From Search Optimization to AI Cognition Management

When Google became the dominant way people found information, an entire industry emerged around one question: how do I rank higher?

Companies like SEMrush and Ahrefs built billion-dollar businesses answering that question. They reverse-engineered Google's algorithms, tracked rankings, analyzed backlinks, and turned the black box of search into something marketers could actually work with.

Now LLMs are becoming the next information gateway. And the same pattern is starting to repeat—but with a crucial difference that most people haven't fully grasped yet.

The SEO Ecosystem Was Built on Reverse Engineering

To understand what's changing, you first have to understand what SEO tools actually do.

Google has the truth. It knows how its algorithm works, what signals it weighs, how its Knowledge Graph connects entities. But brands don't get to see any of that. They just see the results page.

So SEMrush and Ahrefs built their entire value proposition around helping you guess what the truth is. They crawl the web at massive scale, track which pages rank for which keywords, analyze link patterns, and package all of this into dashboards that tell you: here's what's probably working, here's what you should probably do.

This model worked because Google, despite being a black box, was a consistent black box. Rankings were stable enough to track. Changes were gradual enough to reverse-engineer. The search results page was structured enough that "ranking #3 vs #7" was a meaningful metric.

For 15+ years, this ecosystem matured into a well-defined discipline with established tools, methodologies, and career paths.

Why GEO Is Not "SEO for the AI Era"

On the surface, Generative Engine Optimization (GEO) sounds like the same game. You want your brand to show up when people ask AI assistants for recommendations—just like you wanted to show up on Google.

But the mechanics are fundamentally different.

Google returns links. LLMs give answers.

When Google shows your site on the results page, you still have a chance to make your case. The user clicks through, sees your content, decides for themselves. Your job was to get the click.

But when someone asks ChatGPT "what's the best CRM for small businesses?" and your product isn't mentioned—or worse, is mentioned unfavorably—there's no click to win. The AI has already made the judgment for the user. You're either part of the answer or you're invisible.

Google rankings are trackable. LLM responses are probabilistic.

You can check your Google ranking for a keyword every day and get a consistent result. But ask Claude the same question twice and you might get different recommendations, different orderings, different framings. The "ranking" doesn't exist in a stable, measurable way.

SEO optimizes landing pages. GEO optimizes perception.

Traditional SEO is about making your pages more relevant, more authoritative, more technically sound. But LLMs form impressions of your brand through two layers: training data (what the model "knows" out of the box) and real-time retrieval (search results, RAG, cited sources).

The first layer is almost impossible to trace—you don't know which articles, reviews, Reddit threads, or Wikipedia entries shaped the model's baseline understanding. The second layer is more transparent, but now you're dealing with multiple models that each select and synthesize sources differently. Perplexity, ChatGPT, Claude, Gemini—they don't all trust the same sources or weigh them the same way.

The Reverse Engineering Problem Gets Much Harder

Here's where it gets really uncomfortable for anyone trying to build "SEMrush for AI."

With Google, you could at least observe the outputs systematically. Run a thousand queries, track the rankings, spot patterns, make educated guesses about what the algorithm values.

With LLMs, this reverse engineering approach runs into serious obstacles:

There's no stable "ranking" to track. Every response is generated fresh. The same query can yield different results based on context, conversation history, or just probabilistic variation. What do you even measure?

Training data is a black box within a black box. You don't know what articles, forum posts, or documents shaped the AI's understanding of your brand. You can't get a list of your "backlinks" equivalent. The AI's perception was formed during training, in ways that are fundamentally opaque.

There's no single algorithm to reverse-engineer. ChatGPT, Claude, Perplexity, Gemini—each has different training data, different architectures, different behaviors. Optimizing for one might not help with the others. The market is fragmented in a way that Google search never was.

This is why I think the traditional SEO tool playbook—track rankings, analyze competitors, surface optimization opportunities—will struggle to translate directly to the AI era.

The Real Game: AI Cognition Management

If you can't reliably track rankings, and you can't reverse-engineer training, what can you do?

I think the frame needs to shift from "optimization" to "cognition management."

The question isn't "how do I rank higher?" It's "what does AI think about my brand, and how do I shape that?"

This means monitoring a different set of signals:

• How does AI describe your brand? Does it understand your positioning correctly, or does it have misconceptions?
• Who does AI compare you to? Are you being grouped with the competitors you want to be grouped with, or with the wrong peer set?
• What does AI get wrong about you? Every knowledge gap is an opportunity to close.
• In what contexts does AI recommend you? Are you showing up for the use cases you care about?

These are perception questions, not ranking questions. And managing them requires different tools and different thinking.

Who Becomes the New SEMrush?

This is something I've been actively thinking about—and building toward.

My hypothesis is that the winning company in this space won't be a "monitoring tool" in the traditional sense. It will be something more like an AI cognition management platform.

The core assets won't be data alone. They'll be:

Methodology for measuring AI perception. How do you systematically test what multiple LLMs think about a brand? What prompts do you use? How do you account for the probabilistic nature of responses? This is still an unsolved problem.

Playbooks for influencing AI perception. Once you know what the AI gets wrong, how do you fix it? The answer probably involves structured data, authoritative content creation, and strategic information placement—but the specific tactics are still emerging.

Becoming a trusted information source for LLMs. Here's the most speculative—and potentially most valuable—part.

Google spent 15 years building its Knowledge Graph—a structured representation of entities and relationships that powers search. What if the equivalent play in the AI era is building a richer, more authoritative knowledge graph that multiple LLMs learn to trust and reference?

The key insight is that different models have different retrieval behaviors, but they all need reliable sources. If you can become the canonical source of truth for a category—structured, comprehensive, constantly updated—you're not just monitoring AI perception. You're shaping it at the infrastructure level.

This means the winning strategy might not be optimizing for each model separately, but creating information assets that all models converge on because they're simply the best source available.

The Market Question

Will this market mature the way SEO did? Will there be a clear "GEO tools" category with established players in five years?

I'm genuinely uncertain.

The SEO market emerged because Google was so dominant and so stable. There was one algorithm to optimize for, and it changed slowly enough that an industry could form around understanding it.

The AI landscape is much more fragmented and much more volatile. Models are improving rapidly. Multiple players have significant market share. The way people use AI assistants is still evolving.

It's possible that GEO tools become just as established as SEO tools. It's also possible that the market evolves in a completely different direction—maybe toward AI-native content creation, or toward direct partnerships with model providers, or toward something we haven't imagined yet.

What's Actually Changing

Here's what I keep coming back to:

GEO isn't replacing SEO. The two aren't in conflict—in fact, they share the same foundation.

Google's core technology includes Knowledge Graph. LLMs doing RAG also rely on structured knowledge and entity relationships. Recent data from Backlinko shows that 76% of AI Overview citations come from pages that already rank in Google's top 10. Good SEO is the foundation for GEO, not the opposite.

So SEO remains SEO. What GEO adds is making the connections between entities more explicit and organic—helping AI understand not just that your brand exists, but how it relates to categories, use cases, and other players in the space.

This brings us to an open question I don't have a complete answer to: why do certain products show up when someone asks "what's the best CRM for small businesses"?

My hypothesis: it's because they're key players, and being a key player creates a reinforcing cycle. Users talk about them, media covers them, Reddit threads recommend them, Google results feature them—and all of this feeds into both training data and real-time retrieval. Each signal reinforces the others. The more you're mentioned in the right contexts, the more AI learns to associate you with those contexts, the more you get mentioned.

This creates a compounding advantage that's hard for newcomers to break into through brute force.

But here's where I see opportunity for new players:

Find a more precise niche. If you can't win "best CRM," maybe you can own "best CRM for real estate agents" or "best CRM with WhatsApp integration." Perplexity, for instance, surfaces an average of 13 brands per query compared to ChatGPT's 3-4—there's more room for niche players in some AI contexts than others.

Occupy new ecological positions. Categories evolve. New use cases emerge. When AI starts recognizing "Revenue Operations Platform" as a distinct category, the brands that moved early to define that space will have an advantage.

Build authority where established players aren't looking. Reddit threads, YouTube reviews, niche industry publications—these feed into AI perception. If the incumbents are focused on traditional PR and you're building authentic presence in the communities AI actually draws from, you can shift the balance.

The meta-point is this: GEO isn't about gaming a new algorithm. It's about understanding that AI's perception of your brand is shaped by a web of signals—and those signals can be influenced, but not easily faked.

Search engines changed how brands get discovered. LLMs are changing how brands get understood.

The first was a traffic war. The second is a cognition war. But unlike traffic, cognition compounds. The brands that establish the right associations early will be hard to displace.

We're still in the early innings of figuring out exactly how this works. That uncertainty is uncomfortable—but it's also where the opportunity lives.