Feed strategy starts with a specific constraint
Product feed optimisation is a programme of controlled changes to eligibility, product understanding and commercial segmentation. It is not a one-off rewrite of every title. The right strategy depends on the constraint: missing products, stale offers, weak matching, poor variant data or no way to separate inventory by economics.
Establish a baseline before selecting a tactic: active catalogue count, revenue-weighted approved coverage, products receiving impressions, item-level CTR and conversion rate, spend beyond the decision threshold without a purchase, and contribution by product group.
Strategy 1: recover lost coverage
Export Merchant Center issues and join them to product value. Correct account-level or policy risk first, followed by disapprovals affecting high-priority products. Fix the source system, resubmit, inspect the processed value and verify the item is eligible in the intended country and marketing method.
Use the Needs attention workflow rather than counting only visible examples. The measurement is approved revenue coverage and recovered impressions, not the number of warnings closed.
Strategy 2: increase offer freshness
Map the real rate of price and stock change against the data-source schedule. If stock turns hourly, a nightly file is structurally stale. Increase update frequency or use an API-based route, then align product structured data on the page.
Google’s automatic item updates can correct some temporary price, sale-price, availability and condition discrepancies. Treat the number of automatic corrections as a diagnostic: repeated corrections mean the primary submission needs work.
Strategy 3: engineer category-specific titles
Build title rules from reliable structured attributes and real query language. Use different orders for footwear, appliances and furniture because shoppers distinguish them differently. Stay within the 1–150 character title limit, front-load product identity and avoid promotion or repetition.
Prioritise products with enough impressions to evaluate, weak relevance or vague source names. Test a fixed cohort while holding bids, images, price and promotion stable. More impressions alone are not success if post-click quality falls.
Need a feed roadmap, not a generic cleanup?
Upscale sizes the opportunity, chooses the smallest useful cohort and measures changes against commercial outcomes.
Book My Free Ad AuditStrategy 4: strengthen variant integrity
Give every sellable variant a stable unique ID; group siblings with item_group_id; populate relevant variant attributes; and align title, image, price, availability and selected landing-page option. This is especially important where a parent page contains dozens of combinations.
Create a variant-family QA report. Flag shared IDs, missing group IDs, identical images across visually different colours, parent-level stock, and prices that do not match the selected variant. Measure reduced mismatches and improved eligible coverage before looking for a media effect.
Strategy 5: improve product identity
Collect manufacturer-assigned brand, GTIN and MPN accurately. Do not manufacture identifiers. Strong identity data helps Google understand the offer and compare it to the correct product, but the requirement depends on the product; consult the product data specification.
Prioritise high-volume branded categories and product families with identifier warnings. Audit the source of each identifier and preserve leading zeros in GTINs.
Strategy 6: build a commercial taxonomy
Use product type for a clean retailer-defined hierarchy and Google product category for Google’s taxonomy. Then use custom labels for states not represented by the product itself: margin, stock cover, lifecycle or experiment cohort.
For a hypothetical retailer, custom_label_0 might hold margin_low, margin_mid or margin_high; custom_label_1 might hold stock bands. The definitions and thresholds belong to the business. Google allows five custom-label fields, one 1–100-character value per field per product, and up to 1,000 unique values per field account-wide.
Strategy 7: improve images as a cohort
Separate compliance from creative quality. First remove blocked URLs, placeholders, promotional overlays and wrong-variant images. Then test clearer crops, higher-quality product views and useful additional images within a defined category.
Keep image URLs stable until the asset changes. Measure eligible products and image issues for compliance work; use CTR and post-click quality for a controlled visual test. Changing title and image simultaneously prevents attribution.
Size the opportunity before building rules
| Opportunity | Simple sizing method | Primary success metric |
|---|---|---|
| Disapproved hero range | Recent net revenue of affected IDs | Revenue-weighted approval |
| Weak titles | Impressions on vague-title cohort | CTR plus qualified conversion |
| Stale stock | Mismatches and cancelled orders | Recurrence and cancellation rate |
| No margin control | Spend across margin bands | Contribution after ads |
| Poor variants | Affected variant revenue and issues | Eligibility and variant conversion |
This prevents a six-week metadata project on products that have no demand while a source-sync failure affects best sellers.
Run feed tests like product releases
- Write the constraint, hypothesis and affected item IDs.
- Define a holdout or stable pre-change baseline.
- Diff submitted values and inspect processed values.
- Monitor eligibility after processing.
- Wait for an appropriate conversion window.
- Report volume, rates and commercial value together.
A hypothetical title cohort receives 100,000 impressions before and 118,000 after; CTR rises from 0.8% to 1.0%, but conversion rate falls from 2.5% to 1.6%. Clicks rise from 800 to 1,180 while orders fall from 20 to about 19. The title produced more traffic, not a meaningful commercial gain. That is why the measurement plan must extend beyond CTR.
A 90-day feed roadmap
Weeks 1–2: baseline the catalogue, fix enforcement and measurement risks, document source ownership. Weeks 3–6: repair price, availability, identifiers and variants in high-value cohorts. Weeks 7–10: run one title or image test and implement commercial labels. Weeks 11–13: review outcomes, automate QA and choose the next constraint.
The sequence is deliberately narrow. A catalogue-wide rewrite creates too many simultaneous changes and makes it difficult to learn which intervention mattered.
Strategy selection by evidence
| Evidence | Likely strategy | Required guardrail |
|---|---|---|
| High-value items not approved | Coverage recovery | No review request before live verification |
| Frequent price corrections | Freshness and structured-data repair | Automatic updates are not the source of truth |
| High impressions, weak relevance | Title/taxonomy cohort test | Hold price, bids and image stable |
| Variant mismatch issues | Variant integrity programme | Sample complete families |
| Blended ROAS hides margin | Commercial labels and allocation | Finance-approved definitions |
Choose the strategy with the strongest evidence and commercial exposure. Do not pick title optimisation merely because it is easy to demonstrate in a before-and-after screenshot.
Worked strategy prioritisation
A hypothetical fashion retailer has three candidate projects. Project A would rewrite 4,000 low-impression descriptions. Project B would repair missing size and colour attributes across 600 variants that generate £200,000 quarterly revenue. Project C would add margin labels to 2,000 products accounting for 80% of ad spend.
The team scores expected commercial impact, evidence confidence, effort and risk. Project A is low impact and hard to measure. Project B has direct variant and eligibility evidence, so it starts first. Project C follows after finance confirms the contribution calculation and refresh process. The description project is narrowed to products where customer and search evidence shows missing information.
For Project B, acceptance means each variant has a unique ID, common group ID, accurate variant fields and a matching selected page state. Media evaluation begins only after processed values and eligibility are confirmed.
Failure cases in feed strategy
- Catalogue-wide simultaneous rewrite: too many variables change to attribute the result or roll back safely.
- Rules built from unreliable fields: polished titles amplify source errors.
- No holdout or item list: seasonality is credited to the feed change.
- CTR-only success: broader matching increases clicks but reduces orders or contribution.
- Labels without refresh ownership: campaigns make decisions on stale margin or stock.
- Approval treated as delivery: products become eligible but are excluded from active campaign inventory.
The measurement document
Before release, record the hypothesis, affected IDs, comparison cohort, old/new values, primary metric, guardrails, minimum observation window and known confounders. For a title strategy, the primary metric might be qualified orders per thousand impressions; guardrails include eligibility and contribution per click. For freshness, the primary metric might be mismatch recurrence with cancelled-order rate as a guardrail.
At review, report absolute and relative change, processing status and any catalogue movement. Keep the cohort fixed even if items later move campaigns. The feed implementation checklist provides the release controls behind this strategy.
Set a stopping rule before launch. Stop a test if priority products lose eligibility, mismatch issues appear, contribution deteriorates beyond an agreed guardrail or the transformation changes items outside the cohort. A feed strategy should be reversible; otherwise the business cannot distinguish experimentation from uncontrolled catalogue change.
After a successful test, expand in stages rather than copying the rule across the entire catalogue. Revalidate each new category because source attributes and buyer decision criteria differ. The winning furniture-title order may be wrong for electronics, and a stock label threshold suited to replenishable goods may be dangerous for made-to-order products.
Retain the original cohort definition through rollout so later campaign movement does not rewrite the comparison. Archive old and new processed values, issue status and the commercial review alongside the rule version.
Name the owner of the next review.