Published August 18, 2026.
AI image optimization is now about more than alt text and file names. It is about making images work as visual proof inside AI answers from ChatGPT, Gemini, Perplexity, and Google AI Overviews.
The key ideas are original imagery, contextual anchoring, image context decoding, and preferred image signals such as primaryImageOfPage and og:image.
| Format | Notes |
|---|---|
| BMP | Supported by Google in search. |
| GIF | Supported by Google in search. |
| JPEG | Supported by Google in search. |
| PNG | Supported by Google in search. |
| WebP | More efficient compression than JPEG at similar quality. |
| SVG | Supported by Google in search. |
| AVIF | More efficient compression than JPEG at similar quality. |
The article describes multimodal systems using shared embedding space. It references CLIP, introduced by OpenAI in 2021 and trained on roughly 400 million image-caption pairs.
Images and text that match closely land near each other in that space. Generic or vague descriptions land farther apart.
This makes specific, accurate captions and surrounding text important for retrieval and reuse in AI answers.
In March 2026, Google added guidance on specifying a preferred image with metadata.
Google lists three ways to influence which image may be shown:
These signals do not make a page eligible for AI Overviews. They only help choose which image appears once a page is already being used.
Google Search Console and a standard image sitemap can confirm indexing. They do not show whether an image appears inside a ChatGPT, Gemini, or Perplexity answer.
The article says Verseodin tracks citations across ChatGPT, Gemini, and Perplexity at the domain level and can show when a citation pulls in a specific image versus surrounding text.
It is traditional image SEO plus context decoding, contextual anchoring, and preferred image signals for generative systems.
An image acts as evidence for a specific claim rather than decoration.
No. They only influence which image is selected after a page is already eligible.
The article says dedicated citation tracking is needed to see whether images appear in AI answers.