What you will find in this article
- How the Asset Report works and where to read the image rating in PMax
- What Low, Good, Best mean: the retrospective statistical logic of the system
- Which factors Google has officially confirmed as determinants of the rating
- Smart Cropping and Image Enhancements: how Google's AI modifies your images
- Technical image requirements for the three formats required in PMax
- Reasonable inferences on the asset selection and distribution mechanism
- Practical patterns for optimising the rating based on real Asset Report data
PMax campaigns, a campaign type that automatically manages all Google channels, distribute your ads across Search, Shopping, Display, YouTube, Discover, Gmail and Maps. At the heart of this system is the asset data: images, headlines, descriptions and videos you upload to the asset group. But how many Performance Max campaigns have you seen where all images appear to carry equal weight? The system does not work that way.
Google assigns each image a performance rating, Low, Good, or Best, visible in the Asset Report. This PMax asset rating determines how frequently the image is selected for each auction. The problem is that most advertisers read it as an aesthetic traffic light, when in reality it is a far more complex statistical signal. Understanding how it works radically changes the way you manage creative assets in PMax campaigns.
The Asset Report: where to read the rating and what it shows
The Asset Report is one of the most actionable management reports in Google Ads. Unlike dedicated asset management software, the platform integrates this data directly in the campaign view. The starting point is knowing where to find it: the image rating in Performance Max campaigns is in the Asset Report, accessible via Campaigns → select the PMax campaign → Assets tab.
The report includes for each image: the format (landscape, square, portrait), the number of impressions generated, and the current rating. An important detail that many overlook: in the rating column you can find four states, not three.
| Status | Meaning | Recommended action |
|---|---|---|
| Pending | Insufficient data to assign a rating. The image was published recently. | Wait 7–14 days. Do not intervene. |
| Low | The asset performs worse than others of the same type in the group. | After 3 weeks with sufficient impressions, replace with a variant. |
| Good | Average performance within the group. Contributes in a balanced way. | Keep. Test variants to try to reach Best. |
| Best | The highest-performing asset in the group for that format type. | Keep. Analyse what sets it apart and replicate it in variants. |
PMax asset rating Low, Good, Best: the underlying logic
The first mistake I see repeated is interpreting the rating Low Good Best Performance Max as an aesthetic judgement that Google passes on images before they go to auction. It does not work that way. The rating is a retrospective statistical evaluation, built on accumulated performance data over time after publication.
The relativity of the rating has an important practical consequence: if you have only one asset per format (a single landscape image, one square, one portrait), that single asset will automatically be Best, because it has no internal competitors to be compared against. This does not mean it is performing well in absolute terms; it simply means it is the only option available to the algorithm.
The second critical aspect is the stabilisation time window. I have observed across multiple accounts that ratings shift significantly in the first 2–3 weeks. An image that starts at Pending status can move to Best after 10 days, or to Low after 3 weeks with more data. Unlike dashboards that update in real time, the PMax asset rating is a lagging indicator, intervening on assets before this window is one of the most frequent and costly mistakes in PMax management.
Factors that influence the rating: what Google has confirmed
Google has not published a complete formula for calculating the rating. What it has officially confirmed concerns the performance signals used and the technical requirements that images must meet to be eligible for distribution.
Confirmed performance signals
Confirmed signals therefore include CTR (click-to-impression ratio), conversions attributed to assets, and engagement signals, a term Google uses without defining precisely, but which in practice manifests as time on page and post-click interactions. All of this falls within the performance data the system accumulates before stabilising the rating.
Technical image requirements: what the system checks automatically
| Format | Aspect ratio | Minimum size | Recommended size | Max file |
|---|---|---|---|---|
| Landscape | 1.91:1 | 600×314 px | 1200×628 px | 5 MB |
| Square | 1:1 | 300×300 px | 1200×1200 px | 5 MB |
| Portrait | 4:5 | 480×600 px | 960×1200 px | 5 MB |
Providing all three formats is not merely an aesthetic recommendation: it is a matter of placement coverage. Images and video assets cover different placements, the portrait format is the one with the fastest growth in 2026 on Discover, Gmail and Maps on mobile, driven partly by social media consumption habits shaping vertical content expectations, and omitting it means automatically excluding yourself from a growing portion of PMax inventory.
❌ Images with excessive overlaid text
Google can disapprove images where text covers more than 20% of the surface. In PMax campaigns text is added via headlines, do not burn it into the image.
❌ Images with added borders or padding
Black, white or coloured borders added to simulate other formats are detected by the system and penalise distribution. Do not artificially extend images.
❌ Collages of multiple images
Multiple subjects side by side in a single image make Smart Cropping difficult and produce visually fragmented ads on small placements.
❌ Logo overlaid as a watermark
Google adds the logo separately as an asset. Having it in the image too creates visual duplication in layouts where both are shown together.
✅ Clear and centred main subject
Facilitates automatic Smart Cropping and ensures the visual message remains readable even in small formats or when the image is cropped.
✅ High resolution with a safety margin
Uploading at the recommended size (not the minimum) ensures that automatic crops maintain visual quality across all placements, including high pixel-density ones.
Smart Cropping and Image Enhancements: the AI that modifies your images
One of the least-discussed aspects of Performance Max campaigns is that the images you upload may not be exactly what the end user sees. Google has introduced two automatic modification functions that operate on uploaded assets.
Smart Cropping
The practical mechanism works like this: every image online in PMax must work across multiple formats and placements. If you upload a 1200×628 landscape image, the algorithm must make it usable on square and portrait placements as well, without you uploading specific variants. Smart Cropping identifies the main subject, using object detection techniques similar to those described in Google's patents on visual ad classification, and crops around it.
The practical consequence for image composition is direct: subjects positioned at the edges of the image are cropped out in the adapted versions. A product placed in the lower left corner of a landscape image can disappear completely in the automatically cropped square version. This is one of the most common compositional errors that silently degrades asset quality.
Image Enhancements
Image Enhancements has included since 2024 the ability to generate alternative backgrounds using generative AI, if the system detects that the original background is generic or not optimal for a specific placement, it can automatically replace it with a contextually more appropriate one. This function was updated in 2025 with the ability to generate images of adults for lifestyle creatives.
Inferences on the asset selection and distribution mechanism
Google has never published the exact formula by which it calculates the asset rating, or with what probability a Best image is selected over a Good one in a specific auction. What follows are reasonable observations based on repeated patterns.
Best assets appear to receive preferential exposure in high-value auctions. Across several accounts I have observed that during hours with higher CPC and more qualified audiences, Best images dominate impressions compared to Good ones. The most reasonable interpretation is that the system maximises conversion rates by favouring historically more effective assets when the stakes are high.
Stylistic diversity appears to influence the rating more than quantity. Unlike financial reports that express absolute figures, the PMax asset rating is always relative to the group. Uploading 15 very similar images, same photographic style, same composition, same colour palette, generally produces a cluster of Good assets with a single Best. Uploading 6 visually diverse images, different photographic styles, different compositions, different usage contexts, appears to produce a more balanced rating distribution and higher total impressions. My interpretation is that the algorithm implicitly penalises visual redundancy, probably because different users respond to different visual stimuli and variety maximises audience coverage.
The rating appears to be influenced by auction context, not just absolute historical performance. I have observed cases in which an image with a Good rating in the general report dominates impressions on a specific placement (e.g. Gmail). This suggests the system evaluates assets also as a function of placement, an image can be Good on average but Best on a specific channel, and the aggregate report does not show this granularity.
Creative fatigue appears to manifest in PMax faster than in traditional Display campaigns. Unlike physical assets that depreciate through use, digital image assets lose effectiveness through overexposure. Best assets that remain unchanged for more than 8–10 weeks tend to lose their rating over the long term. The most likely mechanism is that the same user, repeatedly exposed to the same image, stops engaging , and the falling CTR is reflected in the rating. Periodically refreshing assets is not just an aesthetic best practice, it is a signal the system interprets as the ability to maintain relevance over time.
How to optimise Performance Max campaigns based on image rating
Do not touch anything for the first two weeks. This is the most important and most frequently violated rule. The system needs time to collect sufficient data, intervening during the Pending phase resets the learning process and requires further weeks before ratings stabilise. I have seen campaigns that never found equilibrium precisely because the advertiser replaced assets every week on the basis of insufficient impressions.
Always upload all three formats from the start. Landscape, square and portrait are three different inventories. Missing portrait means forgoing Discover, YouTube Shorts and Gmail mobile, the fastest-growing placements in PMax in 2026. It is not possible to recover these impressions retroactively by adding the format later without impacting the optimisation process.
Analyse the rating by format, not only in aggregate. Use the Asset Report to track assets across placements and channels: an image showing Low in the aggregate report might have Best on a specific placement. Before replacing an asset, filter by placement or channel to understand where the image actually performs and where it does not. Blind replacement based on the aggregate rating often removes assets that have value on specific channels.
Replace Low assets with variants, not completely different images. If an image has a Low rating, the most productive hypothesis is that the style or composition does not work for that audience, not that the subject is wrong. Improving asset performance through progressive variation, different composition, different shooting format, different background, produces more useful information than completely changing the image. Radical replacement disperses the accumulated learning signal.
Disable Image Enhancements for assets that require precise brand control. For e-commerce with product images that must adhere to precise background and composition standards, Image Enhancements can introduce unapproved modifications. Bear in mind that Google activates these modifications precisely because it believes they improve the chances of driving conversions, but creative control takes priority if you are protecting a precise brand identity. Treat periodic asset refresh as preventive maintenance: it is a signal the system interprets as the ability to maintain relevance over time. Disabling is done at individual asset level in the upload interface.
Use Best images as a creative reference for new variants. When an image achieves Best, analyse what sets it apart from the Good and Low images in the same group, composition, subject type, colour palette, presence or absence of people, usage context. Also consider consistency with the destination landing pages: images visually aligned with the arrival page tend to maintain their rating longer because they reduce the post-click abandonment rate. Replicating winning patterns in new variants is more effective than any generic best practice.
Build a solid asset inventory: maintain a minimum of 5–6 images per format in each asset group. With fewer than 3 images per format, the rating is distorted, as described above, the only available image is automatically Best. With 5–6 different images per format the algorithm has sufficient material for a meaningful comparative evaluation and to adapt to different audiences and contexts.
