On August 13, 2026, a recipe publisher documented that Google's AI Overviews had generated a cartoon-style illustration for a spaghetti squash query - with no image request, no AI-generated label, and no source attribution. On August 17, Google VP of Product Robby Stein confirmed to Search Engine Land: "This was a small experiment we're no longer running."
What actually happened
For four days, from August 13 to 17, nobody publicly explained what was going on or under whose authority the images were being generated. Clarity only came after direct questions from journalists: Google confirmed AI Overviews has at least two distinct code paths for generating images.
The first is the officially announced feature from July 2026 (known as "Nano Banana"), which only generates an image when a user explicitly requests one. It continues to run unchanged.
The second was an unannounced path that triggered automatically: the system decided on its own that an answer deserved an illustration and generated one without any user request. This is the path Google paused.
Why this matters for content strategy
The incident exposed something previously discussed only in theory: AI Overview can visually interpret a page's content and present its own illustration as part of the answer - without linking to the source or notifying the content owner. For recipe publishers, how-to guides, and any content that naturally invites illustration (step-by-step processes, comparisons, diagrams), this is a direct signal.
No corrective action is needed right now - the unannounced path has already been shut down by Google itself. But it's worth keeping in mind when planning AI Overview content strategy going forward: the system is capable of autonomous decisions about how it presents content, decisions that were never publicly announced and could resurface in a different form.
What this reveals about AI Overview's opacity
Four days without an official explanation is telling on its own - it shows that even inside Google, there isn't always immediate clarity about which system made a given decision about presenting content. For SEO practitioners, the practical takeaway is: AI Overview behavior observed today doesn't guarantee the same behavior next week, and some changes will never be announced publicly at all - they'll only surface because an outside observer happened to notice and document them.
There's also a practical monitoring habit worth building from this: periodically saving screenshots of search results for key queries. That's precisely how this incident became public in the first place - without that documentation, the recipe publisher would have had no way to prove the illustration appeared unprompted, and the story likely never surfaces at all.