On August 13, the person behind recipe blog Inspired Taste posted a screenshot on X. A search for "spaghetti squash recipe" had returned an AI Overview with a step-by-step illustration summarizing the cooking process, start to finish — nobody had asked for an image, there was no "AI-generated" label, and no credit to any source site. As creators pushed back, Google's VP of Search, Robby Stein, confirmed on X on August 17: "This was a small experiment we're no longer running." He added that it was separate from the "Nano Banana" image feature announced in July, which only generates images when a user explicitly asks for one.
Why this is worth noting
Most of this project's past coverage of AI Overviews has been about whether a site gets cited in an AI answer — clicks shrink, but a citation is still organic exposure. This case is different in kind. A recipe creator buys the ingredients, cooks the dish, shoots and edits the photos. Here, an AI drew that output for them and showed it in the search result — not a citation, a replacement. Creators' objection was direct: Google was using AIO to stand in for the work they put into ingredients, shooting, and editing.
Summarizing text and recreating an image are different orders of problem. Text summarization compresses information; a photo is itself the creator's finished output. AI summarizing text has been happening for years, but once an AI starts recreating the output's form too, one more reason to click through to the original page disappears.
What's confirmed and what isn't — that the test existed, and that Google confirmed shutting it down on August 17, is reported consistently by both Search Engine Land and Search Engine Roundtable, each quoting Robby Stein's X post directly (checked August 31, 2026). What isn't public: how widely the test ran across recipe queries, or how many users actually saw it — the only documented case so far is the one screenshot that started this.
If your site gets 100 visitors a week
This project's own content — news summaries — carries the same exposure, arguably worse: text is far easier for an AI to "write instead of you" than an image is to redraw. So the lesson here is defensive. Put things in your content that an AI can't fabricate on your behalf: a verification date and number ("as of August 31, 2026"), a link to the primary source, a detail you actually experienced yourself. That's exactly why this project's writing rules require "don't make things up" and a "what's confirmed and what isn't" box in every piece — an AI can imitate a plausible summary or a plausible illustration, but it can't manufacture "I checked this myself." One thing to do today: find a recent post that's missing a confirmed-as-of date and source, and add one.