AI Overviews Now Generates Images: Brand Visual Stakes

Google AI Overviews can now generate images directly inside the answer box. Announced July 14, 2026 as part of Google Images’ 25th anniversary, the feature turns a user’s text prompt into a brand-new AI visual rendered by Google’s own model — not a link to an existing photograph. For the first time, the default search surface doesn’t just summarize the web’s words. It draws its own pictures.

The stakes for brands are structural, not cosmetic. When AI Overviews took over the top of the results page, brands lost control of the click — a shift we’ve tracked in detail in our zero-click search data update. This launch extends that loss of control to the visual layer: when Google’s model draws “your” product or category, there is no photographer, no publisher, no attribution, and no click path back to the real thing.

This guide covers exactly what Google announced, the Nano Banana model family behind it, why generated images are categorically different from the visual results AI Mode already showed, where brand control disappears across Google’s four image surfaces, and what a practical brand-visual playbook looks like while the rollout is still in its opening weeks.

The announcement came in Google’s official “Google Images: 25 years of visual search innovation” post, published July 14, 2026 and authored by Brad Kellett, Senior Engineering Director for Search. Search Engine Roundtable broke it down the next day. Two things ship together as anniversary features: AI image generation inside AI Overviews, and a redesigned, browseable Google Images homepage.

The image-generation piece is the consequential one. Based on the user’s text prompt, AI Overviews will render a new visual inside the answer itself — Google says the update “transforms a simple text prompt into a high-quality, custom visual made completely from scratch, seamlessly bridging the gap between imagination and reality.” The rollout is staged: it begins “over the coming weeks,” English only, limited to regions that already support image creation in AI Mode. No firm date was committed for either feature.

The companion redesign gives Google Images “a brand new browseable home,” in Google’s words — a dynamic, personalized gallery of images from across the web, with a Collections feature for saving tabs. It rolls out desktop-first, U.S. only, in English, and the personalized feed requires a signed-in Google Account. Google frames both launches with a potted history of visual search — the 2001 Google Images launch (famously prompted by the search demand text results couldn’t satisfy), 2011’s Search by Image, 2018’s Lens in Search, through 2024’s Circle to Search and 2025’s Lens plus AI Mode integration.

Google’s copy attributes the feature to its “latest Nano Banana model” — and stops there. No specific version was named in the announcement, so it would be a mistake to assert which model ID is actually rendering these images. What we can state is the family’s lineage. Nano Banana Pro launched November 20, 2025, built on Gemini 3 Pro and marketed on giving professionals control over camera angle, scene lighting, depth of field, focus, and color grading. Nano Banana Pro (gemini-3-pro-image) and Nano Banana 2 (gemini-3.1-flash-image) both reached general availability on May 28, 2026, and Nano Banana Lite (gemini-3.1-flash-lite-image) followed on June 30, 2026.

Image generation in search is also not brand new — it has been a feature in AI Mode for some time, as Search Engine Roundtable’s Barry Schwartz notes. What changed on July 14 is the surface. AI Mode is the opt-in conversational tab; AI Overviews is the default answer box shown to ordinary searchers who never chose an AI experience. Moving generation from the former to the latter is the difference between a feature for enthusiasts and a change to what search is. That same surface has been absorbing more of the results page all year — see our analysis of AI Mode’s Gemini 3 upgrade and its rankings impact for how the two surfaces relate.

The structural distinction matters more than the feature itself. Search Engine Journal’s reporting draws the line precisely: Google’s prior “visual results” in AI Mode linked to external source images — publisher-owned photographs that generated an outbound click and carried attribution. The new feature has the model create a wholly new image, with no source attribution and no outbound link at all. One is retrieval. The other is manufacture.

The traffic implication follows directly, and the trade press didn’t soften it. Queries with a visual-first intent — the “show me ideas for,” “help me visualize” class — can now be satisfied entirely inside the answer box, without a click to a brand site, a stock library, or a publisher’s gallery. That compounds the zero-click dynamics AI Overviews already created for text, which our AI Overviews optimization guide treats as the baseline condition of 2026 SEO.

Here is the connection the launch coverage hasn’t made. When Google first brought visual results into AI Mode, a Google spokesperson told Search Engine Journal that Google’s systems “did not explicitly distinguish real photos from AI-generated images” on that surface. Read that admission against the new feature: a surface that already struggled to separate real from synthetic imagery now manufactures synthetic imagery itself, inside the most authoritative pixels on the results page.

Play the scenario forward. A user asks Google to show them a brand’s new product — or ideas “like” a brand’s signature design. If the model hasn’t seen the real thing, or blends it with a competitor’s aesthetic, AI Overviews can render a plausible-but-wrong visual under Google’s own authority. The brand has no recourse in that moment: no publisher to contact, no image credit to dispute, no click path on which to correct the record. The user, meanwhile, has little reason to doubt what the answer box just showed them. That is a reputational exposure that didn’t exist when every image in search traced back to someone accountable for it.

The gap is sharpest when you compare Google’s two generative surfaces against each other. In Google Ads Asset Studio, Nano Banana Pro launched with professional control knobs — camera angle, scene lighting, depth of field, focus, color grading — so advertisers can direct exactly how a generated visual represents them. Organic AI Overviews image generation ships with none of that. No brand controls were announced, no opt-out, no preview. The entity being depicted has strictly less influence over the picture than a media buyer has over an ad unit.

No coverage we reviewed puts Google’s image surfaces side by side, so we built the comparison ourselves. The matrix below maps each of Google’s four visual surfaces to where its images come from, how much control a brand has, and whether the web economy — clicks and attribution — still functions there. Read left to right, it shows exactly where in Google’s stack a brand loses the ability to control how its products are depicted.

The pattern is stark once tabulated. Moving down the matrix, brand control degrades from full to indirect to none — and then snaps back to fine-grained control on exactly one surface: the one you pay for. The organic generative surface is the only cell in Google’s image stack where the depicted entity has no documented influence, no click, and no attribution simultaneously. That’s the cell every brand’s search visibility is now drifting toward.

The strategic response taking shape in the trade press deserves attention — and a clear label. One early analyst take frames the launch as expanding generative engine optimization beyond “getting cited” into “shaping how visuals are generated.” The argument: high-value visual-first queries now get satisfied by Google itself, so brands should influence the model’s raw material — structured data and semantic product descriptions, consistent visual brand signals (colors, motifs, shapes) across all published imagery, and consistent authoritative terminology so a brand’s language becomes the model’s default vocabulary for a category.

Be precise about what that is: analyst speculation from a single third-party site, not a confirmed Google mechanism. Google has published nothing that says structured data or visual consistency influences what its image models generate inside AI Overviews. The honest framing is that these tactics are low-cost hedges with plausible mechanisms — models train on and retrieve from the web your brand publishes — not levers with documented effects. Anyone selling “visual GEO” with guaranteed outcomes this month is ahead of the evidence.

Our own position is that the citation layer is still where the measurable game is played. Our 1,000-query AI Overviews citation study found that structural factors — schema, freshness, content depth — correlate with getting cited in the text layer, and text citations remain the only AI Overviews outcome you can currently audit at scale. Treat visual-generation influence as a research question to monitor, not a deliverable to buy. It’s the approach we take in our agentic SEO engagements: instrument what’s measurable, hedge what’s plausible, and skip what’s invented.

There is also a wider industry current worth naming. In the visual licensing market, buyers are increasingly demanding provenance signals — such as C2PA manifests from the Coalition for Content Provenance and Authenticity — to prove whether an image is AI-generated, driven by brand-safety concerns as much as copyright, according to industry analysis of the stock-photography market. That trend predates and stands apart from this Google launch — no source ties the two — but it points the same direction: as generated imagery spreads, proving what’s real becomes a brand asset in its own right.

The rollout is measured in weeks, English-first — which means there is a short window to act before generated visuals become the default answer for your category’s visual-intent queries. Four moves, ranked by evidence.

The forward projection worth planning around: generated visuals will not stay confined to generic inspiration queries. The history of AI Overviews is a history of scope expansion — from simple questions to product research to local intent — and the visual layer should be expected to follow the same curve. Brands that build the content pipeline now — real photography, consistent visual identity, structured product data published at scale — will be the ones whose raw material the models have actually seen. If that pipeline is the bottleneck, our content engine builds exactly that publishing capacity.