Diagnose the failure before choosing the fix
“Google Shopping is not performing” can describe four different failures: products are not eligible, eligible products are not receiving impressions, clicks are not turning into orders, or orders are not profitable. Those failures have different evidence and different remedies. Changing the ROAS target before identifying the stage is usually guesswork.
| Symptom | First evidence to inspect | Do not start with |
|---|---|---|
| Few products serving | Merchant Center status by country and destination | Bid increases |
| Approved but low impressions | Product report, titles, bids, budget and demand | Landing-page redesign |
| Clicks but no baskets | Search relevance, item ID, page speed, price and offer | More traffic |
| Baskets but few purchases | Checkout funnel, delivery cost, payment errors | Feed descriptions |
| Revenue but poor profit | Margin, returns, discounting and new-customer economics | Account-level ROAS alone |
1. Prove the measurement
Place a test order and trace it through the purchase event, Google Ads conversion action, analytics and order system. Verify transaction ID, currency, item value, tax and shipping treatment, duplicate prevention, consent behaviour and refunds. Compare at least a full week of Ads purchase value with backend gross revenue on the same attribution basis; document the expected difference rather than demanding an impossible exact match.
If purchase volume is being inflated by duplicate events or primary conversions include add-to-basket, any bidding diagnosis is contaminated. Fix that before deciding that Shopping needs a different structure.
2. Check eligibility and source freshness
Open the Needs attention view in Merchant Center, turn off prioritised fixes if you need the complete list, and download affected items. Google distinguishes products that are approved, limited and not approved; its product visibility and status guide explains those states.
Measure eligibility against the products that matter, not only total SKU count. If 98% of items are approved but the 2% not approved generated 35% of last month’s revenue, this is the first problem. Check country and marketing method because the same item can have different status across destinations.
Then inspect the last successful data-source update and a sample of volatile SKUs. Price and availability must agree between product data, landing page, structured data and checkout. Automatic item updates can correct temporary discrepancies, but frequent mismatch is a source-system problem.
3. Separate demand from ranking and allocation
An approved product may receive few impressions because search demand is small, the submitted data is vague, bids or targets constrain entry, or another campaign is serving the inventory. Use item-level impressions over a meaningful period and compare with seasonality, category demand and previous periods.
Inspect titles as product data, not ad slogans. Google permits up to 150 characters for the title attribute, but visible titles may be truncated. Put the terms that identify the product and variant early. For a running shoe, “Acme TrailPro Women’s Waterproof Trail Running Shoes, Navy, UK 6” is useful; “Amazing Acme TrailPro Shoes – Free Delivery” introduces promotional text and weakens specificity.
Next, inspect campaign inventory. In Standard Shopping, confirm the item is included in an active product group and that “Everything else” has not been accidentally excluded. In Performance Max, inspect listing groups and the Products report. If Standard Shopping and PMax both target the item, record which campaign received the impressions before changing bids.
Shopping spend moving but sales are not?
Upscale can trace the failure from product eligibility through to checkout and give you an ordered fix list.
Book My Free Ad Audit4. Find where clicks stop
Build a product-level table with impressions, clicks, cost, product views, add-to-baskets, checkouts and purchases. Use rates between stages. This prevents a high-volume category from hiding a specific leak.
Consider a hypothetical 30-day comparison:
| Product group | Clicks | Add-to-basket | Purchases | Likely investigation |
|---|---|---|---|---|
| Dining chairs | 1,000 | 90 (9%) | 27 (2.7%) | Traffic is plausible; inspect checkout and delivery |
| Desk lamps | 800 | 16 (2%) | 8 (1%) | Search relevance, price and product page |
| Replacement shades | 240 | 48 (20%) | 4 (1.7%) | Compatibility detail or checkout friction |
For low add-to-basket rate, compare the query intent, submitted variant, primary image, advertised price and first mobile viewport. For a strong basket rate but weak purchase rate, test delivery cost, delivery date, payment failures, coupon-field distraction and out-of-stock variants. Do not call every post-click failure “CRO”. Name the stage.
5. Test price and offer competitiveness
Shopping exposes product, price and image before the click. A weak CTR can be a relevance issue, but it may also reflect an uncompetitive visible offer. Compare like-for-like products: model, pack size, condition, delivery charge and delivery speed. Cutting bids cannot make an overpriced commodity more attractive.
Where the retailer cannot change price, the next question is whether the page communicates a defensible difference: warranty, bundle, authorised-retailer status, returns, stock depth or delivery promise. Those claims must be true and visible. Do not put promotional claims into product titles contrary to Google’s editorial requirements.
6. Stop waste with a decision window
“Spent with no sales” is incomplete without a threshold. Choose a decision window based on conversion lag and expected cost per order. If the allowable acquisition cost for a group is £35, an item with £8 spend and no order has not failed. At £140 with enough clicks and no meaningful basket activity, it deserves intervention.
Possible actions are not limited to exclusion. Correct the title if queries are wrong; isolate the product if it needs a different target; fix a broken variant landing page; lower exposure if economics are weak; or exclude when no credible fix remains. Record the reason so an automated rule does not reintroduce it later.
7. Change one layer and measure it
For each fix, write the hypothesis and primary metric. A title test might use item-level CTR and qualified-session rate while holding bids and price stable. A checkout change might use purchase rate from checkout, payment errors and net revenue. A target change should be judged after allowing for conversion lag, with spend, volume and contribution reported together.
Review results by item or a pre-defined group, not only campaign. Keep a change log with implementation date, affected IDs and confounding events such as a promotion or stock outage. Without that, ordinary volatility gets credited to the last person who touched the account.
Worked diagnosis: from symptom to action
Take a hypothetical cookware retailer whose Shopping revenue falls 30% week on week. Account-level ROAS also falls, so the initial suggestion is to reduce spend. A staged diagnosis produces a different answer.
- Tracking reconciliation shows Ads and backend orders moving together, so the decline is unlikely to be a tag artefact.
- Merchant Center approval is stable overall, but revenue-weighted coverage falls from 96% to 71% because a price-mismatch issue affects the best-selling pan sets.
- The product source updates nightly while a promotion engine changes sale prices at noon.
- Unaffected products retain normal impression and conversion rates.
The priority is to synchronise promotional price data and verify the affected variants, not to lower bids across the account. The team should restore accurate source data, inspect processed prices, verify page and checkout, and follow the issue-specific review path if Merchant Center requires it. The success metrics are recovered revenue-weighted eligibility and impressions on the affected item cohort.
Thresholds that prevent premature decisions
Define the minimum evidence for each diagnosis. For item-level waste, a useful threshold can be spend as a multiple of allowable acquisition cost after normal conversion lag. For landing-page diagnosis, require enough qualified sessions to separate a real rate difference from a handful of visits. For a title test, require meaningful impressions before reading CTR.
Example: a category has a £40 allowable acquisition cost and 90% of purchases arrive within seven days of the click. An item with £25 spend and no purchase is still inside one allowable acquisition cost. At £130 spend, no baskets and more than seven days since the last click, the evidence for intervention is much stronger. The team might correct query relevance, isolate the item or pause it; the threshold does not dictate which action is right.
Failure cases that mimic poor Shopping performance
- Promotional comparison: this month appears weak only because the comparison period contained a site-wide sale.
- Attribution change: conversion settings or consent implementation moved, changing reported value without the same change in backend sales.
- Stock substitution: high-converting variants sold out and spend shifted to weaker alternatives.
- Returns lag: reported ROAS looks healthy before a high-return category is adjusted to net sales.
- Campaign migration: product IDs reset or inventory groups changed, making a like-for-like item comparison invalid.
Use the Merchant Center disapproval workflow when eligibility changes, and the product-page CRO guide when the leak is demonstrably post-click.
End the diagnosis with a falsifiable statement: “Revenue fell because 63 high-volume variants lost eligibility after noon promotion prices stopped reaching the feed.” Name the evidence that would disprove it. This prevents a plausible story from hardening into a conclusion before the fix is measured.
Record whether the evidence confirmed the diagnosis, what changed in the affected cohort and which remaining uncertainty deserves the next test.