What you will find in this article
- What Smart Cropping is and where Google officially documents it today (updated URLs)
- How the focal point detection mechanism works, what Google has confirmed and what is inferable
- The formats and placements that trigger automatic cropping in PMax and Display
- How the algorithm behaves on images with edge subjects, collages, and lifestyle images
- Inferences from observed behaviour on real campaigns, clearly flagged as such
- How to optimise image composition to avoid problematic crops
You upload a landscape 1200×628 image to Google Ads and the campaign uses it on square and portrait placements too. You did not upload separate versions for those formats. Who decided how to crop the image? Smart Cropping did, Google Ads’ AI automatic cropping system.
Most advertisers know this feature exists but do not know exactly how it works. This is not a theoretical problem: a subject cropped out by the automatic crop is an ad that loses its visual message on that placement, and no alert signals it in the interface. The problem is silent and systematic.
This article clearly separates what Google has officially confirmed, what emerges from verified documentation, and what I have observed directly while managing Display and Performance Max campaigns.
What Smart Cropping Is in Google Ads: The Official Documentation
Smart Cropping is the function included in Image Enhancements in Google Ads that automatically adapts uploaded images to different placements and formats by identifying the focal point of the image and cropping around it.
Two important points many overlook. First: Smart Cropping is not only for Responsive Display Ads — it is documented for Performance Max, Demand Gen, and Gmail too. Second: the control level differs by campaign type: for PMax it is at campaign level (opt in/out in settings); for RDAs it is per asset via asset enhancements.
How Focal Point Detection Works: Confirmed and Inferable
Google has never published technical documentation of the exact model used for focal point detection. What is confirmed is the output — the system identifies the main subject and crops around it. The mechanism is only documented indirectly, through the composition requirements and recommendations Google provides to advertisers.
What is officially confirmed
What is reasonably inferable
The algorithm likely uses salient region detection techniques. Salient region detection is a family of computer vision algorithms that identifies the regions of an image that most capture human visual attention. Smart Cropping’s focal point detection behaves consistently with these models: it favours subjects with high contrast against the background, centred subjects, and subjects with sharp contours.
Human faces likely receive greater weight in focal point detection. On campaigns with images featuring people, the automatic crop almost always preserves the face, even when this means cutting part of the surrounding context. This is consistent with what is documented about Google Vision AI’s Face Detection API.
The system appears to penalise compositions with a dominant background. Images where the subject occupies less than 30–40% of the frame are cropped in ways that are often hard to predict. This is consistent with Google’s documented requirement that images must have no significant empty space before upload.
Formats and Placements: Where Smart Cropping Activates
Smart Cropping activates every time Google needs to adapt an image to a format different from the one it was uploaded in. The three main aspect ratios used across Google’s placements each serve a distinct set of inventory.
Landscape
1200×628 px recommended
Horizontal banner placements, Gmail, Display header
Square
1200×1200 px recommended
Square placements, YouTube companion, Discovery feed
Portrait
960×1200 px recommended
Discover, YouTube Shorts, Gmail mobile, mobile feed
When you upload only the 1.91:1 landscape image, the system must generate the square version by cropping in a 1:1 ratio, removing roughly 47% of the horizontal content. If the subject is not centred, it is mathematically inevitable that it gets partially cut in the portrait version.
| Campaign type | Required formats | Smart Cropping activates when | Control level |
|---|---|---|---|
| Performance Max | Landscape, Square, Portrait | One of the 3 formats is missing | Campaign level opt-in/out |
| Responsive Display Ads | Landscape, Square | Square format missing or adaptation needed | Per-asset (asset enhancements) |
| Demand Gen / Discovery | Landscape, Square, Portrait | One of the 3 feed formats is missing | Campaign level, enabled by default |
| Gmail | Landscape, Square | Adaptation to Gmail templates | Per-asset |
Inferences from Observed Behaviour: Real Cases
Edge subjects are systematically cropped out in landscape-to-portrait conversions. A product positioned in the lower-left corner of a 1200×628 landscape image is almost always excluded or pushed to the margin in the automatically generated portrait version. The system keeps the central vertical portion and discards the lateral sections.
Images with multiple subjects produce unpredictable crops. On campaigns with group images or product collages, Smart Cropping behaviour is the hardest to predict. In some cases the system selects the largest subject, in others the one with the highest contrast, and in others an intermediate subject that is not clearly the main focus.
Text in the image confuses focal point detection. Images with large text claims show unusual behaviour: in some cases the automatic crop orients towards the text area rather than the visual subject, because the system interprets high spatial frequency areas (sharp letter edges, strong contrast) as saliency signals.
Plain backgrounds improve crop precision. Images with white or very uniform backgrounds produce more predictable crops than images with complex or textured backgrounds. A uniform background makes it easier for the algorithm to isolate the subject through segmentation.
How to Optimise Images for Smart Cropping
Centre the subject within the central 70–80% of the frame, both horizontally and vertically. A centred subject is mathematically safe from any automatic crop — the central area is always preserved regardless of the target format.
✓ Centred subject, clean background
The subject occupies 60–70% of the frame, centred horizontally and vertically. Uniform or very simple background. Predictable crop across all formats.
✓ Single clear main subject
No ambiguity about which is the focal point. The system has only one candidate and preserves it. Works for products, people, and single objects.
✗ Subject positioned at the edge
Classic “rule of thirds” composition, subject to the left or right of the frame. On conversion to portrait or square, the subject may be cropped out.
✗ Collage of multiple subjects
Multiple products or people side by side. The system does not know which to preserve, the crop result is unpredictable and may break the original composition.
⚠ Text integrated in the image
High spatial frequency text can become the perceived focal point, shifting the crop towards the text area instead of the visual subject.
⚠ Complex or busy background
Backgrounds with complex textures or rich patterns can create ambiguity in focal point detection, making the crop less predictable.
Always upload all three formats rather than relying on automatic cropping for the missing ones. Smart Cropping is a safety net, not a creative strategy. The ideal portrait image is composed with the portrait format in mind, not a cropped landscape image.
Use Google Ads Creative Preview before launch to verify automatic crops. The Creative Preview function lets you see how the ad looks across different placements and devices before going live. Checking that the subject is intact across all formats takes 5 minutes and can prevent weeks of campaigns with visually compromised assets.
Disable Smart Cropping only when you have direct control over all formats. Disabling image enhancements for a PMax campaign without uploading format variants means giving up the placements that require those formats.