Google is bringing back the ability to generate images on the fly inside AI Overviews, for some searchers. The feature was first announced in July 2026 and paused in August after controversy — and its quiet return raises fresh questions about what AI Overviews will show instead of the open web.
Google first announced image generation inside AI Overviews in July 2026, letting the AI answer create an image to illustrate its response rather than pulling one from a publisher's page. The test did not last long. Google paused it in August after controversy, and at the time Google's Robby Stein explained the decision plainly: "This was a small experiment we're no longer running. FYI this is not the Nano Banana feature we announced in July, which only triggers when users explicitly ask to generate an image." That distinction mattered: the paused feature generated images automatically as part of an answer, while Nano Banana only creates an image when a user explicitly asks for one.
Now the automatic version appears to be back, at least for some searchers. The return was spotted by Derek Hobson, who posted about it on LinkedIn, and Search Engine Roundtable's Barry Schwartz replicated the behavior in some browsers but not others. In browsers where it does not work, the AI Overview replies "I cannot generate images directly." That uneven rollout is typical of Google's search experiments: a feature can be live for one user and absent for another, with no announcement and no way to know which side of the test you are on.
What is not yet clear is how widely the feature has returned, or whether this is a limited test or the start of a broader rollout. Google has not announced the return, and the spotting so far comes from individual users rather than any official channel. That leaves the SEO community in the usual position of watching, screenshotting, and comparing notes while Google says nothing.
The August pause is worth remembering because it tells us something about how Google itself views this feature. A small experiment that gets switched off within weeks after controversy is not a feature Google was confident about, and bringing it back three months later suggests either that the underlying concerns were addressed or that the strategic pull of image generation inside search proved stronger than the backlash. Either way, the history argues for caution: a feature that has already been killed once can be killed again, and building a strategy around it would be premature.
There is also a meaningful difference between an image a user asks for and an image an AI answer invents on its own. When a user explicitly requests an image, the generation is a tool doing what it was asked to do. When an AI Overview generates an image unprompted, it is choosing to show synthetic visuals where a real photograph, chart, or illustration from a publisher might otherwise have appeared. That substitution is the part publishers should watch, because it is a direct trade between content Google creates and content the open web creates.
Barry Schwartz speculated that the return may be related to the Gemini 4 Hydrogen update in Google Search, but that is unconfirmed speculation, not a fact. It is a plausible guess — a new model generation often unlocks capabilities that were previously held back — but Google has said nothing linking the two. The honest position is that the connection is a theory, and theories about Google's internal wiring have a poor track record.
For SEOs, the immediate action is observation, not reaction. Check whether AI Overviews in your key query areas are showing generated images, note which browsers and accounts show them, and keep screenshots with dates. The feature is inconsistent across browsers right now, so a single check proves very little; the pattern across a week of checks is what will tell you whether this is expanding or stalling.
For link builders, the return of generated images is a reminder that the visual assets you create have two audiences now: users who find them through search, and AI systems that may or may not show them. Original photography, diagrams, and illustrations remain the assets that AI cannot generate from nothing — they are grounded in real products, real places, and real expertise. That grounding is exactly what makes them worth linking to, and it is the quality that a generated image cannot fake.
The bigger lesson is about the direction of travel. AI Overviews keep absorbing more of the search result into themselves: text, definitions, sources, and now images again. Each addition makes the Overview more self-contained and the click-through less necessary. The sites that keep earning traffic in that world are the ones whose content is genuinely better than what an AI can synthesize — original, specific, and worth the visit.
Sources: Search Engine Roundtable, November 13, 2026.
Related reading on this site: Google Tests Moving AI Overview Source Icons to the Bottom, Google Expands AI Overview Definitions by Default, Google Moves Brand AI Overviews to the Right Column, Gemini 4 Hydrogen Is Now the Default Model in AI Mode for Everyone, YouTube Leads AI Overview Citations at 24.5%, and AI Overviews Now Cover Over 80% of UK Keywords, SISTRIX Finds.
Linkbaits helps you create content that earns citations and backlinks. Start with the toolkit and build your first linkable asset this week.
Start free trial