Are your charts and infographics getting cited by AI? Here's how we optimized our visual content

Visual content optimization for AI is increasingly important as systems become multimodal. Here’s what works:

Alt text best practices:

Don’t describe WHAT the image is. Describe what INSIGHT it provides.

Bad alt text: “Bar chart showing revenue by quarter”

Good alt text: “Bar chart showing Q4 revenue growth of 25% year-over-year, outperforming Q1-Q3 averages by 12 percentage points”

The second version gives AI extractable information it can cite.

Optimal length: 80-125 characters. Long enough to convey insight, short enough to be useful.

The processing chain:

AI systems use multiple signals:

1. Alt text (primary for non-multimodal queries)
2. Caption text
3. Surrounding paragraph text
4. Filename
5. ImageObject schema
6. Visual analysis (for multimodal systems)

Optimize all of them, not just one.

Schema markup absolutely helps for AI visibility.

ImageObject implementation:

{
  "@type": "ImageObject",
  "contentUrl": "/images/revenue-chart.png",
  "caption": "Q4 2025 revenue growth of 25% YoY",
  "description": "Bar chart comparing quarterly revenue with 25% growth in Q4",
  "representativeOfPage": true
}

Why it works:

1. Explicit signals - Tells AI exactly what the image represents
2. Removes ambiguity - AI doesn’t have to guess from alt text alone
3. Priority indication - representativeOfPage marks key images

Testing results:

Sites with ImageObject schema on key visuals see 35% higher AI citation rates for image-related content.

Quick implementation:

Most CMS platforms have schema plugins. Add ImageObject to featured images and key data visualizations.

We changed our content process to optimize visuals for AI from creation.

The new workflow:

1. Planning: Define the key insight the visual will show
2. Design: Ensure all labels are in the image, not implied
3. Alt text: Write before the image is created (insight-focused)
4. Caption: 40-80 words explaining the takeaway
5. Context: Surrounding paragraph explicitly references the visual

The insight-first approach:

Before creating any visual, we ask: “What specific claim do we want AI to be able to cite from this?”

Then we design and optimize the entire visual package around that citeable claim.

Results:

Visuals created with this process get cited 4x more than our legacy visuals.

On the question of whether AI can read visuals directly - yes, increasingly.

Current state:

• GPT-4 Vision: Can interpret charts and extract data
• Gemini: Strong multimodal understanding
• Claude: Solid visual analysis capabilities
• Perplexity: Primarily text-based retrieval still

But here’s the catch:

Even with visual understanding, AI systems still rely heavily on text signals. Why?

1. Text is faster to process at scale
2. Text signals are more reliable
3. Visual analysis has higher error rates

Practical implication:

Don’t rely on AI’s visual understanding. Optimize text signals (alt, caption, context) as if AI can’t see your images at all. Visual understanding is a bonus, not a baseline.

We publish original research with lots of data visualizations. Here’s what we’ve learned:

What gets cited most:

1. Comparison charts - “[A] vs [B]” visuals
2. Trend charts - Showing changes over time
3. Stat highlights - Large numbers with context
4. Tables - AI loves structured data

What doesn’t work:

1. Complex multi-variable charts - Too hard to parse
2. Artistic infographics - Style over substance
3. Charts without axis labels - Incomplete information
4. Images with text burned in - AI can’t read overlay text reliably

The golden rule:

Every visual should be citeable as a single, specific claim. If you can’t express it in one sentence, the visual is too complex for AI to cite.

Accessibility optimization and AI optimization overlap significantly.

The connection:

Both require visuals to be understandable without seeing them:

• Accessibility: For screen readers and users who can’t see
• AI: For systems that process text signals first

What accessibility taught us:

1. Alt text should convey the PURPOSE, not just appearance
2. Complex visuals need extended descriptions
3. Data should be available in text form (table alternatives)
4. Color shouldn’t be the only differentiator

Double benefit:

Properly accessible visuals are inherently more AI-friendly. You’re optimizing for both at once.

Quick audit:

If a screen reader user could understand your visual from its text signals, AI probably can too.

Video perspective: similar principles apply to video thumbnails and frames.

What we’ve learned:

1. YouTube video descriptions get cited, not the video itself
2. Thumbnails with clear text get more AI mentions
3. Video transcripts are goldmines for AI citations
4. Chapters/timestamps help AI find specific moments

For static visualizations:

Consider creating video explainers for key data. The transcript gives you another text signal layer, and YouTube is heavily indexed by AI systems.

Example:

A 2-minute video explaining our annual survey data gets more AI citations than the static infographic, because the transcript provides rich text context.

This discussion has given me a complete optimization framework.

Key takeaways:

1. Alt text should describe the INSIGHT, not just the visual
2. ImageObject schema increases citation rates by ~35%
3. Captions and surrounding text are critical signals
4. Simple, citeable visuals outperform complex ones
5. Accessibility optimization = AI optimization

Our new visual content checklist:

Before publishing any visual:

• Alt text (80-125 chars, insight-focused)
• Caption (40-80 words, key takeaway)
• ImageObject schema markup
• Surrounding paragraph referencing the visual
• All axis labels and data points visible
• Single citeable claim identifiable

Process change:

We’re now writing alt text BEFORE creating visuals. Define the insight, then design to support it.

Tracking:

Using Am I Cited to monitor visual content citations and iterate on what works.

Thanks everyone for the practical guidance - this will significantly change how we approach data visualization.