Multimodal AEO

AI answers increasingly include and cite video, images, and audio — and the engines "understand" that media largely through its text: transcripts, captions, alt text, and structured data. This skill makes a brand's multimedia surfaceable and citable by answer engines. It works for any business, B2B or B2C. The honest core principle: engines mostly read the text around and inside media, so the highest-leverage move is almost always to give them clean, accurate text.

When to use this skill

• Optimizing video or YouTube content to be cited in AI answers.
• Making images and diagrams understandable to multimodal models.
• Getting podcasts/audio surfaced via transcripts.
• Planning media that AI answer surfaces will pull from.

Before you start

Read aeo-foundation.md for the audience, prompts, and which media the brand has or can make. If it is missing, point to the aeo-foundation skill.

The levers

1. Transcripts and captions — non-negotiable. Provide accurate, full transcripts and closed captions for every video and audio asset. Text is how engines parse, quote, and cite audiovisual content; without it, the asset is largely invisible to them.
2. YouTube presence. YouTube is a Google-owned surface that Google AI Overviews and Gemini draw on heavily. Use clear, question-shaped titles, thorough descriptions, chapters, and uploaded transcripts. Pairs with gemini-optimization and google-ai-overviews-optimization.
3. VideoObject schema. Mark up videos (name, description, thumbnail, uploadDate, transcript/caption URL) so machines understand them — see structured-data and the schema cookbook.
4. Answer-first in the media itself. State the answer early (the first lines of a video or its first chapter), and name chapters as the questions people ask — the same extractability logic as answer-engineering, applied to time-based media.
5. Images and diagrams. Use descriptive alt text, real captions, surrounding context, and meaningful filenames. Make the image genuinely answer a visual question; multimodal models read both the pixels and the text around them.
6. Audio and podcasts. Publish full transcripts and detailed show notes; that text is what gets indexed and cited, not the audio.
7. Text twins. Pair each major media asset with a written version (article, summary, or transcript page). Engines cite clean text most readily — see ai-capability-signaling.

Workflow

1. Inventory media tied to high-value prompts (videos, key images/diagrams, podcasts).
2. Add or fix transcripts/captions for every audiovisual asset — accurate and complete.
3. Optimize YouTube items: question-shaped titles, full descriptions, chapters, uploaded transcripts.
4. Mark up videos with VideoObject and images with descriptive context.
5. Apply answer-first structure inside the media (early answers, question chapters).
6. Publish text twins for the assets that matter.
7. Verify the text is reachable and readable (see ai-retrievability).

Output format

• A media optimization plan: which assets, what to add (transcripts, schema, twins).
• Transcript/caption status and gaps per asset.
• YouTube and image optimizations to apply.
• (assumption) tags where media inventory or platforms were unknown.

Common pitfalls

• Audiovisual content with no transcript — effectively invisible to engines.
• Assuming engines "watch" or "look" rather than reading the surrounding text.
• Ignoring YouTube when targeting Google/Gemini answers.
• Thin alt text and missing captions on images that answer visual queries.
• No text twin, so clean media never becomes quotable text.

Related Skills

• structured-data — VideoObject and image markup.
• answer-engineering — answer-first structure, applied to media.
• gemini-optimization / google-ai-overviews-optimization — the engines that lean on YouTube and visual results.
• ai-capability-signaling — text twins of media for clean ingestion.