Why I Finally Stopped Thinking About "SEO" and Started Thinking About "Being Found"

Last month, I was debugging why a client's perfectly optimized blog post wasn't showing up in ChatGPT's search results, even though it ranked #1 on Google. I spent an evening digging through their robots.txt, checking canonicals, validating schema markup, all the classic SEO checklist items. None of it mattered. The post was ranking. It was indexed. But it was invisible to AI search. That's when I realized I've been thinking about this entire problem wrong.

The article I read this week finally explained why. It's not about ranking a URL anymore. It's about whether your content survives four separate filtering stages that have nothing to do with traditional SEO metrics. And honestly? Most of us building for the web right now don't even know these stages exist.

The Four Places Your Content Gets Filtered Out

Here's what I didn't understand until recently: when someone asks a question in ChatGPT or Perplexity, the system doesn't just run one search. It breaks the question apart into multiple sub-queries, what the article calls "fan-out"-and searches for each piece separately.

Think about it practically. Someone asks: "What's the best laptop for machine learning on a budget?" That's actually three different questions hiding in one. The system might search for "budget laptops 2025," "machine learning GPU requirements," and "laptop cooling for compute-heavy tasks" as separate queries. Your article about budget laptops might answer one of those perfectly, but if you only focused on ranking for the original phrase, you'd never know.

This changed how I think about content structure. A #1 ranking means nothing if your page doesn't cleanly answer any of the hidden sub-queries the system actually sends. That's stage one where you lose.

Then comes retrieval, the system pulls pages from its index. If you're not indexed, you're dead. I've seen this with sites using too-aggressive JavaScript rendering or robots.txt blocks. The crawler never saw the content, so it doesn't exist.

Even if you're retrieved, reranking kills a lot of content. The system chunks your entire article into fragments and scores each one. Your 3,000-word guide competes as 20+ separate candidates. One paragraph might be the answer, but if it's buried after three paragraphs of preamble, the reranking model might pick someone else's more direct response.

Finally, even if your fragment gets selected for the answer, the model has to cite it. I've seen cases where content clearly influenced the answer but wasn't attributed. Game over for visibility.

Why This Matters More Than You Think

The uncomfortable truth: ranking #1 on Google now guarantees nothing. You can win the old game and lose the new one simultaneously.

I've started auditing our clients' content differently. Instead of asking "does this rank?", I ask: "does this answer a specific, atomic question clearly and completely?" Does a single paragraph, read in isolation, make sense? Can a reranking model immediately identify it as the answer without reading the surrounding context?

This means our writing needs to change. Introductions with context are beautiful for readers. They're death for AI retrieval. The clearest, most specific statements need to come first. Think like you're writing for a model that will cut your paragraph out of context and compare it to thousands of alternatives.

The other shift: we need to stop obsessing over single keywords. Query fan-out means your content's value comes from covering related subtopics thoroughly and distinctly. A guide on budget laptops should have clearly separated sections on GPU requirements, thermal performance, and specific models, not because it looks good, but because each section might match a different sub-query.

The Honest Bit

I'm still figuring this out. The article mentions that these fan-out queries are hidden, Search Console won't tell me what sub-queries Google actually used. I'm working partially blind, making educated guesses based on the search intent topology.

What concerns me: we're optimizing for systems we can't fully inspect. The SEO community's built on transparency and measurable signals. RAG-based search removes that. We're back to guessing, but at a more sophisticated level.

What excites me: once you understand these four stages, you can actually build content that wins at scale. It's not luck. It's architecture.

What Would You Do Differently?

If you're running a content business or building products that depend on being discoverable through AI search, I'm genuinely curious: have you noticed these patterns? Are you already restructuring content for this reality, or are you still optimizing for the old playbook?

Source: This post was inspired by "How AI search picks fragments: query fan-out and RAG explained" by Dev.to. Read the original article