Google has set a hard minimum on product image size, and the deadline is far enough out that most teams will file it and forget it. That is the mistake. From January 31, 2027, product images smaller than 500 by 500 pixels will be disapproved and pulled from Shopping ads and free listings. Warnings started appearing in April of this year, PPC Land reported the confirmed timeline on August 22, and the practical deadline for a large catalogue lands well before the stated one. The Google Merchant Center image requirements are now a visibility deadline, not a feed hygiene task.
What the Google Merchant Center image requirements now say
The specification itself is short. Every product image must be at least 500 pixels on each side, effective January 31, 2027. Google recommends 1500 by 1500 or higher for best performance across listing formats. The upper bounds are unchanged: 64 megapixels and a 16 MB file size ceiling. There is no category carve-out in the announcement, so apparel, hard goods and everything else face the same floor. Google documents the change in its own Merchant Center product data specification update, which is the version worth sending to an engineering team, since third-party summaries have already started disagreeing with each other about the automatic-optimization caveat.
Two details in the announcement deserve more attention than they are getting. The first is that Google says some smaller images will be automatically optimized to meet the requirement, preventing disapproval without action from the merchant. That sounds like relief until you notice the scope is unspecified. Nobody outside Google knows which images qualify, what the upscaling does to visual quality, or what share of a given catalogue it will cover. Planning on it is planning on a number that has not been published.
The second is the framing. Google positions the change as ensuring best performance in all listing formats, and the phrase all listing formats is doing real work. Product imagery flows through the Shopping Graph into Shopping ads, free listings, AI Mode, Gemini experiences and Google Lens simultaneously. One asset, five destinations. Raising the floor on the asset raises it everywhere at once. That is a meaningful change in who should care about this announcement. A specification update that affects only Shopping ads is an ads problem with an ads owner and an ads budget. A specification update that affects the structured product data behind visual and generative search is a company problem, and in most organisations it currently has no owner at all, because the feed sits in one team's tooling and the consequences land in another team's reporting.
| ATTRIBUTE | BEFORE | FROM JAN 31, 2027 | CONSEQUENCE IF MISSED |
|---|---|---|---|
| Minimum image dimensions | No enforced floor, warnings from April 2026 | 500 x 500 pixels | Product disapproved, removed from ads and free listings |
| Recommended dimensions | Larger is better, unquantified | 1500 x 1500 or above | Weaker performance across listing formats |
| Maximum resolution | 64 megapixels | 64 megapixels | Image rejected on upload |
| Maximum file size | 16 MB | 16 MB | Image rejected on upload |
| Automatic optimization | Not applicable | Applies to some smaller images, scope undefined | Cannot be relied on for planning |
Why product imagery stopped being an ads-only concern
For most of the last decade, the Merchant Center feed was owned by whoever ran paid shopping. It was an ads asset, judged by ads metrics, and organic teams paid attention to it roughly never. That division made sense when the feed powered Shopping ads and little else. It stopped making sense when Google connected the Shopping Graph to its generative surfaces.
A product that is disapproved for a small image is not simply missing from a paid placement. It is missing from the structured product data Google uses when someone asks an AI surface to compare options, find something similar to a photo, or recommend a product in a category. The absence is silent. There is no impression loss to explain, because there was never an impression, and no ranking to track, because these surfaces do not report positions. The product just does not come up. This is the failure mode that makes structured data problems so persistent. A ranking drop generates a ticket because somebody watches rankings. A disapproved product generates a warning inside a tool that only the paid team logs into. And an absence from an AI shopping answer generates nothing at all, because no dashboard anywhere is counting the answers you were not part of. The first person to notice is usually a competitor's customer.
That is the same structural failure we described in what happens when agents cannot find your price: the machine reading your catalogue does not degrade gracefully when a field is missing or malformed. It skips you and answers with somebody whose data is complete. Image dimensions have now joined price and availability on the list of fields that can remove you from consideration entirely.
The recrawl problem nobody has priced in
Here is the operational detail that turns a five-month deadline into a shorter one. Google recrawls product image URLs on its own schedule, and for URLs whose content has not changed, that interval has been reported at up to six weeks. Replacing the file behind an existing URL does not guarantee a prompt refresh. You can fix an image in November and still be showing Google the old dimensions in December.
The workaround is not exotic, but it does need to be decided early because it affects how you do the remediation. Publishing corrected images at new URLs, and updating the image_link value in the feed, forces a fetch rather than waiting for one. That is straightforward for a catalogue of a few hundred products and a genuine engineering task for a catalogue of half a million, particularly where images are served from a digital asset manager with its own URL conventions and cache behaviour.
Illustrative remediation timeline for a large catalogue, working back from the January 31, 2027 deadline
The proportions above are our planning estimate rather than measured data, and they are deliberately weighted toward asset creation, because that is where these projects actually stall. Finding the non-compliant products takes an afternoon. Producing several thousand replacement images at adequate quality, for products that may have been photographed years ago by a supplier who no longer exists, is the part that takes a quarter.
Auditing your catalogue against the new floor
Do the audit now, while the deadline is still comfortably distant, because the output of the audit determines whether this is a two-week fix or a two-quarter project. The steps below assume access to the Merchant Center diagnostics and the underlying product feed.
That last point is worth pressing on. Every ecommerce catalogue audit we have run has turned up more than the thing we were looking for, and the secondary findings are usually worth more than the primary one. Image dimensions are unusually easy to detect and unusually easy to fix compared to, say, inconsistent product identifiers across a merged catalogue. Use the deadline as the reason to open the file, then do the rest of the work while it is open. The economics favour it heavily: the fixed cost of a catalogue audit is mostly in access, tooling and getting the right people in a room, and that cost is the same whether you fix one attribute or six. Running the same project again in March to correct product identifiers means paying the setup cost twice for no reason. Our ecommerce engagements have consistently found that feed quality and organic product visibility move together, because they are fed by the same underlying data.
A five-month remediation sequence
Work backwards from January 31 and the sequence is straightforward. Finish the audit and the revenue segmentation this quarter, so you know the size of the asset-creation job before budgeting season closes. Produce and publish replacement images for the top revenue decile first, at new URLs, through October and November. Handle the long tail through December. Leave January entirely as buffer for re-review and exceptions, on the assumption that some share of your corrections will be rejected for something unrelated to size and will need a second pass.
Then check the thing most teams will skip. After the corrected images are approved, verify that the affected products actually appear in AI Mode and Lens results for queries where they should, rather than assuming approval equals presence. Approval is a feed state. Presence in a generative answer is a separate outcome, and the whole reason this deadline matters more than a specification change normally would is that the two are now connected. Our SEO and ecommerce and retail teams treat feed compliance and AI surface presence as one workstream for exactly that reason, and the January date is a good excuse to merge them at your end too.
See where you are cited today
A free snapshot audit of your rankings and AI citations before we ever talk.
Tyler leads work at the intersection of SEO and generative engines at Something Inc., helping B2B brands get ranked and cited across every major AI engine.