Google AI Overviews stopped linking to your brand photos on July 14, 2026. Now they draw their own.
The announcement came buried in a Google Images anniversary post. Brad Kellett, Senior Engineering Director for Search, wrote that the update "transforms a simple text prompt into a high-quality, custom visual made completely from scratch." No photographer. No publisher credit. No click path back to your website. Google's model renders a brand-new image inside the answer box that sits at the top of search results.
If you're a marketer and you didn't hear about this, that's the point. It shipped on a Monday, framed as a celebration of 25 years of Google Images. The trade press caught it. The brand teams didn't.
What Actually Changed
Here's the distinction that matters. Google's AI Mode (the opt-in conversational tab) has been surfacing images for months. But those images were real photographs owned by publishers and brands. They carried attribution. Clicking them sent traffic to the source.
The July 14 update moved image generation from AI Mode into AI Overviews, the default answer box shown to every searcher who never chose an AI experience. The model doesn't retrieve an image. It manufactures one. Search Engine Roundtable's Barry Schwartz put it plainly: "This will lead to fewer clicks on links and fewer people looking at Google Images from publishers."
That's not a prediction. It's already happening. AI Overviews have driven a 58% decline in click-through rates for the top-ranking result since early 2025. The zero-click rate when AI Overviews appear now sits at 83%. Google was already eating your text traffic. Now it's eating your visual traffic too.
The Two-Tier Control Gap
Here's where it gets uncomfortable for anyone who doesn't buy Google ads.
Google's Nano Banana Pro model gives advertisers detailed creative controls inside Google Ads Asset Studio. You can adjust camera angle, lighting, depth of field, bokeh, focus, and color grading. You can feed reference images to maintain brand consistency. Paying advertisers get the knobs.
Organic AI Overviews image generation ships with zero stated brand controls. Google's announcement didn't mention sourcing, attribution, or how a brand might influence what gets drawn. If you don't pay, Google's model generates whatever it thinks your product looks like based on whatever it scraped from the web.
This is a two-tier internet for visual brand identity. Advertisers shape their synthetic images. Everyone else gets whatever the model decides. We've already covered how AI Overviews are reshaping citation authority, but this goes further. It's not just about what Google says about your brand. It's about what Google shows.
Why This Is Different From Text
You might be thinking: brands already lost control of text in AI Overviews. Google summarizes your product description however it wants. Why are images worse?
Because images carry trust differently. A text summary that's slightly off can be corrected with better content and structured data. A synthetic image that misrepresents your product is a completely different problem. You can't SEO your way out of a wrong picture.
NewsGuard already ran a red-teaming audit on Nano Banana Pro and found it generated images advancing all 30 false claims they tested. All 30. The model that Google is now using to draw your brand in search results has a demonstrated inability to distinguish fact from fabrication.
If someone searches "show me what a [your brand] product looks like" and Google draws a synthetic image that's wrong, misleading, or visually off-brand, you have no tool to fix it. There's no edit button. There's no claim process. There's no brand verification portal. The image exists for the duration of the search session and disappears, but the impression stays.
This matters in regulated industries especially. We've written about how AI detection evasion affects regulated markets and the liability traps in AI Overviews. A cannabis dispensary that gets a synthetic image with the wrong product labeling, wrong health claims, or wrong compliance messaging embedded in it has a problem that goes beyond marketing. It's a regulatory exposure created by a platform the brand cannot control.
The Visual GEO Problem
GEO (Generative Engine Optimization) has been focused on text. Get your content cited, your data structured, your entity recognized. That playbook still matters. We've covered the GEO visibility traps brands are already falling into.
But visual GEO is a different animal. When Google draws your brand, it pulls from whatever visual data it has. If your product images across the web are inconsistent, low quality, or outdated, that's what the model trains on. If your competitors have better, more consistent visual footprints, the model may draw something closer to their version of the category than yours.
The third-party analysis from Digital Applied argues brands can influence generated visuals through structured data, consistent visual identity, and authoritative terminology. That's plausible. It's also unconfirmed by Google. Nobody outside Google knows what signals the Nano Banana model weights when it draws a brand image inside AI Overviews. Treat the visual GEO playbook as informed speculation, not a proven system.
What Brands Should Actually Do
Three things, in order of urgency.
First, audit your visual footprint across the web. Every product image, every press photo, every social media visual. If the images Google can find are inconsistent, the model will draw inconsistent results. Standardize your visual identity before the rollout reaches full coverage. This means consistent backgrounds, color profiles, product angles, and metadata across every image you publish.
Second, use the advertiser's shortcut if you have the budget. Google Ads Asset Studio gives you Nano Banana Pro controls. Even if you're not running large campaigns, generating brand-consistent images through the ads platform may train the model on what your brand actually looks like. It's a paid hack for an organic problem, but it's the only confirmed mechanism right now.
Third, accept that visual GEO is the next frontier. The brands that win will be the ones whose visual identity is so consistent and so widely distributed that the model can't get it wrong. That means investing in image quality, structured data for visual assets, and distribution across platforms Google indexes.
The Honest Shrug
Here's what nobody can tell you yet: whether this matters at scale. The feature is rolling out "over the coming weeks" in English only, limited to regions that already support image creation in AI Mode. It may take months to reach full coverage. It may never reach every query type.
But the direction is clear. Google went from linking to your images, to summarizing your text, to drawing its own pictures of your brand. Each step removed one more reason for a user to click through to your site. The visual layer was the last place where a brand's authentic presence still lived in search results. That layer is now synthetic.
You can't opt out. You can't request removal. You can only make sure that when Google's model decides what your brand looks like, it has enough good material to get it right.
The brands that treat this as a visual data problem, not a design problem, will be the ones that survive the transition. The ones that don't will wake up one morning to find Google drawing a version of their brand they've never seen before, shown to every searcher, with no way to correct it.