Not Just Text: How to Optimize Video and Infographics to Appear in Multimodal AI Answers

Search engines have stopped answering with text alone. Google AI Overviews, ChatGPT, Perplexity, and Gemini weave images, charts, and clips into their answers, and the model decides which ones to show. The visual asset itself is a separate signal here, not just a decoration for the text.

Models are getting better at recognizing images, but in search they rarely analyze every pixel on a page, because at the scale of the whole internet that is too costly and slow. Most often they reach a graphic through the layer of text and metadata: alt text, caption, transcript, structured data, and the surrounding text, and it is that layer that decides whether the material even enters consideration.

Full machine “vision” does happen, especially when an image goes straight to the model, for example uploaded in a query or through Google Lens. That is why, for now, solid attributes and a text layer are the safer route, and classic image SEO, focused on speed and thumbnails, is no longer enough.

Optimizing visual content for AI has two goals: getting the model to show your image or clip in an answer, and getting it to cite what they convey. The same layer of text and data that describes the visual leads to both.

A video without text is almost invisible to a model, so a transcript is your best friend when optimizing video for AI.

The order of actions:

1. Publish the transcript in your page code: visibly, in an expandable “Show transcript” block, or in VideoObject structured data. Add captions to the recording itself, ideally edited, not only automatic.
2. Split the material into chapters with timestamps, and phrase the chapter titles as questions.
3. Fill in VideoObject: name, description, thumbnailUrl, uploadDate, duration, and transcript.
4. Use hasPart with startOffset and endOffset so the model can point to a specific moment.
5. Prepare your own thumbnail with readable text, not a random frame.
6. Host the recording on YouTube and embed it on your site.

The difference comes down to the goal: SEO fought for a thumbnail in the results, AISO fights for a citation in the assistant’s answer. You will prepare the same file differently for a search engine and for a model.

Optimizing visual content works best on a well-structured page, where all the content is written to be cited by AI. A model cites what it can extract as a whole and match to the question. Eight universal tips:

1. Start with the answer. Put a concise, self-contained answer of 30 to 50 words at the start of the page and every section.
2. Phrase headings as questions, the way a user would ask an assistant.
3. Use tables and lists, because a model pulls facts from them most easily.
4. Give statistics with a link to the source. Adding statistics, quotations, and references boosts visibility in generative engines by up to 40%, as shown by the GEO study from a team at Princeton and IIT Delhi (1)(2).
5. Write self-contained paragraphs; each should carry a full thought and stand on its own.
6. Add an FAQ section and mark it up with FAQPage data, and the article with Article data including datePublished and dateModified.
7. Update the content every 3 to 6 months. Freshness of information matters here.
8. Mind E-E-A-T: a named author, consistent terminology, and semantic code.