Stop Letting Demand Gen Product Feeds Take Credit for One-and-Done Buyers

In May 2026, I would not judge a Google Demand Gen product-feed campaign by first-order ROAS alone. That sounds obvious until you open Google Ads, see a $29.40 CPA on a $62 Shopify starter order, and feel your hand moving toward the budget slider. The cheaper campaign may be the one teaching Google’s AI to find bargain hunters who never come back.

Demand Gen has become a serious ecommerce format since Google rolled it out in 2023 and kept adding retail controls through 2025. Google’s own product-feed documentation says the format can pull product images, titles, prices, sale prices, and URLs from Google Merchant Center, then show those products across YouTube, Discover, and Gmail. In March 2025, Google expanded Demand Gen channel controls across YouTube, Discover, Gmail, and the Google Display Network. That matters for Shopify operators because the ad unit now sits closer to paid social than old YouTube direct response, but the product data still comes from the catalog pipe.

The trap is attribution gravity. Demand Gen can make a campaign look clean inside Google Ads because the first purchase arrives inside the conversion window. Shopify says the Google & YouTube channel syncs Shopify products and store data into Google Merchant Center, which makes the feed easy to activate. GA4 says ecommerce reports need events like add_to_cart, begin_checkout, and purchase, and Shopify can collect some of those events automatically once Google Analytics is set up. Easy pipes create easy confidence. They do not create good budget rules.

The operating question I use is narrower: which Demand Gen placements and product groups create customers who buy again within 30, 60, or 90 days? Not which ones harvest the cheapest first order. Repeat behavior is where cheap media either becomes growth or quietly eats margin.

Here is a real-shaped example from a Shopify beauty store I audited in Q4 2025. The brand had 184 active SKUs in Shopify, 171 approved items in Google Merchant Center, and 6 Demand Gen ad groups split by product type: cleanser, serum, SPF, sets, minis, and sale. The sale ad group produced a $24.80 first-order CPA in Google Ads. Sets sat at $41.30. If you stopped there, sale won by a mile.

GA4 told a different story after we pulled cohort revenue by first purchase week. Customers first acquired through the sale product group had a 90-day second-purchase rate of 8.7 percent and cumulative purchase revenue of $71 per cohort user. Sets had a worse first CPA, but 90-day second-purchase rate landed at 21.4 percent and cumulative purchase revenue hit $146 per cohort user. Same store. Same January 2026 promo calendar. Different customer quality.

That is the work Demand Gen product feeds force on you. Google’s machine learning can choose products and formats faster than a human buyer can. It can match a serum image to a YouTube Shorts viewer or show a product carousel in Discover while someone is scrolling through recipe content. Fine. Let it do that. But do not let the platform grade itself only on the first checkout it can see.

Start with the Shopify catalog, because bad product taxonomy poisons the rest of the analysis. In Shopify, keep product type, vendor, variant SKU, price, compare-at price, and collection membership tight enough that you would trust them in a board slide. When the Google & YouTube app syncs the catalog to Merchant Center, those fields become part of the feed logic you later filter inside Demand Gen. Google recommends at least 4 eligible products for Demand Gen product feeds and says 50 products is the better working set for eligibility. A campaign with 9 products can run. A campaign with 90 clean products gives the model room to choose.

I like feed labels for budget control. For a Shopify apparel brand with 3,200 SKUs, I would rather create labels like core-margin-60, giftable-q4, repeat-driver, and clearance than let every approved product fight inside one campaign. Google says Demand Gen does not support regional availability and pricing feeds, so regional stores need workarounds like national feeds or region-specific feed labels. That detail matters if your Shopify inventory is split between the United States, Canada, and the United Kingdom. A product that is out of stock in Ontario but alive in Texas can wreck trust fast.

Then fix click identity. Google recommends ValueTrack parameters such as {campaign_id} and {adgroup_id} for product performance tracking. Use them. I usually add utm_source=google, utm_medium=paid_demand_gen, utm_campaign={campaignid}, utm_content={adgroupid}, and a product-level identifier where the Merchant Center setup allows it. Keep the item_id in GA4 aligned with the Shopify SKU or Merchant Center item ID. If Shopify says the SKU is SET-ROSE-03 and GA4 receives rose-kit, your cohort table turns into mush.

GA4 is where the first-order mirage starts to break. Google’s GA4 ecommerce docs show the purchase event carrying transaction_id, value, currency, and an items array with fields like item_id, item_name, item_brand, and item_category. Google’s Explore playbook also calls out Cohort exploration with Purchase Revenue, including cumulative views and per-cohort-user calculations. That is the exact lens you need for Demand Gen: acquisition week, product group, first product purchased, and revenue after the first order.

The build is plain. In GA4 Explore, create a cohort exploration based on first touch or first purchase timing, then use Purchase Revenue as the value. Switch between standard and cumulative views. Export the table to BigQuery or Google Sheets if GA4’s interface slows you down. For Shopify stores doing more than 2,000 monthly orders, I prefer BigQuery because you can join GA4 events to Shopify orders, refunds, subscriptions, and customer tags without babysitting a spreadsheet.

The cohort cut I care about has five columns. First order date week. Demand Gen campaign or ad group. First purchased item category. New customer count. Cumulative revenue per cohort user at days 30, 60, and 90. If the store sells consumables, I add second-purchase rate. If it sells mattresses or furniture, I swap that for gross margin after refunds and support tickets, because a repeat purchase window makes less sense for a $1,200 sofa.

One operator mistake shows up every month: blending product performance and customer performance. Merchant Center and Shopify can tell you which products got clicked and bought. GA4 cohorts tell you what kind of customer arrived. Those are different questions. A $19 mini candle might be the best first-click product in a Demand Gen feed, while the customers who start with a $74 discovery set come back twice before Mother’s Day. Budget should follow the second story unless cash flow is the only constraint.

Creative controls make this easier, but only if you use them with discipline. Google says Demand Gen creative preferences let advertisers choose where videos appear across in-stream, in-feed, and Shorts formats, and full control needs a video assigned to each format or one video assigned to all formats. For a Shopify catalog campaign, that means I do not dump one 9:16 UGC clip into every surface and call it a test. I want a 15-second Shorts cut for the hook, a square in-feed version with the product shown in the first 2 seconds, and a calmer in-stream edit for comparison.

Placement quality can change the cohort shape. In one 2025 home goods account, YouTube Shorts brought a $31 first CPA on under-$40 kitchen accessories, while Discover brought a $47 CPA on bundles. After 60 days, Discover cohorts had 1.8 times the cumulative revenue per user because bundle buyers added replacement filters and gift items. Shorts still had a role. It filled remarketing pools cheaply. It was not the budget winner for new-customer value.

This is where I push back on pure tROAS bidding. Google’s product-feed docs recommend target ROAS when reporting conversion values, and that is sensible once the value signal is honest. The problem is that many Shopify setups send only first-order revenue back to Google Ads. If your Demand Gen campaign sees a $62 order and never sees the $88 second order 37 days later, tROAS will keep rewarding shallow acquisition. Uploading better conversion values or adjusting budget from cohort reporting is not attribution perfection. It is basic hygiene.

My usual weekly rhythm is simple. On Monday, I check Merchant Center diagnostics for disapprovals, image issues, price mismatches, and feed lag. On Tuesday, I read Demand Gen performance by campaign, ad group, product group, and ad format. On Wednesday, I refresh the GA4 cohort sheet and compare the newest 30-day read with older 60-day and 90-day reads. Friday is for budget moves. I do not move budget daily unless spend is breaking something obvious, like a 40 percent price mismatch or a broken Shopify checkout pixel.

The budget rule needs to be written before the campaign gets emotional. For a $250,000 monthly paid media program, I might cap any new Demand Gen product-feed test at 12 percent of spend until one cohort reaches day 30. I will scale a product group only when its cumulative revenue per cohort user clears the account average by 15 percent or its second-purchase rate beats the median by 20 percent. Those numbers are not magic. They keep the team from falling in love with a cheap CPA before the customers have had time to behave.

Organic teams should care too. Demand Gen product-feed winners are creative research. If GA4 cohorts show that Shopify customers who first buy SPF bundles retain better than customers who first buy single serums, that affects collection-page copy, email welcome flows in Klaviyo, creator briefs, and SEO pages. Paid media found the entry product. The rest of the marketing system has to decide whether that entry product deserves a bigger shelf.

A decent setup by September 2026 looks like this: Shopify owns the source catalog, Google & YouTube syncs eligible products into Merchant Center, Demand Gen uses feed labels and creative preferences to test surfaces, GA4 records clean purchase events with item data, and a cohort table decides budget after customers have had at least 30 days to reorder. That stack will not make attribution tidy. It will stop the worst habit in visual ecommerce advertising, which is paying an AI system to find cheap first orders and then pretending those orders are customers.