Words by
Jemma
Most ecommerce brands running Performance Max campaigns have the budget right. They have the feed right. What they get wrong is the creative, and Google's AI makes them pay for it in wasted spend.
PMax is a fully automated campaign type. You hand over assets, Google decides where to run them, in what format, to whom, and at what bid. The only real lever you still control is the creative quality you feed it. Get that wrong and the machine optimizes toward the floor.
The Problem With How Brands Feed PMax
I see the same mistake repeatedly. Brands upload three or four product images, write two headlines, and call the asset group complete. Google marks it as poor strength, the campaign launches anyway, and the brand blames PMax for bad results.
The asset strength rating is not cosmetic. It directly influences how aggressively Google's AI will bid your campaign into competitive placements. Poor strength means the system doubts your assets can convert, so it bids lower. You end up in cheaper, lower-intent placements and wonder why ROAS is soft.
The root issue is that most brands treat PMax as a campaign to set up once. Good performance requires treating it as a creative testing system you actively feed with new assets every few weeks.
What Google's AI Actually Does With Your Assets
PMax runs across Search, Shopping, Display, YouTube, Gmail, Discover, and Maps. Each placement has different aspect ratios, video requirements, and copy lengths. Your single square product photo cannot serve all of those well. Google's AI assembles combinations on the fly from whatever you provide.
If you only upload square images, Display placements that need landscape images auto-crop your asset or fall back to your product feed images. If you skip video assets, Google auto-generates one from your statics. I have seen these auto-generated videos and they are not good. They look like a slideshow from 2012.
The system needs at least 3-5 images in each major ratio (landscape 1.91:1, square 1:1, portrait 4:5), multiple headlines of varying lengths, and at minimum one video you actually produced. The more good material you give it, the more placements it can compete for with real creative.
The Creative Volume Brands Are Getting Wrong
A standard product shoot gives you 20 to 30 final images. After selecting the best angles and white-background shots for your Shopify store, you might have four or five that feel right for ads. That is not enough variety for PMax to learn.
Google recommends uploading up to 20 images per asset group. That recommendation exists because the algorithm needs variety to find which combinations perform against different audiences. If you give it five photos, it runs those five photos in rotation and calls it done. The machine learns from limited data.
The math problem is straightforward. To run 20 images across multiple aspect ratios, you need around 60 images total assuming three crop variants each. A traditional shoot does not produce that. The cost would not make sense anyway.
This is the specific gap AI creative production fills. Not replacing a hero campaign shoot. Filling the volume requirement that performance campaigns need but studios cannot economically supply.
Image and Video Specs That Actually Matter
For images, PMax requires a minimum of 314 x 314 pixels for square assets and 600 x 314 for landscape. In practice, upload at 1200 x 628 for landscape and 1200 x 1200 for square, minimum. The main subject should sit within the center 80% of the frame, because Google auto-crops at different ratios and anything near the edges gets clipped.
For video, the minimum length Google accepts is 10 seconds. But anything under 30 seconds struggles to run on YouTube pre-roll where most of the video inventory sits. Upload 16:9, 9:16, and 1:1 versions if you can. That unlocks YouTube, Shorts, and Display inventory simultaneously.
The first five seconds of any video are the only seconds that matter for PMax. Bumper ads run at six seconds. Skippable pre-roll gets skipped at second five. Lead with the product, not a brand animation. Show the thing you are selling before the viewer checks out.
How AI-Generated Creative Changes the Math
The bottleneck for PMax creative is not skill. It is production capacity. A brand manager knows they need 60 images across three ratios, lifestyle context, white background, and close-up variants. Knowing and being able to produce are different problems.
AI creative tools now let you generate lifestyle images, background variations, and crop-ready versions of a product shot in hours rather than days. A shoot that would cost four thousand dollars and take a week of coordination can become a same-day workflow.
The quality question comes up every time. My answer: for PMax Display and Discovery placements, AI lifestyle images convert at the same rate as studio-shot alternatives when the product itself is accurate. The format rewards context and relevance, not production budget.
Where AI creative pays back most clearly is in seasonal rotations. Refreshing PMax assets for summer, back to school, and holiday used to mean three additional shoots per year. Now it means briefing a creative system and uploading new batches. The refresh cycle shrinks from months to days, which means your campaigns stay current instead of running January creative in October.
Building a PMax Creative System That Scales
The brands I see performing well on PMax treat creative uploads as a recurring operation, not a launch task. They have a cadence: new assets go in every three to four weeks, poor performers get removed, and at least one new angle or visual style gets tested each cycle.
Asset group performance reporting in Google Ads shows you which combinations drive the most conversions. Use that data to brief your next creative batch. If lifestyle images outperform white background, make more lifestyle. If one headline consistently appears in top combos, write variants around that angle.
PMax works best when you give the algorithm something worth optimizing. A sparse asset group gets sparse results. The brands that complain about PMax efficiency are almost always the ones who set it and forgot it.
If you want to see what a proper AI creative pipeline looks like for PMax, krev.ai is built specifically for this. Generate product images in every format, at the volume Google actually needs, without the production cost that makes most brands underinvest in the first place.
