How GEO forces web accessibility

SEO vs GEO: what changes when the engine answers for you

SEO and GEO share a goal — making your content visible — but they don't play the same game.

The fundamental difference: in SEO, Google ranks pages. In GEO, the LLM selects passages — a paragraph, a list, a number — and weaves them into its response. Your content is no longer a result. It becomes an ingredient.

This means a poorly structured 3,000-word page loses to a shorter article where each section clearly answers a question. The AI doesn't scan an entire document hoping to find the right information. It chunks, evaluates, and selects — passage by passage.

The good news: practices that work for GEO don't contradict SEO. They reinforce it. Clear, structured, well-sourced content is also content that Google ranks well. It's not a choice between the two — it's the same work serving both.

What LLMs read

An LLM doesn't see your page the way a user does. It doesn't see your colors, your layout, or your animations. It sees text and structure. Exactly like a screen reader.

Headings

An LLM chunks your page into passages. Headings are its cut points. A clean H1→H2→H3 hierarchy means content the AI can segment, understand, and cite passage by passage.

<!-- ❌ Flat — the LLM sees a wall of text -->
<div class="section-title">Our commitments</div>
<div class="subsection-title">Environment</div>
 
<!-- ✅ Structured — the LLM sees a tree -->
<h2>Our commitments</h2>
<h3>Environment</h3>

A skipped heading level (H1 then H3 with no H2) breaks the hierarchy. The screen reader loses a navigation level. The LLM loses a level of context. The result is the same: your content is harder to use.

Landmarks

Semantic elements — <header>, <nav>, <main>, <article>, <section>, <aside>, <footer> — aren't decoration. They're markers that screen readers use to navigate and LLMs use to separate useful content from noise (menu, footer, sidebar).

<!-- ❌ A div soup — everything looks the same -->
<div class="wrapper">
  <div class="top-bar">...</div>
  <div class="content">...</div>
  <div class="bottom">...</div>
</div>
 
<!-- ✅ Semantic landmarks — structure is explicit -->
<header>...</header>
<main>
  <article>...</article>
</main>
<footer>...</footer>

An LLM hitting a page of nested <div>s has to guess what's content and what's chrome. An LLM hitting an <article> inside a <main> knows exactly what to extract.

Lists and tables

Pages cited by generative engines overwhelmingly contain lists or tables. That's not a coincidence: a list is a structure the AI can reuse as-is in its response. A paragraph of prose containing the same information requires extraction and rephrasing — with a risk of loss or distortion.

<!-- ❌ Prose — the LLM must extract and rephrase -->
<p>We offer an initial audit, followed by a remediation plan,
then six months of support and finally a verification audit.</p>
 
<!-- ✅ List — the LLM can cite directly -->
<ol>
  <li>Initial audit</li>
  <li>Remediation plan</li>
  <li>Six months of support</li>
  <li>Verification audit</li>
</ol>

Structured data

schema.org markup lets you explicitly declare what your content represents: an article, a FAQ, a product, a review, an event. Brands that use structured data make it easier for generative engines — and increase their chances of being cited.

It makes sense: instead of guessing that a block of text is a Q&A, the LLM knows because you declared it. Less ambiguity, more chances of being selected.

Image alt text

LLMs don't see images — they read their alt. A descriptive, natural alt text makes an image that AI can understand, index, and cite in the right context. We cover this in detail in The two lives of an image.

What you can do right now

You don't need to choose between accessibility, SEO, and GEO. The work is the same.

Check your heading structure. Every page has one H1. Levels follow in order with no skips. You can verify this in 10 seconds with the WAVE extension.

Replace your <div>s with semantic HTML. <main>, <article>, <section>, <nav>, <header>, <footer> — each element says what it contains. Screen readers navigate by them. LLMs do too.

Get your alt text right. Informative images need an alt that says what they show — a natural sentence, not stacked keywords. Decorative images need alt="" to tell screen readers and LLMs to move on. Both cases matter. We cover this in detail in The two lives of an image.

Structure your data as lists and tables. If your content is an enumeration, write a <ul>. If it's a comparison, write a <table>. The AI can cite a list. It can't cite a paragraph hiding a list.

Add structured data. schema.org markup to declare what your content represents — article, FAQ, product, review. Less ambiguity for the LLM, more chances of being cited.

Bring evidence. Generative engines favor content that cites sources, statistics, and experts. According to Aggarwal et al., an article with verifiable numbers has up to 40% more visibility in generative answers than one without.

These six actions serve three goals. One effort, three results — exactly what I explained in Three bills, one codebase, zero overhead. Except now, the bills are four.