What you'll find in this article
- How Google analyzes images in Responsive Display Ads, what is officially documented
- The 20% text rule: why it exists and how it penalizes creatives
- The white background paradox: why Display and Shopping have opposite rules
- What Ad Strength really measures and how to use it as a diagnostic tool
- What can be reasonably inferred about creative ranking in the auction
- Practical patterns I've tested directly that change performance
Google Display Ads use computer vision to analyze every image you upload, but few advertisers know exactly how this evaluation works. Most follow generic rules, "use high-quality images", "be relevant", without understanding the underlying mechanism that determines whether a creative is served frequently or relegated to less visible placements.
Google has never published complete technical documentation on how it evaluates display ad images. But it has published best practices, policies, and constraints that, read carefully, reveal a lot about how the system works. This article separates what is documented from what is inferable, and adds observations from direct practice.
Google Display Ads: two formats with different evaluation logic
Before going into detail it's essential to understand that there are two fundamentally different types of Google Display Ad campaigns, and each has completely separate image evaluation logic.
Responsive Display Ads (RDA) are the default format since 2021. You provide separate assets, images, headlines, descriptions, logos, and Google assembles them dynamically into different combinations for each placement. In this format, Google Ads computer vision analyzes each image individually using artificial intelligence models to determine how compatible it is with different contexts, sizes, and text combinations.
Uploaded image ads are static or animated GIF Google Ads banners that you upload already assembled. Here you have full control over the layout, but you lose access to many placements, in 2026 about 72% of Display inventory is served via RDA, not via static banners. The trade-off between creative control and coverage is real and applies to all Google Ads campaigns using this format.
Confirmed rules: what the system checks
In the official best practice guide for RDAs, Google has documented a series of criteria that images must meet. These are not suggestions, they are constraints that the system actively verifies and that influence the distribution and performance of assets.
The 20% text rule
The logic is simple: in RDAs, Google adds your headlines and descriptions on top of the image as separate elements. If the image already has a lot of embedded text, the final result of the assembled ad is visually cluttered, text layered on text. The system anticipates this by penalizing images with too much copy burned in.
This is one of the most frequent mistakes I've seen in inherited campaigns. The designer creates a creative with the claim written large in the image, "Discover our exclusive offer", because it worked in static banners. In RDAs that same image gets penalized, and the advertiser doesn't understand why that "nice" image receives no impressions.
The 80% empty space rule
Very minimalist images with a small element at the center on an otherwise empty field are penalized. The system interprets excessive empty space as a signal of low visual quality, the image is neither informative nor engaging.
What is explicitly prohibited
❌ Logo overlaid on the image
Google automatically adds the logo to the assembled ad. If you also include it in the image, you create visual duplication in some layouts. Logos naturally integrated into the photograph are accepted.
❌ Fake buttons
Buttons with text like "Play", "Download", "Close", "Click here" burned into the image violate Google Ads policy. They promise features that don't exist and generate disapprovals.
❌ Collages of multiple images
Multiple products or subjects side by side in a single image. Google explicitly recommends single images: Smart Cropping doesn't work well on collages and the visual message is fragmented when images are resized for small placements.
❌ White or digitally composited backgrounds
Google specifies to avoid products on white or digital composite backgrounds. The image blends with the background of the page where the ad appears, making it invisible. (See dedicated section below.)
✅ Text naturally integrated in the photo
Text that is part of the photograph, a street sign, a book cover, writing on a product. It is not overlaid text, it is part of the image. Allowed.
✅ Logo integrated in the photograph
If the logo appears naturally in the photographed scene, on a product, a building, a fabric, it is accepted. The prohibition applies to logos added in post-production as overlays.
The white background paradox: Display vs Shopping
If you manage both Google Shopping campaigns and Display campaigns, you face an apparent contradiction:
In Google Shopping, a white or neutral background is recommended, it ensures the product is the dominant subject, adapts to carousel templates, and meets the visual expectations of someone actively searching for a product.
In Google Display, Google explicitly specifies to avoid white backgrounds or digitally composited backgrounds for Responsive Display Ad images.
The reason is contextual and once understood becomes obvious. The Google Display Network shows ads while users browse editorial content, articles, blogs, news sites. Most of these sites have white or very light backgrounds. A product image on a white background in that context literally blends with the page background, the product seems to float in the void, with no visible borders separating it from the surrounding environment. The ad loses visibility.
Shopping, on the other hand, appears in a structured context, the Google search page with its product carousel, where all listings share the same neutral background. There, visual consistency is an advantage, not a disadvantage.
The practical pattern that emerges: never use the same product image in Shopping feeds and Display RDA campaigns. You need creatives specifically designed for each context.
Ad Strength: what it really measures and how to use it
Ad Strength is one of the most visible signals in the Google Ads interface for RDAs, and also one of the most misunderstood. Most advertisers treat it as a goal, "I need to reach Excellent", without understanding exactly what it measures.
Ad Strength does not tell you whether your images are beautiful, relevant, or effective. It tells you how many options you are giving the algorithm to work with. A Poor Ad Strength with three outstanding images is a signal that you are limiting the algorithm, not that the images are bad.
Conversely: an Excellent Ad Strength with fifteen mediocre images does not guarantee performance. Google itself has stated this explicitly, quality matters more than quantity. Five excellent, varied images beat fifteen similar and mediocre images, even if the latter have a higher Ad Strength.
How to use it in practice: Ad Strength is a structural diagnostic tool, not a creative quality diagnostic. If it's Poor or Average, add more asset variety. If it's already Good or Excellent and performance is low, the problem is image quality, not quantity.
What can reasonably be inferred about creative ranking
The algorithm appears to favor images with a clear, centered main subject. This is consistent with the documented workings of Smart Cropping, which uses object detection techniques to identify the focal point of an image, and with the policy against collages. An image with an ambiguous or multiple subject is harder to crop, resize, and adapt to different placements without losing the visual message.
Contextual relevance appears to influence distribution. Google matches creatives to Display Network sites in part based on the visual content of the ads. An image that clearly shows what a company sells is better matched to relevant sites compared to a generic or purely emotional image.
Images with faces tend to generate higher CTR. This is a pattern documented in visual marketing literature and consistent with advertising data. The RDA optimization system learns from this signal and tends to distribute images with people more than those with products alone, especially for awareness campaigns.
Creative fatigue manifests in Display too, probably faster than in Search. Observing campaigns with static assets over long periods, performance tends to degrade over time, CTR drops even with the same audience and bids. Regular creative refresh, every 6–8 weeks, is an empirically solid pattern even if the exact mechanism is not documented.
How to optimize images for Google Display campaigns: from practice
Remove all text from images destined for RDAs. No claim, no CTA, no price in the image. All copy goes in the headlines and descriptions, which is exactly where Google expects it. I've seen campaigns where removing text from images and moving it to headlines increased impressions on the same assets by 40–60% within two weeks.
Create images specifically for Display, don't recycle Shopping images. The workflow that works: photograph your products both on a white background (for Shopping) and in real-life context (for Display). Same photography sessions, two different sets of shots.
Center the subject in 70–80% of the frame. Both for Smart Cropping and for readability on small placements. Centered versions systematically receive more impressions on small placements (300×250, 160×600) which are still the most common in the GDN.
Provide all three formats: landscape, square, and portrait. In 2026 the portrait format is the most neglected and the one with the fastest growing inventory (Discover, Shorts, Gmail mobile). Not having it means progressively missing placements without anyone noticing it in reporting.
Use images with people for awareness campaigns, product images for retargeting campaigns. For awareness you want to create emotional connection, people do this better than an isolated product. For retargeting the user already knows the brand and wants to see the specific product they visited.
Refresh creatives every 6–8 weeks. You don't have to replace everything, replacing the 2–3 lowest-performing assets with new variations is enough to keep performance stable.