, plus an honest take on when native Shopify search is still enough.