The Modular Content Debate Misses the Actual Problem

On January 9, 2026, the industry experienced a brief, frantic collective shiver. Danny Sullivan, speaking from within the Google apparatus, told us to stop "chunking" our content for machines. It was a directive that felt like a betrayal to the thousands of publishers who had spent the last eighteen months atomizing their expertise into disconnected, flavorless snippets in a desperate bid for relevance. But the betrayal wasn't Google's. It was the industry's own obsession with what Ars Technica would later identify as "SEO superstition." We mistook fragmentation for structure. We mistook "bite-sized" for "clear."

The center was not holding.

Barry Schwartz reported Sullivan's position for Search Engine Land: Google "doesn't want people to have to be crafting anything for Search specifically." The statement landed with the force of a rule change announced mid-game, which is exactly what it was. Publishers had been told (by whom? by consultants, by case studies, by the breathless optimism of early GEO practitioners) that LLMs preferred knowledge in small, digestible units. Build "knowledge blocks." Write "answer-first." Segment everything. The promise was visibility. The result was incoherence.

Sullivan was describing the wreckage, not the architecture.

The Evidence: Passage-Level Coherence and the 35% Inclusion Rate

Search Engine Land reported research showing that pages with concise summaries at the top demonstrate a 35% higher inclusion rate in AI citations. This is the kind of data point that makes the industry salivate (because we are always looking for the edge, the hack, the thing that separates first-page from oblivion). But the interpretation matters more than the number.

The 35% advantage is a concession to a human in a hurry; the machine is simply the courier that arrived first. Direct answers near the top of comprehensive content make extraction easier across all retrieval systems, traditional search included. This has been true since Google introduced passage-level indexing in 2021. The technical implementation hasn't changed. What changed is that more systems now work this way, and publishers finally noticed because the traffic patterns shifted in ways they couldn't ignore.

The mechanics of retrieval demand semantic coherence at the passage level. LLMs don't read linearly. They evaluate relevance within passages and need each section to be independently coherent while still contributing to the larger argument. Your content needs to make sense in chunks without being written as chunks.

This is the paradox at the heart of Generative Engine Optimization. The system penalizes fragmentation while rewarding passage-level clarity. If your passages aren't coherent, Google's AI Overview won't just ignore you. It will misrepresent you. Or worse, it will cite someone else who said what you meant to say, but clearer. The retrieval layer doesn't care about your intent. It cares about what it can extract without ambiguity.

The Technical Layer: Structure Is Not Manipulation

The publishers who got burned weren't using information architecture. They were atomizing articles because someone told them it would work.

Look at what they built: Recipe blogs that required scrolling through 847 words about a grandmother's porch in rural Tennessee, the quality of afternoon light through lace curtains, the specific sound of cicadas on a July evening, before revealing that you need three cups of flour and two eggs. News sites that broke investigative features into seventeen-slide galleries with each slide requiring a new page load and three banner ads. Business blogs that turned comprehensive guides into seven disconnected posts with internal links binding them together like duct tape on a cracked foundation. SaaS companies that created separate landing pages for "enterprise project management software," "enterprise PM tools," "business project management systems," and "corporate PM platforms" because someone in a conference room decided that keyword variations were the same as topical authority.

The center was not holding, and everyone just kept building.

Ars Technica called this "SEO superstition" for a reason. In a tumultuous digital environment with inconsistent traffic, publishers attributed any positive changes to their latest optimization tactics. The correlation was spurious. We saw patterns that weren't there. We built entire content strategies around algorithmic quirks that were never ranking signals to begin with.

The broken foundations were everywhere. You could see them if you looked: pages with six H1 tags because someone heard that "more keywords in headers" mattered, schema markup claiming a blog post was simultaneously an Article, a BlogPosting, a NewsArticle, and a TechArticle because why not cover all the bases, meta descriptions that were just keyword salad hoping the algorithm would notice, internal linking structures that connected everything to everything because "relevance" and "topic clusters" sounded scientific.

We used to call it manipulation. Now, in the wreckage, we simply call it what it was: desperation masquerading as strategy.

What Survives: Infrastructure for Truth

Semantic HTML is the correct way to structure information on the web. Schema markup provides machine-readable context about what your content actually means. Clear, logically organized sections with proper heading hierarchy make content navigable for humans and extractable for systems.

This is the foundation of Answer Engine Optimization. You become citeable by being the most extractable source of truth in the room. Clear attribution. Direct answers followed by depth. Structured signals about expertise that don't require the system to guess what you know. AEO is infrastructure for truth in an environment where the machine needs to know, definitively, who said what and why they should be believed.

The distinction matters because proper information architecture benefits both human readers and machine extraction without degrading either experience. You can build something that serves both without fragmenting your expertise into disconnected pieces.

The Three-Layer Problem and the Only Remaining Exits

The center is not holding, but some structures remain standing. Discoverability in 2026 requires addressing three overlapping challenges. Pretending you can focus on just one is the kind of strategic self-deception that leads to invisibility.

Building topical authority through backlinks and comprehensive content coverage remains fundamental. This hasn't changed since 2004. It won't change. What changed is that it's no longer sufficient on its own, which is a harder truth than most businesses want to accept. You can't buy your way to authority anymore (the link farms died, then came back as "guest posting networks," then died again). You can't fake it with domain age (plenty of 15-year-old domains rank for nothing). You can't manufacture it with content volume (publishing 847 thin pages doesn't equal expertise, it equals noise).

Making content citeable requires structured signals about expertise, clear attribution, and direct answers that make extraction straightforward. Schema markup and semantic HTML matter because they make the extraction process unambiguous across multiple retrieval systems with different architectures attacking the same problem from different angles. The systems don't collaborate. They independently conclude you're worth citing, or they don't.

Ensuring semantic coherence at the passage level means your content makes sense in sections without requiring the full narrative context. Each passage needs to be independently evaluable while contributing to the larger expertise you're establishing. The LLM reads paragraph seven before it reads paragraph three. It evaluates section four without the context of section one. Your content either survives that evaluation or it doesn't.

These aren't separate strategies. They're overlapping requirements for the same outcome: being found by people looking for what you know.

The choice isn't between "writing for AI" and "writing for humans." That's a false binary designed to make you feel better about inaction. The actual choice is between being discoverable or being a ghost in the machine. Between establishing enough coherent expertise that multiple retrieval systems independently conclude you're worth citing, or hoping that things go back to the way they were when ten blue links and meta descriptions were enough.

They won't go back.

We moved from ten blue links to a world where the answer is the entire journey, where the snippet is the destination, where being cited is the new first page. The only way forward is a fundamental shift in Modern SEO Services—building content that actually stands up to scrutiny across all three layers: authority, extractability, and coherence.

Sullivan's right that you shouldn't write for AI. Structure for retrieval is recognizing that multiple systems with different architectures all need to extract meaning from your content, and proper information architecture makes that possible without fragmenting the narrative or atomizing your expertise into disconnected pieces.

The systems that win aren't the ones gaming algorithmic quirks. They're the ones that understand information architecture has always mattered, and it matters more now that extraction happens at scale.