For thirty years, web design has meant designing for humans. People browse, read, click, scroll, and decide. Every design choice—typography, layout, imagery, interaction—optimizes for human perception and human behavior.

That era isn’t ending, but it’s no longer the complete picture. AI agents are becoming primary consumers of web content. They read pages on behalf of humans. They summarize, compare, and recommend. They extract data and synthesize answers. They make decisions about what to surface and what to ignore.

This changes what “good design” means. When the first reader isn’t a human but an AI that will interpret and represent your content, you need to think about design differently.

How AI agents use the web

Understanding the shift requires understanding how AI agents actually interact with web content.

Search assistants Like perplexity and searchgpt read web pages to answer user questions. They extract relevant passages, synthesize information from multiple sources, and present summarized answers with citations. Your page might provide the definitive answer to a question, but if the AI can’t extract that answer cleanly, a less authoritative source might be cited instead.

Shopping and research agents Compare products, analyze reviews, and synthesize recommendations. They need to extract structured information: prices, specifications, ratings, availability. If your product page buries this information in images or complex javascript, it becomes invisible to these agents.

Coding assistants Read documentation, tutorials, and API references to help developers. Clear, well-structured technical content gets surfaced; dense, poorly organized content gets ignored.

Personal assistants Help users with tasks that require gathering and comparing information: travel planning, event research, decision-making. They need to extract dates, locations, prices, and other factual content efficiently.

In each case, the AI is reading on behalf of a human who will never see your actual page. They’ll see the ai’s interpretation of your page. You’re not designing for the end user directly—you’re designing to be accurately represented by an intermediary.

What AI readers need

AI agents parse web content differently than humans. They excel at some things and struggle with others.

They’re good at:

• Extracting text from semantic HTML
• Understanding document structure from headings and sections
• Parsing structured data (schema.Org markup, tables, lists)
• Following explicit relationships (links, navigation)
• Reading alt text for images
• Processing consistent, predictable formats

They struggle with:

• Content rendered in images or video
• Information that requires javascript execution
• Implicit context that humans infer from visual design
• Content that’s visually prominent but semantically unmarked
• Dynamic content that changes based on user state
• Information spread across multiple pages without clear linking

The implication is clear: if important content isn’t in the HTML, accessible without JavaScript, marked up semantically, and structured clearly, AI agents may not be able to access it at all.

Practical design implications

This understanding leads to specific design recommendations.

Semantic structure matters more than ever. Headings should be actual headings (H1, H2, H3), not styled divs. Lists should be lists. Articles should use article tags. Sections should use section tags. This isn’t just accessibility practice anymore—it’s how your content gets understood and represented by AI intermediaries.

Put critical information in text, not images. AI agents can read alt text but can’t interpret complex information graphics. If key data lives in an infographic or a hero image, it’s invisible to most AI readers. That doesn’t mean avoiding imagery—it means ensuring that anything important enough to be in an image is also available as text.

Structured data is your metadata layer. Schema.Org markup tells AI agents exactly what your content contains: is this a product, an event, an article, a recipe? What are its key attributes? This structured data is increasingly how AI agents make sense of pages. Implementing it thoroughly and accurately improves how your content is understood and surfaced.

Answer questions directly. AI search assistants are looking for answers. If your page is about a topic, consider what questions users might ask and ensure those questions are answered clearly. A page about your product should answer: what is it, who is it for, what does it cost, how do I get it. These answers should be findable by automated parsing.

Maintain content stability. Content that changes frequently based on user state, a/b tests, or personalization is harder for AI agents to understand. They may cache versions that don’t match what users eventually see. Core informational content should be stable and consistent.

Consider the summarization. If an AI were to summarize your page in two sentences, what would it say? Is that summary accurate? Does it capture what you’d want conveyed? If your page is so unfocused that a good summary is impossible, that’s a problem for human readers too—but AI intermediation makes it visible.

The design for humans vs. Design for AI tension

There’s a genuine tension here. Some things that work well for humans work poorly for AI, and vice versa.

Rich visual design delights humans but may obscure content for AI. Interactive experiences engage humans but may be unparseable by AI. Conversational, personality-rich copy connects with humans but may be harder for AI to extract facts from.

The resolution isn’t to choose one audience over the other. It’s to layer the experience: a foundation of semantic, structured, accessible content that serves both AI readers and humans with assistive technologies, topped with visual design and interaction that enhances the experience for those who can access it.

Think of it as progressive enhancement for the age of AI: everyone gets the content, humans with full browsers get the rich experience, AI readers get what they need to represent you accurately.

What good looks like

A web page designed for both human and AI readers has certain characteristics.

The source HTML is readable and meaningful even without styling. Headings create a clear outline of the content. Important information is in text, not just images. Structured data provides machine-readable metadata.

The visual design enhances this foundation without replacing it. Imagery adds emotional resonance but doesn’t carry essential information alone. Interactive elements reveal content but don’t hide it behind inaccessible patterns.

The content itself is clear and direct. Questions are answered. Key information is easy to locate. The page can be summarized accurately.

When an AI agent reads this page, it can extract the key information, understand the structure, and represent the content accurately. When a human reads it, they get a rich, engaging experience. Neither audience is served at the expense of the other.

The emerging optimization game

Just as SEO emerged when search engines became primary discovery mechanisms, a new optimization discipline is emerging for ai-mediated discovery.

Today’s AI agents are relatively unsophisticated in how they parse web content. But they’re improving rapidly. The investments you make now in semantic structure, structured data, and content clarity will compound as AI agents get better at understanding and utilizing them.

There will be temptations to game this system—just as there were with SEO. Content farms will try to generate ai-bait content. Sites will manipulate structured data. New forms of spam will emerge.

But the fundamental principle is the same as with SEO: the most sustainable strategy is genuine quality. Make content that’s accurate, well-structured, and genuinely useful. The AI agents that thrive will be the ones that surface good content. Aligning with that goal—rather than trying to game it—is the long-term play.

Design is still for humans

A final note: none of this means design stops being for humans. Humans are still the end users. They’re just increasingly reached through AI intermediaries.

The goal isn’t to make your website pleasant for a language model. The goal is to ensure that when an AI represents your content to a human, that representation is accurate and favorable. You’re designing for humans, through AI, rather than directly.

This adds a layer of complexity to web design—but it’s a layer we can’t ignore. The question isn’t whether to design for AI readers, but how to do it without sacrificing the human experience. The answer is that good design for AI and good design for humans overlap more than they conflict, and the foundation of both is clarity, structure, and substance.