Guide · Images

Where do images in AI answers come from? How to get yours used

Updated 2026-11-07 · kagentx

Short answer

No official document says how images in AI answers are chosen. Google's AI features documentation says there are no additional technical requirements: the page just needs to be indexed and eligible to be shown with a snippet. The OpenAI and Perplexity pages we read do not mention images, and ChatGPT's help article could not be read, so that is marked not verified. What you can do is ordinary image optimization: descriptive alt on content images, images in the body as img, and where useful an image sitemap and ImageObject; none of them guarantees display.

What do the three official docs say about images in AI answers?

Only Google's documentation has explicit image rules, and they are the ordinary image-search rules, not a special clause for AI answers. The official OpenAI and Perplexity pages we read do not say how images are chosen. The table below is what we opened on 2026-11-07; anything we could not read is marked "not verified".

ProductWhat the official documentation saysStatus
Google AI Overviews / AI ModeLink requirement: the page is indexed and eligible to be shown with a snippet; "no additional technical requirements", no special optimization or new markup needed. The general practices list includes "high-quality images and videos, when applicable"ai-features page, read 2026-11-07; the page shows no update date
Google Imagesalt is the most important image attribute; supported formats are BMP, GIF, JPEG, PNG, WebP, SVG and AVIF; an image sitemap can be submitted; structured data can add a badgegoogle-images page, read 2026-11-07; the page shows no update date
ChatGPT searchOAI-SearchBot lets a site appear in ChatGPT search; robots.txt changes take about 24 hours; the crawler page does not mention imagesdevelopers.openai.com crawler page, read 2026-11-07; the help-center article returned 403, so how images are chosen: not verified
PerplexityThe relevant developer-doc pages say nothing about images or how sources are chosen. Third-party reports mention a licensing deal with an image library; we did not read an official statementRead docs.perplexity.ai on 2026-11-07; image source: not verified

So no document promises "do X and AI answers will show your image". What you can do is make images readable the way search systems usually read them; the steps below follow that idea.

Step one: is the alt text on your images right?

Start with alt. Google's image documentation calls it the "most important" attribute for adding information to an image: describe what the image shows, fit the context of the page, and do not stuff keywords. An empty alt is only for purely decorative images (such as a logo). If a content image has none, search systems can only guess from the file name and the surrounding text.

<!-- Bad: empty or filename -->
<img src="/img/IMG_2043.webp" alt="">
<!-- Better: says what the image shows, in the context of the page -->
<img src="/assets/guide/size/canvas.webp" width="900" height="802"
     alt="A sticker on a 370×320 px canvas with the 10 px margin marked">

We checked one MochiStamp guide page

On 2026-11-07 we read the HTML of MochiStamp's size guide page with curl: 9 img tags in total; 2 body images have descriptive alt (the sticker canvas diagram and the three file sizes side by side) and 7 have empty alt (2 site icons and 5 related-card images at the bottom). Both body images sit in a figure with a figcaption.

CheckResult on 2026-11-07
Total img tags9
With descriptive alt2 (the two body images)
With empty alt7 (2 site icons, 5 related-card images)
figure + figcaption2 pairs
image in the Article structured dataAn array listing the URLs of the 2 body images
ImageObject typeNone
og:image/assets/og-en.png
Images in the sitemap7,132 image:loc entries in sitemap.xml (image sitemap extension)

What this page does right: alt on the body images, figcaption, and image in the Article data. What it has not done yet: alt on the 5 card images (they are previews of other pages, so the target page's title would do) and ImageObject. We have no data on this page's images being used in AI answers, so we cannot say how much these practices help.

For comparison, kagentx.com's own sitemap.xml, read on 2026-11-07, has no image extension (0 image:image entries). Our images are mostly interface screenshots and icons, so this is a gap we have not closed.

Is an image sitemap worth it?

Yes, but it is not the first priority. Google's sitemap documentation requires only two tags: image:image, which wraps one image, and image:loc, the image URL. Each url can hold up to 1,000 images, and the caption, geo_location, title and license tags have been removed from the documentation, so you no longer need them. An image sitemap may also list image URLs on other domains, so images on a CDN are fine.

<url>
  <loc>https://example.com/guide/size</loc>
  <image:image><image:loc>https://example.com/assets/guide/size/canvas.webp</image:loc></image:image>
  <image:image><image:loc>https://example.com/assets/guide/size/pack.webp</image:loc></image:image>
</url>
<!-- needs xmlns:image="http://www.google.com/schemas/sitemap-image/1.1" on <urlset> -->

It helps most when images only load through scripts or a page has many images, because those may not be discovered directly; for ordinary pages with img tags in the body it is a supplement. Whether AI crawlers run scripts is covered in Do AI crawlers render JavaScript.

What should ImageObject structured data contain?

Google's image license metadata documentation says contentUrl is required, plus at least one of creator, creditText, copyrightNotice and license; the Licensable badge also needs license. It also states that structured data is "not guaranteed" to be shown, and the page must be crawlable without login. So it suits sites with original images that want to state copyright and licensing; it is not a switch for display.

{
  "@context": "https://schema.org",
  "@type": "ImageObject",
  "contentUrl": "https://example.com/assets/guide/size/canvas.webp",
  "creator": { "@type": "Organization", "name": "Example" },
  "creditText": "Example",
  "copyrightNotice": "© 2026 Example",
  "license": "https://example.com/image-license",
  "acquireLicensePage": "https://example.com/image-license"
}

For structured data in general, see Does structured data help AI search?.

Do it in these steps

  1. 1. Write alt for every content image

    Check page by page: fetch the HTML with curl and count img tags with empty alt; add a description to content images and leave decorative ones empty.

  2. 2. Put images in the body with an img tag

    Google's image documentation says it reads the src of img. Background images (CSS) and images produced only by scripts are more likely to be missed.

  3. 3. Make the file name and surrounding text say what the image is

    Do not name files IMG_2043; put a sentence or a figcaption next to the image.

  4. 4. Add an image sitemap, and ImageObject for original images

    The first helps discovery and the second states copyright; neither guarantees display.

  5. 5. Make sure crawlers can get in

    The page holding the image must be crawlable first. OpenAI's page says to allow OAI-SearchBot to have a chance of appearing in ChatGPT search; how to write it is in How to write robots.txt for AI crawlers.

  6. 6. Ask an AI your own questions

    Ask an AI with web search on the questions your buyers ask, and see whether your pages appear in the sources. Video works similarly, see How YouTube videos show up in AI answers; for Google, see How to appear in Google AI Overviews.

FAQ

Where do the images in AI answers come from?

Google's AI features documentation has no image-specific rules; it only says the page must be indexed and eligible to be shown with a snippet. The OpenAI and Perplexity pages we read do not say how images are chosen, and ChatGPT's help article could not be read, so those two are marked not verified.

Does alt text affect whether an image appears in AI answers?

Google's image documentation calls alt the most important attribute for adding information to an image. Whether it directly decides which image appears in an AI answer is not stated officially; what is certain is that it helps search understand the image, and lets people and programs who cannot see it understand it too.

Do I have to make an image sitemap?

Not necessarily. Google says an image sitemap helps discover images it "might otherwise miss"; img tags in the page can already be discovered. It helps more when images load through scripts or sit on a CDN under another domain.

Does ImageObject structured data guarantee that an image is shown?

No. Google's documentation states that structured data or IPTC metadata is not guaranteed to be shown. It states copyright and licence information and gives images a chance at the Licensable badge.

How do I know whether AI used my images?

No provider offers an "image cited" report. What you can do is ask an AI your buyers' questions and see whether your images or pages appear in the answer and sources, then check the Image search type in Search Console for impressions.

Sources

kagentx

Can crawlers read your pages?
Let the evaluation tell you

The evaluation checks whether crawlers can get in and whether the text is served directly, then asks ChatGPT, Perplexity and Claude (with web search on) your buyers' questions and sees whether you are mentioned. The Basic plan is $1, and results usually take 2–3 minutes.