Google just made AI Overviews generate their own images

Google's AI Overviews can now generate images from a text prompt using the Nano Banana model. If you asked Google to visualise a nautical bedroom yesterday, you got a grid of thumbnails linked to interior design blogs. Today you get a generated bedroom, rendered on the page, with follow-up prompts to refine it.

The Google Images homepage is being redesigned at the same time — from a search box into a browseable, personalised gallery of images from across the web.

Both are rolling out over the coming weeks. Both are being framed as anniversary features for Google Images turning 25. Neither is really about anniversaries.

What Google has quietly done is close another loop where the SERP used to hand traffic to publishers. Image search was one of the last surfaces where a query reliably ended at somebody else's URL. That's what's changing.

The loop that just closed

There's a pattern to how AI features arrive on the SERP. First they answer the question. Then they answer the follow-up. Then they generate the thing the query was looking for in the first place.

Text queries hit that final stage two years ago. "How do I write a polite decline email" doesn't need to send you anywhere — the AI Overview writes the email. "What's a good hook for a LinkedIn post about pricing" doesn't need to send you anywhere either.

Image queries were the holdout. If you searched for "nautical bedroom inspiration," Google could summarise the concept, but the actual thing you wanted — the image — still lived on somebody else's site. Pinterest boards. Interior design blogs. Retailer product pages. Etsy listings. That was Google Images' job: point at where the images actually lived.

That job is now optional. If the AI Overview generates a nautical bedroom on the page, the trip to the interior design blog doesn't happen. If the generated image is close enough and the follow-up prompt refines it further, the trip never happens at all.

That's the loop. And it just closed on images the same way it closed on text.

The zero-click problem gets a visual sibling

Everyone in SEO has spent two years arguing about zero-click search for text queries. What share of AI Overviews reduce clicks. What share increase them. Whether informational content is dead, wounded, or fine. Whether transactional queries are safe. The Bocconi clickstream data I wrote about last week suggests the damage is heavily concentrated in informational queries — which sounds like commercial content is safe until you notice that a lot of top-of-funnel discovery is informational.

The image is the destination now. Not a link to the destination.

Image search sits awkwardly across that split. A search for "modern kitchen ideas" is informational in shape but commercial in outcome — it's how people find the designer, the tiles, the tap, the interior blog whose newsletter converts them later.

Generated images inside AI Overviews break that funnel in a specific way. The generated result satisfies the aesthetic query without ever surfacing the source that would have converted the browser into a lead. There's no attribution, because there's no source — Nano Banana isn't citing the interior designer whose portfolio it trained on. It's producing an image that lives entirely inside Google's surface.

The image is the destination now. Not a link to the destination.

That has consequences most people haven't sat with yet.

Who this hits hardest

Not everyone equally. The publishers and businesses that will feel this fastest are the ones whose commercial model depends on being *discovered visually*:

Interior designers and architects whose portfolio pages have been ranking on Google Images for style queries. Photographers licensing stock. Wedding planners, cake makers, florists, tattoo artists — every creative business where "show me what this could look like" is the top of the funnel. Ecommerce brands where the product image itself is the ad. Recipe sites where the finished-dish photo is what drives the click.

For most of them, Google Images was a discovery channel they didn't have to optimise for — it just worked as a side effect of publishing images that people wanted. That side effect is now being competed with by an infinite generator sitting between the query and the source.

The other group getting hit is the layer of the industry that sells "image SEO" as a discipline. Alt text audits, structured image sitemaps, Pinterest-adjacent tactics. Not that any of that becomes worthless — Google Images the product still exists, and the new gallery homepage arguably surfaces more images to more people — but the assumption that image search is a stable, growing, low-competition discovery channel has just quietly stopped being true.

The new homepage is doing something different

The gallery redesign is the part I find more interesting than the generation feature, because it's a smaller change on the surface with a larger implication underneath.

The old Google Images homepage was a search box. You brought intent, it fetched results. The new one is a feed — personalised, real-time, with saved collections as tabs across the top. Signed-in users see images tailored to their interests. The behaviour it's borrowing from is obvious: Pinterest. Instagram Explore. TikTok's For You page.

That's a shift from *retrieval* to *discovery*. From "I have a query, find me matches" to "here's a stream, browse it." Which changes what winning image search looks like. In retrieval, you win by matching a query. In discovery, you win by being the kind of image that gets served to somebody who's browsing — which means being part of a graph of images the system already associates with the interests of the user.

Being served in that feed is going to look a lot more like being recommended on a social platform than being ranked on a search engine. Signal quality. Engagement. Whether people save you to collections. Whether people who have interests overlapping with yours also engage with you. It's a recommendation problem, not a ranking problem — and most sites still treat their images as ranking artefacts.

What this actually means for the reader

If your business has been quietly relying on Google Images traffic — and a lot of businesses have been, without ever really auditing it — now is the time to look. Open Search Console, filter to Image search, look at the last twelve months. If it's a meaningful share of your organic traffic, you have a channel that's about to be reshaped, and probably compressed.

The strategic response is not to try to out-optimise the AI Overview image generator. You can't. It's a compounding advantage on Google's side. The response is to stop treating image discovery as an SEO problem and start treating it as a brand problem. People search for interior designers by name. They save wedding photographers to Pinterest boards by name. They follow illustrators by name. Named entities survive AI generation because the AI can't generate *you specifically*. It can only generate the aesthetic category you sit inside.

Which brings this back to the argument Tom Critchlow made and I've been making variations of for a while now: the discovery outcomes that matter in AI search are increasingly downstream of brand, not downstream of technical SEO. Image generation inside AI Overviews is another data point on that graph. The businesses that will keep being found are the ones with enough gravitational pull that people search for them by name — and enough recognisable visual identity that no generator can substitute for them.

Everyone else is competing with Nano Banana. That's a competition you can't win.

The measurement gap widens

One last thing worth naming. There is still no reliable way to measure how much of your image search traffic is being replaced by generated images inside AI Overviews. Search Console doesn't split it out. Third-party tools guess. Log files can tell you what didn't happen but not why. The AI visibility dashboards I wrote about a few weeks ago don't cover image surfaces at all — they're built around text citations, and image-generation displacement is invisible to them.

So we're going to spend the next six months in the same fog we were in for text AI Overviews when they launched — knowing the traffic has moved without being able to prove exactly where. This is not a reason to panic. It is a reason to stop making budget decisions from dashboards that were built for a SERP that no longer exists, and start making them from the harder-to-quantify question of whether your brand is one people search for on purpose.

The image side of that question just became urgent.