Three things to do this quarter to improve AEO

Your content is being used without you

I've been in this space long enough to remember when "SEO" meant stuffing keywords into meta tags. We've come a long way since then, but the fundamental assumption has stayed the same: create good content, get found, get clicks, get conversions.

That assumption broke in 2025.

The Similarweb/SparkToro Zero-Click Study found that 58.5% of U.S. Google searches and 59.7% of EU searches now end without a single click to an external website. BrightEdge measured AI Overviews appearing on 48% of all searches as of February 2026, up 58% year over year. Google is answering questions with your content but not sending people to your content.

The numbers on the other side tell a sharper story:

The gap between being cited and being ignored has never been wider. And it's growing every quarter.

I keep having the same conversation

Three years of helping organizations migrate from legacy CMS platforms to modern content architectures, and the pattern repeats almost word for word. The marketing team wants to talk about the redesign. IT wants to talk about the migration timeline. Nobody wants to talk about whether the content itself is fit for purpose.

Then I ask a simple question: "Pick a product page from your website. Tell me, is that product stored as a structured object with typed properties and explicit relationships to other content? Or is it a page with a title field and a rich text body?"

The room goes quiet. Because the answer is almost always the second one.

And that answer explains why their content isn't showing up in AI Overviews, isn't being cited by ChatGPT, isn't being referenced by Perplexity or Claude. AI agents don't read websites the way humans do. They don't scroll. They don't interpret. They look for structured data: content types with defined properties, relationships between entities, metadata like author, date, category, and tags. Schema.org markup in JSON-LD format. Machine-readable content with clear semantics.

If your content is free-form text inside a WYSIWYG editor, AI agents skip it. They move to the next source. And that next source is the competitor who modeled their content properly.

Semantic debt

I've started calling this semantic debt in conversations with clients, and the term sticks because it describes something everyone recognizes but nobody has named.

Semantic debt is the accumulated cost of content that lacks structure, metadata, relationships, and machine-readable markup. Every product page without Schema.org markup. Every FAQ stored as one long text block instead of structured question-and-answer pairs. Every team member mentioned by name in body text instead of linked as a structured Person entity with title, email, photo, and department.

This kind of debt was annoying when Google ranked pages based on links and keywords. In a world where AI agents decide which sources to cite in real time, it's the single biggest factor determining whether you're visible or invisible.

And unlike technical debt, which engineers can see and estimate, semantic debt tends to be invisible to the people making platform decisions. It lives in the gap between what your CMS stores and what AI systems need.

What structured content actually looks like

I want to make this tangible because the concept of "structured content" can sound abstract until you see it applied.

Take a typical B2B company. Products, specialists, customer case studies, a resource center. In a properly modeled content architecture, a Product isn't a "page about a product." It's a content type with typed properties: name, category, description, key benefits, related products, and a reference to the Product Manager who owns it. That Product Manager isn't a name in a text field. It's a reference to a Person content type with name, title, email, photo, and department. Defined once, reused everywhere.

A Case Study has structured references to the customer (an Organization type), the products used, the industry vertical, and measurable outcomes. Everything connected through typed relationships. Change the product name in one place, it updates across every page that references it.

All of this gets exposed through a GraphQL API and rendered with JSON-LD Schema.org markup. When an AI agent reads that product page, it doesn't encounter a wall of text. It sees a structured product entity with typed relationships to people, organizations, and outcomes. That's content worth citing.

This is the kind of architecture Optimizely CMS is built for. Headless-first. Content types defined in TypeScript with full type safety. Everything indexed automatically into Optimizely Graph, a global, edge-cached GraphQL service. The content model you define is the API contract your frontends and AI agents consume. No translation layer. No middleware. The structure is the delivery mechanism.

The operating model shift

Structured content gives you the foundation. But maintaining, optimizing, and scaling that content when you now serve two audiences (humans who want speed and clarity, machines that extract and cite) requires something more. It requires AI agents working inside your CMS, alongside your team, every single day.

This is where I get genuinely excited about what we're building at Optimizely, because Opal isn't a chatbot bolted onto a CMS. It's an agent orchestration platform that changes how digital teams operate.

Generative Engine Optimization (GEO) runs continuously, not quarterly. Opal's GEO Recommendations agent audits your pages for discoverability across ChatGPT, Google AI Overviews, Perplexity, and Claude. It checks whether your content is being found, understood, and cited by these systems and delivers specific, prioritized actions to close gaps. Every day. Not in a report you read three months later.

Content operations move at a pace your team couldn't sustain manually. The platform auto-generates llms.txt files that signal to AI crawlers which pages to index. It creates Q&A pairs from existing content. It populates GEO-specific metadata including EEAT signals (Experience, Expertise, Authoritativeness, Trustworthiness) in bulk across your entire site. Work that takes a content team weeks gets done in minutes.

Your people become orchestrators. Opal has a library of pre-built AI agents and a drag-and-drop workflow builder. Instead of writing prompts one at a time, your team designs workflows where agents handle research, drafting, optimization, and distribution in sequence. Because Opal is connected to your brand kit, your tone of voice guidelines, and your existing campaigns, the output stays on-brand. Your team keeps control of quality. The agents handle the volume. Optimizely's benchmark data shows a 78.7% increase in experiment volume and 53.7% decrease in campaign completion time among organizations using these workflows. That's a fundamentally different way of working.

Experimentation becomes the default, not a project. When AI agents can set up, execute, and analyze A/B tests autonomously, you stop running three experiments a quarter and start running thirty. Personalization stops being something you plan for "next year" and becomes the standard way content reaches your audience.

I've seen what this looks like in practice. Smaller teams producing more, testing more, learning faster. Not because anyone is working harder, but because the operational layer that used to consume 80% of their time is handled by agents. The humans focus on strategy, creativity, and the conversations that actually move deals forward.

Three things to do this quarter