Google reactivates AI image generation in AI Overviews, but only partially and unevenly

Google has resumed generating images with artificial intelligence directly inside AI Overviews, the panel of AI-generated answers that appears at the top of search results. The information comes from Search Engine Roundtable, published on November 13, 2026 by editor Barry Schwartz, who says he managed to reproduce the behavior in some browsers, but not in all of them.

In other browsers, according to Schwartz's own tests, the AI Overview simply replies that it "cannot generate images directly." The feature was first spotted by Derek Hobson on LinkedIn, and the inconsistency between sessions suggests a partial rollout, not an official launch announced by Google.

Not the Same Feature Announced in July

This specific feature, automatically generating an image inside the AI Overview response, had already been announced by Google in July 2026 and paused in August of the same year, after some controversy the company did not publicly detail. In July, Google's statement read as follows:

To help bring those unique ideas to life, we're bringing image generation directly into AI Overviews in Search.

Google, official statement, July 2026

A month later, in August, Robby Stein, from Google, announced the pause and made a point of distinguishing this feature from another one launched around the same time:

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.

Robby Stein, Google

This distinction matters. Nano Banana is an image generator that the user triggers on purpose, explicitly asking for an image. The feature that has now returned, by contrast, triggers on its own, inside the summary the AI already builds for any search, without the user explicitly asking for an image.

How It Works in Practice

When the AI Overview decides that an answer would benefit from an illustration (an abstract concept, a fictional scene, a visual comparison), it calls an image generation model and inserts the result directly into the answer panel, without the user ever leaving search. Barry Schwartz speculates that the feature's return may be linked to an internal update he calls "Gemini 4 Hydrogen" inside Google Search, though he himself acknowledges this is just a guess, without confirmation from Google.

The practical problem, according to the editor's own account, is that the generated image may not resemble what was requested at all. In the example exchanged between Schwartz and Hobson, the same type of request produced quite different visual results on each attempt, including an image that was supposed to resemble a specific person and came out completely different.

What Changes for SEO and Brand Management

Here's the point that matters to anyone working in search, growth, and martech: AI Overview is already the layer that cuts clicks to the source site the most, because it delivers the ready-made answer right on the results page itself. Now that layer gains the ability to create images, not just text, without asking anyone's permission. This raises a new brand representation risk that traditional SEO never had to manage.

• Product image out of your control: if a user searches for something about your product's category and the AI Overview decides to illustrate the answer with a generated image, that image doesn't come from your catalog, your official photo, or your Merchant Center feed.
• Inconsistency between sessions: as Schwartz himself showed, the feature appears for some users and not for others, which makes it impossible to predict when it will trigger on a search relevant to your brand.
• No visible warning that the image was made up: nothing in the report indicates that Google clearly and prominently flags that image as AI-generated rather than a real photo of the product or scene.

In short: technical SEO has always worked to control what appears on the results page (title, snippet, image, rich result). With automatic image generation inside AI Overview, part of that surface is now filled by a generative model that doesn't necessarily use the visual asset the brand produced.

What Can Be Done Now

Since the feature is still in partial rollout and without official confirmation from Google, there's no ready-made setting to "disable" it or "guarantee" that the real image appears. Still, there are steps that make sense for anyone tracking their presence in AI Overviews:

1. Monitor brand and category queries using tools that capture AI Overviews content, not just traditional ranking, to flag when a generated image appears associated with your niche.
2. Keep structured image data (the ImageObject and Product schema) complete and up to date, since this has historically been the way to increase the chance that Google prioritizes a real asset instead of creating one from scratch.
3. Track Search Console to separate a CTR drop caused by ordinary zero-click behavior from one caused specifically by a visual answer that replaces a click to your site.
4. Document screenshots whenever a generated image misrepresents the product or brand; this is the kind of evidence that helps both in reporting to Google and in internal content strategy decisions.

What's Still Unknown

Google has not publicly confirmed the feature's return or explained why it's unstable across browsers. There's also no official indication of whether this is a controlled test, a rollout bug, or the start of a gradual reactivation. Until the company comments, the safest approach is to treat this as a sign that visual generation inside search is a front Google keeps testing, even after pausing it amid controversy, and that it may reappear in other forms before becoming official policy.