Visual findability: Google Images is being rebuilt

Google Images is becoming a gallery you browse

Google is rebuilding Images into something you browse instead of search. On 14 July 2026 the company announced a new starting point for Google Images: a gallery with images from across the web, updated continuously and tuned to the interests of the user. Google itself calls it "a dynamic, immersive gallery of images from across the web". Images you save end up in collections, and those collections appear as tabs above the gallery.

That looks like a cosmetic change, but it moves where the search begins. A results page starts with your search term; a gallery starts with what the system thinks the user finds interesting. You do not win there with the right words in the right place. You win because someone stops at your image while scrolling past dozens of others.

For a webshop there is an uncomfortable conclusion in that. Your photography becomes your visual findability. The image does the first work, and your product title, alt text and H1 only come after.

Before you act on any of this, be clear about how small it still is today. The gallery is rolling out on desktop in the United States, in English. In the Netherlands nothing about your traffic changes this month. That is not a reason to click away. It is the reason to read now, because this is one of the few times you see a shift coming before it shows up in your numbers.

What Google does say, and what it leaves open

Google announced two things at once, and the second is at least as big as the first. Next to the gallery, image generation is coming straight into AI Overviews, the summaries above the search results. Google describes that feature as something that turns a simple text prompt into a high-quality, custom image, built entirely from scratch, with their Nano Banana model. According to Google the rollout starts in the coming weeks, in English, for every region where image generation in AI Mode already works.

Then the gap in the announcement. Google says nothing about how images for that gallery are selected or ranked. Not a word about signals, not a word about sources, not a word about what pushes an image up. Google also says nothing about labelling images the system made itself. Neither point is in the announcement at all.

That matters more than it sounds, for two reasons. The first is practical: anyone who offers you a ready-made optimisation for this gallery today is working from assumptions. There is no published rule to act on. The second is a matter of principle: if a user cannot tell whether an image was generated or photographed, the burden of proof shifts to the brand. You then have to make clear yourself that what you show really looks like that.

Do not wait until that right-hand column is filled in. Get your imagery in order so it holds up in both columns: recognisable, real, and complete enough to understand without a caption.

Visual findability starts with recognition

Searching by reading and searching by browsing ask different things from you. When reading, someone scans a page of text and decides on words whether to click through. When browsing, someone decides on the image, in a fraction of a second, without the context of your page around it. Your photo is then not in your own webshop, but among images from ten competitors selling the same product.

The image that wins there is the one that makes clear straight away what it is and whose it is. Those are two different things, and the second one is almost always skipped.

That bottom row is the real danger. Being shown without being recognised costs you nothing you can see. There is no drop in your numbers, because you were never in them. All you notice is that a category slowly moves towards brands whose photos look alike and therefore stick.

Recognisability is not a matter of sticking your logo in the corner. It sits in the things that come back without anyone naming them. A background that does not jump around per product, light that always falls from the same angle, a camera position you hold per product type. At Castagnola that was exactly the work: taking a brand that did something different on every channel and pulling it back to one visual line across packaging, listing and webshop. Put five of those photos side by side without a logo and you see within a second that they belong together. That is what a gallery asks of you.

At most webshops the problem sits in the consistency, not in the photography. The photos are fine on their own, but together they tell no story. How you build that line and why it pays off is in what professional product photos really earn you.

If AI can make an image itself, why would anyone still want to see your photo?

This is the question to ask yourself, even if the answer does not suit you. Image generation in AI Overviews makes it urgent. If someone asks for a wooden desk lamp with a brass arm, and the system draws that lamp on the spot, what is your real photo still worth?

The honest answer: less and less for inspiration, almost everything for the purchase.

A generated image does not show your product. It shows a plausible version of the kind of product someone asked for. That is a fundamental difference, even when the result looks better than your photo. The colour does not match what is in the box, the proportions are an average of a thousand other lamps, and the finish is invented. For mood and direction that works fine. For someone about to pay, it is worthless, because they are not buying a mood but an object that sits on their desk tomorrow.

As soon as it costs money, the question shifts from "what does this look like" to "will I get this". Only your real image can answer that second question.

That is also exactly where it goes wrong for brands that have their own catalogue generated. The moment your customer opens the package, they compare it with the photo their choice was based on. If that does not match, you get a return and a review that follows you around for months. An image that is too flattering sells once and costs you from then on.

We are not against generated imagery. We use it ourselves for concepts, for variants and to test a direction before a camera comes in. The line sits at the question of who is making a promise. Anything that makes a promise about what the customer gets should be a real shot of the real product. Where exactly that line falls per situation, we worked out in CGI vs. AI product visualisation.

There is an opportunity in this too. If the web fills up with images nobody made, provably real imagery gets scarcer and therefore more valuable. A photo where you see fabric, where light behaves the way light behaves, where a product has small imperfections, will stand out sooner than the perfect render next to it.

What the American numbers say about traffic from AI

There is no data on the new gallery yet, because it is not running. What does exist is measurement data on how traffic from AI behaves once it arrives. Adobe analysed more than a trillion visits to American retail sites and saw that traffic from AI sources grew 138% in May 2026 against a year earlier. Since October 2024, when Adobe started measuring this, that traffic has grown by 1,324%.

Important: this is about American retail sites. These are not Dutch figures and you cannot lay them one for one over your own webshop. They say something about direction, not about your expected revenue.

What makes that direction interesting is the quality of the traffic. Visitors who arrive through an AI source convert 54% better than visitors from other channels. They stay 53% longer on the site and view 23% more pages. That makes sense: the groundwork is done, so whoever clicks through comes to check whether what they were promised holds up.

The same analysis measured, per product category, how much of the content on those sites is machine-readable. There is a pattern in that which has everything to do with imagery.

Look at the bottom. Furniture and home comes out at 47%, while cosmetics sits at 63%. Adobe gives no explanation for that difference, so what follows is our reading and not a finding from the research. What stands out is that the categories at the bottom are exactly the categories where the story sits in the image. You only understand a sofa once you see it standing in a room. If that reading is right, the text falls shortest for precisely the products that depend most on imagery.

You do not close that gap with better photos alone. You close it by recording in text, with every image, what is in it. What the wider consequences of that are for your traffic is in agentic search and your webshop traffic.

Does a Dutch webshop have to act on this now?

Yes, but without haste and without a big project. Once more, so you do not have to find it out later yourself. The gallery arrives first on desktop in the United States, in English. For other markets Google has named no date. Your Dutch traffic does not change this month because of it. Anyone who tells you to intervene now or lose visibility is selling urgency that is not there.

The reason to start anyway is a different one. Image work has a long lead time. Reshooting a range on one visual line takes weeks to months, depending on how many products you have and whether you work with a product photography setup or with lifestyle photography. Once this shift reaches the Netherlands, you cannot put a consistent visual language in place in two weeks. Whoever starts now is ready before it counts.

There is one more reason, and it stands apart from Google. Everything you improve about your imagery now already pays off today in channels that are simply running. Your bol listing, your Amazon listing, your own webshop, your ads, your socials. A recognisable visual line raises click-through and conversion there, whatever Google does with its gallery. At Screenmate we built a brand around a product that sat on the market without a name. The gain was that every image on every channel showed the same brand.

So do not see it as preparation for an American feature. See it as overdue maintenance that happens to secure your position in a new channel as well. That is a considerably better argument for your budget than an announcement out of Mountain View.

What you can do with your own product photos this week

No shoot needed to start. The first four steps cost you an afternoon and show exactly where you stand.

1. Look up your own product in Google Images. Use the search terms a customer would use, not your brand name. Look at what shows up from you, and above all: does it stand out among the rest or does it disappear into the grid?
2. Put your five most important hero images side by side, without a logo. Do you recognise them as one brand? If not, that is your first project, and it is not about quality but about consistency.
3. Look at your hero image at thumbnail size. Scale it back to thumbnail size on your phone. Is the product still recognisable? Many photos that look beautiful large turn into a blur when small.
4. Pick one visual constant and write it down. Background, direction of light, camera angle, shadow: take one or two and write down what the rule is. Every new photo follows that rule. That is the start of a visual language, and it costs you nothing.
5. Make sure every product has at least one situational image. Product in use, in the right setting, with something beside it that makes the size clear. In a gallery that image does the work your packshot on white cannot do.
6. Write down what is in every image. A descriptive file name, descriptive alt text, and the key features as text on the page as well. A system that cannot read your image cannot recommend it either.
7. Take away your weakest image. Not replace, take away. One bad image among good ones lowers the average of what people see of your brand.

Anyone who works through these seven points has not learned a new specialism. They have only done what should always have happened, except now before it costs them anything. The technical side of findability we handle through our SEO approach, and how your content relates to AI search engines is in GEO for webshops.

What we do differently now that images are read by machines too

At Oase Creative we have been building brands for e-commerce for six years, and the photo work itself has not changed. The order in which we make decisions has.

We now start with the question of what job each image has, before we build a setup. Recognition, understanding or trust. An image that tries to do all three at once does none of them well. That split decides how many images a product needs, and which image goes first.

We record the image rules in the brandbook, next to the colours and the typography. Background value, lighting setup, camera angle per product type, how shadow falls. A brand only becomes recognisable once someone else can follow those rules too. With a range that grows over years, that is the difference between a visual line and a pile of separate photos.

And we write down what is in every image, as part of the delivery. That is the point where photography and findability meet: a machine does not see mood, it reads what you recorded about it.

As soon as there is a purchase promise in an image, we never replace a real product photo with a generated one. Your customer notices the moment the package opens.

Curious how your imagery does once it is shown away from your webshop? Put your hero images next to those of your three biggest competitors and see whether someone who does not know your brand can point out which ones are yours. If they cannot, it is not your camera. It is your brand. Book a call if you want us to look at it with you.

The rebuild of Google Images is not news for the Netherlands yet. It is an announcement whose consequences land here later than the headline suggests. But the direction is clear enough to act on now: the more imagery that gets made without you, the more value there is in imagery that is provably yours.