Shopping feed optimisation is the continuous practice of fixing your product data, titles, images and attributes so Google’s AI systems can match your products to the right searches and bid on them intelligently. It is never a one‑off project. The first move for any account is a Merchant Center diagnostic against your top SKUs by revenue, sorted by impressions affected, so you know exactly which fixes will move the needle before you touch a single title.
TL;DR:
- Prioritize diagnosing issues based on impressions affected and revenue contribution, especially focusing on missing key attributes like GTIN, item_group_id, and title mismatches.
- Rewrite titles using structured templates that incorporate brand, product type, and key modifiers, ensuring they match shopper search queries and avoid spammy keyword stuffing.
- Fill primary attributes such as GTIN, MPN, brand, and product category accurately, and use supplemental feeds for backfilling data, to ensure eligibility and competitiveness in the auction.
- Fix image disapprovals by using high-resolution, plain backgrounds, removing watermarks, and automating bulk quality checks for large catalogues.
- Conduct regular, scheduled audits and updates, with a clear owner, impact-effort tracking, and continuous improvement to prevent feed decay and maintain optimal Shopping and Performance Max results.
Table of Contents
- What is shopping feed optimisation and where do you start?
- How do you rewrite titles that actually convert?
- What makes a product description work for shoppers and AI?
- Which attributes do you need to fill first?
- What image specs actually stop disapprovals?
- How do you stop variants from colliding on price?
- How do custom labels feed your bidding strategy?
- How do AI tools fit safely into feed workflows?
- How does feed quality change Performance Max results?
- How often should you audit your feed?
- How Moor Marketing applies this playbook
- Why continuous feed discipline compounds growth
- Get a feed audit and 12‑week optimisation plan
- Sources
- FAQ
What is shopping feed optimisation and where do you start?
Start with a diagnostic, not a rewrite. Open Merchant Center Diagnostics and pull three data sets covering the last 30 to 90 days: impressions and CTR by product, item‑level disapprovals and warnings, and the search terms report from your Shopping campaigns. That third one gets skipped constantly, and it’s usually where the real story is.
Here’s the sequence that actually works:
- Export impressions and clicks per SKU for the trailing 30 to 90 days and rank by revenue contribution, not just traffic volume.
- Sort diagnostic issues by “impressions affected” rather than by error count. A missing GTIN on your top 20 sellers matters more than 200 warnings on dead stock.
- Cross‑reference search terms against product titles to spot mismatch. If shoppers are searching “waterproof hiking boots” and your title says “outdoor footwear,” that gap is costing you impressions.
- Verify conversion tracking on your top SKUs before you change anything. You cannot measure the impact of a title fix if the tracking underneath it is broken.
- Flag candidates for title, attribute and image fixes in a simple spreadsheet with owner and priority score.
Google’s own guidance backs this order: complete, accurate and enriched product data is what widens the surface area AI systems use for matching and bidding, so the diagnostic has to identify gaps before enrichment starts. Merchant Center disapprovals themselves cluster around a handful of causes, mostly missing item_group_id values, title mismatches against landing pages, and absent required attributes — which is exactly why the audit needs to check those three things first.
Pro Tip: Build a “top 50 by revenue” tab in your feed tracker and refresh it monthly. Every fix you make gets tested against this list first, because a 2% CTR lift on your bestseller is worth more than a 20% lift on a $12 SKU nobody searches for.
How do you rewrite titles that actually convert?
Titles remain the single highest‑leverage fix in the entire feed. Practical audits consistently find that title rewrites on top SKUs can multiply impressions and lift CTR when the new title matches how shoppers actually search, not how your product database happens to be structured.
The template that holds up across categories: Brand + Product Type + Key Modifiers (colour, size, material, model number), front‑loaded in that order because Google truncates and shoppers scan left to right. Aim for the first 70 characters to carry the full meaning, since that’s roughly what’s visible before truncation on mobile, even though the field itself allows up to 150.
A few rules worth locking in:
- Never hardcode a price, a promotional phrase like “sale” or “free shipping,” or a size/colour that changes per variant into the title. Keep those in their own attributes.
- Do include the modifier a shopper actually types, like “waterproof” or “cordless,” even if it feels repetitive with the product type.
- Don’t stuff keywords. A title with five modifiers reads as spam to both shoppers and Google’s matching systems.
- Keep brand name spelling and casing identical to your
brandattribute, every time.
For rollout, work in batches. Take your top 80 SKUs by revenue, draft new titles in a spreadsheet next to the originals so you can diff them, push through a supplemental feed or the Content API rather than editing the primary feed directly, then hold for at least two full weeks before judging results. SKU Analyzer’s prioritisation framework puts titles ahead of descriptions and attribute enrichment for exactly this reason: it’s the fastest path to measurable movement.
Pro Tip: Test title changes on your top 20 to 80 SKUs before rolling a template across the whole catalogue. A framework that works for outdoor gear can flop for jewellery, so validate first.
What makes a product description work for shoppers and AI?
Google Shopping only shows a fraction of your description in most placements, and AI‑driven overviews pull from the opening text disproportionately. That means the first 160 characters do almost all the work.
Structure that opening the same way every time: state what the product is, answer the question the shopper is implicitly asking (does this fit my need?), then name the one differentiator that separates it from ten similar listings. A generic blender description that opens with brand history wastes the exact words that would have told a shopper it has a 2‑litre jug and a 10‑year warranty.
Beyond the opening line:
- Use
product_highlightslots for short spec bullets — capacity, material, dimensions, warranty length — since these often surface directly in Shopping results without the shopper clicking through. - Avoid promotional language (“best,” “amazing,” “guaranteed”) in both title and description; Google’s policy systems flag superlatives inconsistently, and it’s not worth the disapproval risk.
- Mirror your landing page copy where it matters for policy: if your feed claims a feature your product page doesn’t mention, that mismatch is a common trigger for suspension.
- Keep the tone plain. Shoppers scanning a grid of thumbnails read fast, and dense marketing copy simply gets skipped.
Descriptions are lower leverage than titles for CTR, but they’re higher leverage for the AI systems now generating shopping overviews, because those systems lean on structured, factual copy far more than persuasive copy.
Which attributes do you need to fill first?
Not every attribute in the Merchant Center schema carries equal weight. A handful control whether your product is even eligible to show; the rest shape how competitively it shows.
The non‑negotiable tier:
- GTIN — without it, many categories simply won’t serve your ads competitively, regardless of bid.
- MPN and brand — required together when GTIN is genuinely unavailable, and they must match your actual packaging.
- google_product_category — use the full five‑level taxonomy path, not just a top‑level guess. A vague category costs you relevance in Smart Bidding’s matching logic.
- item_group_id — mandatory the moment you have variants; more on this below.
The tier that shapes competitiveness rather than eligibility: colour, size, material and the product_highlight slots mentioned earlier. Seer Interactive’s audit checklist treats these secondary attributes as the difference between a feed that’s merely compliant and one that actually wins impressions against comparable listings.
Sourcing them is the practical bottleneck. Manufacturer spec sheets and barcode lookup services solve GTIN gaps for branded stock. For anything private label or bundled, a supplemental feed lets you override or backfill attributes without touching your primary product data pipeline, which matters if that pipeline is owned by a warehouse system you don’t control.
What image specs actually stop disapprovals?
Image issues cause a disproportionate share of avoidable disapprovals, and most of them are fixable in an afternoon.
- Use a minimum 800×800 pixel image for non‑apparel and 250×250 as an absolute floor Google will accept, though anything under 800px looks poor in Shopping’s larger card formats.
- Keep backgrounds plain white or a neutral solid colour for the primary
image_link; save lifestyle shots foradditional_image_link, where Google allows up to ten extra images per product. - Strip watermarks, promotional badges, and border text from the primary image entirely, since these are among the fastest disapproval triggers.
- Never use a placeholder or “coming soon” graphic. If stock photography isn’t ready, delay the listing rather than submit a blank.
Bulk QC is worth automating once your catalogue passes a few hundred SKUs. A script that checks image dimensions, flags broken links, and confirms no two products share an identical image URL will catch the errors a manual spot check misses. Tools built for browser‑based data extraction can pull and validate image metadata across a large catalogue faster than checking listings one by one in Merchant Center.
How do you stop variants from colliding on price?
Missing or inconsistent item_group_id values are one of the most common causes of Merchant Center disapprovals, because Google reads mismatched variants as conflicting data about the same product rather than as legitimate size or colour options.
- Assign one item_group_id per parent product, using your Shopify product ID or an equivalent stable identifier from your PIM, never the individual variant SKU.
- Map every variant (each colour, size, style combination) to that same group ID, while keeping each variant’s own unique
idfield distinct. - Check for orphaned variants after any catalogue migration or bulk import, since these are the usual source of “pricing collision” warnings when two variants of the same product show different prices under what Google reads as separate listings.
- Re‑submit and monitor the Diagnostics tab for 48 to 72 hours after any bulk item_group_id change, because Google’s crawler needs a cycle to reprocess grouped items.
Get this wrong and the practical cost isn’t just disapprovals. It’s shoppers seeing a $40 version of a product ranked oddly against a $65 version because Google can’t tell they’re the same item in different sizes.
How do custom labels feed your bidding strategy?
Custom labels are the plainest, most underused lever in the whole feed. They’re free‑text fields with no fixed vocabulary, which means you can encode whatever your business actually cares about, then build Smart Bidding and Performance Max listing groups around it.
Practical label sets that hold up in real accounts:
margin_band(high/medium/low) so you can bid harder on the SKUs that protect profit, not just the ones that sell fastest.velocity_30dto flag products moving quickly versus sitting flat, feeding automatic bid adjustments before stock runs out.seasonorAOV_tierto separate impulse buys from considered purchases that need different messaging and budget.top_sellerandlow_stockas simple flags that let you pull bids down automatically before you oversell inventory you don’t have.
Calculating these isn’t complicated, but it needs a data source outside the feed itself, usually your order management or inventory system. The workable pattern is warehouse or ERP data feeding a supplemental feed on a daily refresh, so labels stay current without anyone manually tagging products. SKU Analyzer frames custom labels as the bridge between merchandising and finance data on one side and Smart Bidding logic on the other, and that’s the right way to think about them.
Pro Tip: *Start with just two labels, margin_band, and velocity_30d.
How do AI tools fit safely into feed workflows?
AI‑assisted title and description generation genuinely works at scale, but only inside guardrails. Generic prompts run against a whole catalogue risk hallucinated specs and dropped identifiers, which is worse than doing nothing.
The pattern that avoids that: build one few‑shot prompt template per Google Product Category rather than one universal prompt, because a template tuned for apparel will invent nonsense specs when pointed at electronics. Inside every prompt, lock GTIN, MPN and brand as immutable tokens the model is instructed never to rewrite or infer, only copy through.
- Generate new titles and descriptions in a draft feed or spreadsheet, never write directly to your live feed.
- Score and diff every output against the original before approval, flagging anything where a locked identifier changed.
- Bulk approve the clean diffs and export only those as a supplemental feed layered over your primary feed.
- Keep human review on any SKU where the diff touches price, availability, or category, since these carry policy risk if the model gets it wrong.
Google’s open‑source FeedGen pattern follows exactly this workflow: draft feed, score, diff, approve, export as supplemental. It’s a sound reference architecture whether you build custom tooling or adapt an existing one, and it’s the difference between AI that speeds up good work and AI that quietly breaks your catalogue.
Pro Tip: Never let an AI prompt see your whole catalogue at once. Category‑scoped batches of 50 to 100 SKUs keep the model’s context tight and make review manageable.
How does feed quality change Performance Max results?
Performance Max leans on your feed more than any campaign type before it, because the system uses your product data to decide which AI surfaces to enter, not just which keywords to bid on. Thin or generic feed data narrows that surface area; enriched data widens it.
- Run a controlled test on 20 to 80 top SKUs before rolling any title or description template across the catalogue, changing one variable at a time where possible.
- Hold for a full measurement window, generally two to four weeks, since Performance Max needs a learning period after any meaningful feed change before results stabilise.
- Track by custom_label tier, not just account‑wide averages, because a title change might lift your high‑margin tier while doing nothing for low‑velocity stock.
- Watch impressions, CTR, conversion rate and ROAS together, since a title that pulls more impressions but lower CTR usually signals a mismatch, not a win.
Attribution here is genuinely messier than in standard Shopping campaigns, because Performance Max blends channels. Treat directional movement over a full window as the signal, not day‑to‑day swings.
How often should you audit your feed?
Feed optimisation only compounds if it’s a schedule, not a project you finish once. A workable cadence: weekly diagnostics scans to catch new disapprovals early, monthly rollouts of title or attribute batches on your priority SKUs, and a quarterly deep audit that revisits categorisation, images and custom labels across the whole catalogue.
Prioritise with a simple impact‑versus‑effort score: impact as impressions affected multiplied by conversion rate, effort as how readily the fixing data is available. A missing GTIN on a bestseller with the barcode sitting in your supplier’s spec sheet is high impact, low effort — do that first, always.
- Assign one owner for feed health, even if it’s a shared role.
- Keep a documented toolchain, from export to supplemental feed to submission.
- Always have a rollback plan for bulk changes, tested before you need it.
- Verify every major rollout against the Diagnostics tab within 72 hours.
Pro Tip: Put the impact‑versus‑effort score in a shared spreadsheet everyone on the team can see. It stops arguments about what to fix next and keeps the backlog honest.
How Moor Marketing applies this playbook
Some marketing agencies run this exact prioritisation sequence with eCommerce clients: a top‑SKU diagnostic first, title batches second, custom label design third, then a staged implementation plan with verification built in. The approach mirrors what’s driven outcomes like very high revenue sales months for some brands, applied here specifically to feed data rather than campaign structure alone.
The starting point for any engagement is always the same diagnostic described above, scoped to the client’s actual catalogue and revenue mix. From there, Moor Marketing’s ecommerce growth framework folds feed work into the broader campaign and conversion strategy rather than treating it as an isolated task, which is where most in‑house efforts fall short.
Why continuous feed discipline compounds growth
The mistake most accounts make is treating feed work as a launch task, something you get right once and move on from. Feeds decay. Categories get retired, competitors update their titles, shoppers change how they search. A feed that was well‑optimised eighteen months ago is quietly losing ground today, and nobody notices until impressions drop and someone asks why.
The businesses that pull ahead treat the feed as owned infrastructure, not agency output. They track their own custom labels, own the diagnostic cadence, and know which SKUs matter without waiting for a monthly report to tell them. Small, weekly fixes beat a single heroic overhaul every time, because Google’s matching systems reward consistency more than bursts.
— Liza
Get a feed audit and 12‑week optimisation plan
You’ve now got the full sequence: diagnose, prioritise by top SKUs, fix titles and attributes, lock identifiers before any AI touches your data, and measure by custom label tier. Running that properly across a real catalogue, on top of everything else running a Shopping account demands, is a genuine time cost most in‑house teams don’t have spare.
This company is the alternative to a generic feed audit tool or a freelance one‑off fix: a senior strategist runs the diagnostic, builds the title and attribute batches, designs your custom label structure, and stays hands‑on through implementation rather than handing you a PDF and disappearing. No outsourced execution, no templated reports. An experienced team that has driven strong revenue outcomes for other eCommerce brands applies that prioritisation logic directly to your feed and your campaigns together, because the two were never separate problems.
If your Shopping performance has plateaued and you suspect the feed is part of it, the 12 Week Double Your Revenue Challenge is built around exactly this kind of structured, staged optimisation. Otherwise, get in touch to scope an audit and find out what’s actually costing you impressions.
Sources
For the technical detail behind this playbook, Google’s own guidance on feed optimisation in the AI‑driven commerce era explains why enriched data widens AI matching surfaces. The FeedGen README documents the open‑source draft‑score‑diff‑approve pattern referenced above. Seer Interactive’s checklist and SKU Analyzer’s prioritisation guide are both worth bookmarking for ongoing reference.
- Feed optimisation in the AI‑driven commerce era — Google Shopping Solutions
- FeedGen README.md — Google Marketing Solutions (GitHub)
- 9 tips to optimise your Google Shopping feed — Seer Interactive
- Product feed optimisation guide — SKU Analyzer
FAQ
What is shopping feed optimisation exactly?
It’s the continuous process of improving product titles, images, attributes and identifiers in your Google Merchant Center feed so Shopping and Performance Max campaigns match, bid and display more effectively.
How often should I optimise my product feed?
Run weekly diagnostics for new issues, monthly rollouts on priority SKUs, and a full quarterly audit covering categorisation, images and custom labels.
What’s the single highest‑impact fix I can make?
Title rewrites on your top‑revenue SKUs typically deliver the fastest, largest lift, because they directly close the gap between how shoppers search and how your titles are structured.
Can I use AI to rewrite my whole feed at once?
Not safely. Use category‑scoped few‑shot prompts, lock GTIN, MPN and brand as immutable, and always generate into a draft feed for review before exporting a supplemental feed.
What causes most Merchant Center disapprovals?
Missing item_group_id values, title or attribute mismatches against your landing pages, and absent required attributes like GTIN cause the majority of disapprovals.
Does Moor Marketing offer feed audits?
Yes, Moor Marketing runs top‑SKU diagnostics, title batch rewrites and custom label design as part of its hands‑on eCommerce growth engagements, including the 12 Week Double Your Revenue Challenge.
