Why Performance Max Fails: It’s Your Feed, Not Your Bids

Performance Max gets blamed for a lot. Wasted spend. Black-box reporting. Brand traffic cannibalization. But after auditing hundreds of retail accounts, we keep finding the same root cause. And it is almost never a campaign setting.

It’s the product feed. For retail advertisers, Performance Max feed optimization matters more than any bid strategy, audience signal, or asset group tweak. Google’s automation decides when your products show, who sees them, and how much you pay. It makes every one of those decisions based on your feed data.

Feed in weak data, and the smartest bidding algorithm in the world will happily spend your budget on it. This article breaks down why the feed drives retail PMax performance, the five issues we see most often, and how to fix them at scale.

PMax Took Away Your Levers, Except One

Think back to Standard Shopping. You had priority structures, query sculpting with negatives, and bid control down to the product level. Performance Max stripped most of that away. You can’t target keywords. You can’t see full search term data. You can’t set a bid for a single SKU.

So what do you still control? Five things:

• Budget and ROAS targets
• Campaign and asset group structure
• Creative assets (text, images, video)
• Audience signals (suggestions, not targeting)
• The product feed

Of those five, the feed is the highest-leverage lever by far. Why? Because it’s the input to everything else. The feed determines which auctions you can enter, how relevant Google thinks you are, and how your products get grouped and reported.

Here’s the key insight: Smart Bidding optimizes within the opportunities your feed creates. It cannot create new opportunities. If your title doesn’t say “linen,” you will struggle to enter linen-related auctions, no matter how generous your tROAS is.

How Performance Max Actually Uses Your Feed

Your Titles Are Your Keywords Now

In Shopping and PMax, you don’t bid on keywords. Google matches user queries against your product data instead. The title carries the most weight in that matching, followed by description, product type, and category.

That makes the product title the closest thing you have to a keyword list. Google’s own product title specification recommends including brand, product type, and defining attributes like color, size, and material. Most feeds don’t come close.

One more detail that matters: titles can be up to 150 characters, but only the first 70 or so are visible on most placements. Mobile cuts even earlier. Front-load your most important words.

Attributes Decide Which Auctions You Enter

Beyond the title, structured attributes quietly expand or shrink your reach:

• GTIN: connects your product to Google’s product knowledge graph, unlocking more matching signals and competitive benchmarking.
• Color, size, gender, age group: power filtered results and long-tail queries like “women’s beige wool coat size M.”
• Product type and Google product category: help Google classify the item and help you build clean listing group structures.
• Custom labels: don’t affect matching, but they let you segment by margin, season, or stock level, essential for budget control.

Every missing attribute is a set of auctions you silently never enter. The full list of available fields lives in Google’s product data specification.

The 40% Visibility Problem

Here’s a pattern we see in almost every retail account we audit: only around 40% of approved products get meaningful Shopping visibility. The other 60% are technically live but earn almost no impressions. We call them zombie SKUs.

These products aren’t disapproved. Merchant Center shows them as healthy. Yet Google rarely serves them, because their data gives the algorithm nothing to work with: vague titles, missing attributes, thin descriptions.

Why does nobody notice? Two reasons:

1. PMax reporting aggregates everything. Campaign-level metrics look fine because your top 50 products carry the account. The long tail hides underneath.
2. Product-level reports take effort. You have to deliberately pull listing group or product reports to see impression distribution. Most managers never do.

The result: a large chunk of catalog, and the ad spend potential behind it, sits idle. Meanwhile, the team keeps tweaking tROAS and wondering why scaling stalls.

Five Feed Issues That Quietly Kill PMax Performance

1. Vendor-Style Titles

Titles like “BLZ-2291 Navy M” were written for an internal ERP, not for a shopper. They contain no searchable words. This is the single most common and most damaging issue we find.

2. Missing or Generic Product Types

A product_type of “Default” or a one-word category gives Google weak classification signals. It also makes your listing groups a mess. Use a full path: Apparel > Women > Outerwear > Wool Coats.

3. Incomplete Attributes

Color, size, gender, age group, and material are required for apparel in many countries. Beyond compliance, they expand your query coverage. A coat with no color attribute never matches a query that includes one.

4. Identifier Problems

Wrong or missing GTINs, or misuse of the identifier_exists field, lead to limited visibility or outright disapprovals. Audit identifiers first if your catalog includes resold branded goods.

5. Thin or Duplicated Descriptions

Hundreds of stores copy the same manufacturer description word for word. Original, descriptive copy gives Google more matching material and differentiates your listing. It also feeds PMax’s text assets.

How to Audit Your Feed in 30 Minutes

You don’t need a tool to diagnose the problem. Run this quick audit:

1. Pull a product-level impressions report for the last 30 days. Sort ascending. What percentage of products got fewer than 100 impressions? That’s your zombie rate.
2. Open Merchant Center diagnostics. Look past disapprovals and check items flagged as “limited” too.
3. Sample 20 titles from your zero-impression products. Honest question: would a real shopper ever type those words into Google?
4. Compare a top seller with a zombie SKU from the same category. Count the attribute gaps. The difference is usually obvious.
5. Check title truncation. Are your key attributes within the first 70 characters, or buried at the end?

If your zombie rate is above 50%, you have a feed problem, not a bidding problem. Fixing it will outperform any campaign-settings change you could make this quarter.

Fixing Titles at Scale: A Before-and-After Look

For apparel, a reliable title formula looks like this: Brand + Gender + Product Type + Key Attribute + Color + Size. Adapt the order to your vertical: for electronics, lead with brand and model; for home goods, lead with the product type.

Here’s what that change looks like in practice:

The optimized versions contain brand, gender, product type, material, fit, and color — all words real shoppers actually search. Each one creates new auction entries the original title could never match.

Now the hard part: scale. We recently worked on a European fashion retailer’s catalog of more than 11,700 products. At two minutes per title, a manual rewrite would take close to 400 hours. Nobody does that. Which is exactly why most catalogs stay broken.

Manual, Rules-Based, or AI: Choosing Your Approach

There are three realistic ways to fix a feed, depending on catalog size:

• Manual editing works fine under roughly 200 SKUs. Spreadsheet, title formula, a focused afternoon. Done.
• Rules-based feed tools (DataFeedWatch, Channable, and similar platforms) excel at mapping and transforming data: combining fields, fixing categories, applying conditions. Their limit: rules can only rearrange the data you already have. If the source title is “BLZ-2291,” no concatenation rule will turn it into a great title.
• AI-based rewriting (tools like MagicFeedPro) generates search-oriented titles and descriptions from the underlying product data (type, attributes, description fragments) rather than just rearranging fields. This is what makes 11,700 titles a batch job instead of a 400-hour project.

A word of honesty about the AI route: it needs guardrails. Review samples before pushing anything live. Keep your brand naming conventions. Validate output against Google’s product data spec. And roll changes out in batches so you can measure impact cleanly.

Whichever approach you choose, measure it like a campaign change. Watch impressions, CTR, and the share of products receiving traffic over the following two to four weeks. Google’s Performance Max documentation is clear that the system learns from your inputs. Better inputs, better learning.

Conclusion: Treat Your Feed Like a Campaign

Performance Max didn’t remove your control. It moved it. The optimization work that used to live in keywords and bids now lives in your product data. Most advertisers haven’t followed it there, which is precisely why feed optimization is still such an undervalued edge.

Before you touch your tROAS again, do this instead:

1. Pull the product-level report and calculate your zombie rate.
2. Audit 20 titles from your invisible products.
3. Fix titles for your highest-revenue categories first.
4. Measure for four weeks before judging the result.

Your bids are not the problem. Your feed is the part of Performance Max you fully control, start acting like it. Run the 30-minute audit this week, and you’ll likely find more growth sitting in your product data than in any setting inside the Google Ads interface.

Alain Abulafya

Alain Abulafya is a performance marketing consultant and the founder of MagicFeedPro, an AI-powered Google Shopping feed optimization platform. He has managed Google Ads and Meta Ads accounts for ecommerce and lead generation brands for over a decade.