Google has started showing AI generated images inside AI Overviews in Search, according to a Search Engine Roundtable report published on 14 August 2026. The capability is the one Google announced roughly a month earlier, and the sighting that made it visible came from a recipe publisher: Inspired Taste posted an example on X on 13 August 2026 in which an AI Overview rendered AI generated cartoon step illustrations for a spaghetti squash recipe.
This is a rollout story, not an announcement story. Relevant Audience covered the announcement of image generation in AI Overviews on 16 July 2026, when it was a stated capability with no live examples attached. What changed in August is that the images began appearing in real results, and the first people to notice were publishers in a vertical where original step photography is the product.
What Google said it was building
Google's wording when it announced the feature was that it was bringing image generation directly into AI Overviews in Search, framed as a way to help bring ideas to life. That framing treats images as an illustration layer on top of an answer. The objection that surfaced in August is about what that layer displaces.
Search Engine Roundtable, which reported the rollout in this write up on 14 August, recorded the publisher reaction as unsurprising. Schwartz noted the post tagged Google's Robby Stein and Rajan Patel. The source does not report any Google response to the complaint.
The timeline in three dates
Here is the sequence as it stands, using only what has been reported. It is short, and the gap between the middle and the last row is the part that matters for anyone who filed the July news and moved on.
| When | What happened |
|---|---|
| About a month before the sighting | Google announces image generation coming directly into AI Overviews in Search. Relevant Audience covered that announcement on 16 July 2026. |
| 13 August 2026 | Recipe publisher Inspired Taste posts an example on X showing an AI Overview with AI generated cartoon step illustrations for a spaghetti squash recipe. |
| 14 August 2026 | Search Engine Roundtable reports that Google has begun rolling out AI generated images inside AI Overviews. |
The publisher example, and why it landed the way it did
The detail that gives the Inspired Taste post its force is timing. The publisher wrote that they were in the middle of updating that exact recipe when the AI generated version appeared. Their article and recipe video already show how to safely cut a spaghetti squash, which is a genuinely dangerous step, how to add flavour with herbs, and how to cook it faster in a microwave.
Their objection was blunt. They described the AI Overview as "directly trying to replace creators who buy groceries for testing, photography, and filming recipe videos", and said the simplification strips out the work that made the recipe worth publishing. Whether or not you agree with the framing, the underlying observation is testable: the illustrations in that AI Overview were generated, and the photography they sit in place of was not.
The attribution problem is the part with legal and brand edges
Inspired Taste raised a second point that is easy to miss and harder to shrug off. A searcher looking at an AI Overview that cites several websites and displays images may reasonably assume that one of those cited sites produced the images. It did not. The images were generated by Google, and the citations sit next to them.
That creates an association nobody agreed to. If a generated illustration shows a technique done differently from how a cited publisher teaches it, or shows a result that does not match what their recipe produces, the visual and the attribution are pointing at each other in the same block of screen. The source does not report whether Google labels these images in a way searchers notice, and it does not report whether cited publishers can opt out of appearing alongside them.
Why image led verticals lose a differentiator here
Until now, the practical answer to "what can an AI Overview not do to my content" had a reliable item on the list: it could not produce the pictures. Text could be summarised, structure could be flattened, the ordered steps could be extracted. Original photography, step imagery and process shots were the part that required someone to buy the ingredients, cook the thing and photograph it going wrong twice.
If AI Overviews can generate a plausible looking step illustration, that item comes off the list. The verticals where this bites hardest are the ones where the image is the explanation rather than decoration:
- Recipes and food, where step imagery carries the technique and the finished shot carries the promise.
- How to and repair content, where a photo of the actual part in the actual position is the difference between following along and guessing.
- Product demonstration, where the image proves the thing exists and works.
- Travel, where photography is often the entire reason a page gets shared.
- Home improvement, where before and after images do the persuading.
The strategic response is not to stop making images. A generated illustration can look like a technique without being the technique, and the gap between a plausible drawing and a photograph of the real thing is where a publisher still has something to sell. What changes is that the image alone no longer proves you did the work. The proof has to be visible in the specifics: the failure modes, the timings, the substitutions, the shot of what it looks like when it goes wrong. That is the same argument behind serious content marketing in an AI answer environment, and it is worth revisiting how your pages are structured for AI search visibility now that the visual layer is in play too.
What the source does not say
The limits here are wide and worth stating plainly, because this is a single report built on one live example.
- The source does not say which markets or languages the feature is live in.
- The source does not give a rollout percentage or say how many queries trigger a generated image.
- The source does not say whether cited publishers can opt out of having generated images shown alongside their citation.
- The source does not report a Google statement on the attribution concern, or on the publisher complaint at all.
- The source does not say whether generated images appear for query categories other than recipes.
Anyone telling you this is now standard behaviour across all AI Overviews is going beyond what has been reported. One documented live example plus a report that a rollout has begun is what exists as of 14 August 2026.
What to actually do this week
- Run your top image led queries yourself and record what you see, with the date. This is the only way to know whether generated images are appearing for your topics rather than for recipes in general.
- Note whether your pages are cited in those AI Overviews, and whether generated images are displayed in the same block. That pairing is the attribution risk in concrete form.
- Keep a dated record. If the behaviour changes, or if labelling changes, you want a before state to point at.
- Check that your own images are properly marked up and attributable on your pages, so that the version a person can click through to is unambiguously yours.
- Do not rewrite a content strategy on one example. Watch it for a few weeks first and see how far it spreads.
What this means for Thai marketers
The honest starting point is that nobody has reported this appearing in Thailand or in Thai language results, and the source says nothing about markets or languages at all. Do not tell a client this is live here. The correct statement is that it has started rolling out and that the reported example is in English.
The practical move for Thai teams is the same one that has worked for every AI Overviews change: check your own queries rather than reading a foreign report as a forecast. Thai language recipe, travel and home improvement publishers are exactly the profile that would feel this first if it does arrive, and they are also the ones with the most original photography at stake. If you manage an image heavy Thai site, the cheap action is a dated screenshot record of your top twenty queries now, so that a change is visible when it happens. That record also feeds the ordinary work of tracking SEO performance in Thailand, where visual results have been shifting for a while.
Frequently asked questions
Is this live in Thailand?
The source does not say. Search Engine Roundtable reports that the rollout has begun and shows one example, without stating which markets or languages are covered. Until you see it in your own results in Thai, treat it as unconfirmed for this market.
Can I stop Google from generating images next to my citation?
The source does not say whether an opt out exists. It reports the publisher raising the attribution concern and does not report a Google response or any control offered to cited sites.
Does this mean original photography is no longer worth the cost?
No, and nothing in the report supports that conclusion. What changed is that a generated illustration can now occupy the visual slot in the answer, so the photography has to earn its place further down the click, through specifics a generated image cannot fake.
Which types of searches show generated images?
The reported example is a recipe query, and the source does not say whether other categories are affected. Image led categories are the obvious candidates, but that is reasoning about the feature rather than something Google or the report has confirmed.
Do I need to change anything on my site right now?
Not urgently, no. The useful work this week is observation: run your image led queries, record what appears, and note whether your pages are cited alongside generated images. Strategy changes can wait for more than one data point.
If you want help working out which of your pages sit in the categories most exposed to this, and what the citation picture looks like for them, that is a reasonable conversation to have now rather than after the pattern spreads.
