Master AI Engine Optimization (AEO) and its 3 sub-fields: LLMO, GEO, and AAIO.

Author(s): Mohit Sewak, Ph.D.

Originally published on Towards AI.

Alright, grab your cup of chai, pull up a chair, and let’s talk. Because the internet as you know it? It’s over.

And I’m not being dramatic. I mean, I am a storyteller, so a little drama is my jam, but this time, the data is just screaming.

For the last twenty years, we’ve all been playing the same game: Search Engine Optimization, or SEO. It was our digital religion. We prayed at the altar of Google, performed keyword rituals, and built link-towers to the heavens, all hoping to be blessed with that coveted #1 spot.

Well, the gods have changed. And they’re not listening to our old prayers anymore.

Recent intelligence from the front lines is staggering. When Google’s new AI Overviews — those neat, conversational summaries at the top of the page — show up, the top organic search results can lose up to 79% of their clicks (Authoritas, 2025).

Let that sink in. You could be the undisputed king of the search results, the #1 champion, and still have four out of five of your potential visitors simply vanish. Poof. Gone.

This isn’t a future forecast. It’s the rain that’s already soaking you. We’ve fundamentally shifted from a web built for human eyeballs to a web built for machine intelligence. The main visitors clicking around your digital real estate are no longer just people. They are swarms of sophisticated, task-oriented AI agents.

To survive this, you need to stop thinking about Search Engine Optimization and start mastering a whole new discipline: AI Engine Optimization (AEO). It’s not just a new acronym to sound smart in meetings. It’s a completely new martial art for the digital world. And mastering its three core stances — Large Language Model Optimization (LLMO), Generative Engine Optimization (GEO), and Agentic AI Optimization (AAIO) — is no longer a competitive edge. It’s the basic price of admission to the future.

The Day the Ten Blue Links Died

Picture the internet of yesteryear as a bustling, chaotic city. SEO was the art of being the best tour guide. You’d stand on the busiest street corner (Google’s page one) with the biggest, flashiest sign, screaming, “Get your authentic, artisanally crafted information right here!” And people would follow you down your little alleyway to your website.

The goal was simple: get them to click one of those “ten blue links.”

Today, that city has a new central concierge. A giant, omniscient AI sitting in the main plaza.

The new city center. Your old street corner is now a historical landmark.

Tourists don’t wander the streets looking for signs anymore. They just ask the AI concierge, “Hey, where’s the best place to get a taco?” And the concierge doesn’t hand them a map with ten options. It just synthesizes all the information it knows and says, “The best taco is a combination of the salsa from Taco ‘Bout It, the tortilla from Guac ’n’ Roll, and the carnitas from Holy Frijoles. I have summarized the recipe for you.”

That’s the “zero-click” future. The user gets their answer and their journey ends right there, in the plaza. They never even walk down your street.

The data backs up this ghost town effect. The Pew Research Center found that clicks to traditional links get sliced nearly in half when an AI summary appears (Pew Research Center, 2025). A user experience study saw a staggering two-thirds drop in clicks on desktop, with users barely even reading past the first few lines of the AI’s answer (Indig & van Buskirk, 2025).

Your new customer isn’t the tourist anymore. It’s the concierge. You don’t have to convince people to visit your shop; you have to convince the AI that your shop has the best ingredients in the entire city.

“The future is already here — it’s just not evenly distributed.” — William Gibson

The New Defense: Winning in a World of AI Concierges (GEO & LLMO)

So, how do you get the all-knowing AI concierge to recommend your tacos? You can’t charm it. You can’t bribe it. You have to speak its language. This is our first defensive move in AEO, and it’s called GEO and LLMO.

Generative Engine Optimization (GEO) is your direct response to things like Google’s AI Overviews. The goal is no longer to be the #1 link. The new #1 spot is to be cited as a source inside the AI’s answer. That little footnote is the new digital holy grail.

Large Language Model Optimization (LLMO) is the grunt work that makes GEO possible. It’s about renovating your shop — your website — so that AI models can crawl it, understand it, and most importantly, trust it.

Think of it like this: an AI doesn’t “read” your beautifully designed blog post. It ingests it like a protein shake. You need to make sure all the nutrients are perfectly blended and labeled.

For an AI, unstructured content is noise. Structured data is a perfectly labeled meal.

Here’s how you do it:

• Answer the Dang Question: Stop writing flowery intros (yes, I see the irony). Structure your content in brutally clear question-and-answer formats. The AI is looking for the most direct, unambiguous answer to a user’s query. Give it that.
• Use Your Labels (Structured Data): This is the big one. Using Schema.org markup is like putting a “nutrition label” on your website’s data. You’re explicitly telling the AI, “This string of numbers is the price,” “This text is the recipe,” “This date is the event time.” It allows the AI to grab the facts without having to guess by reading the whole paragraph. It’s the difference between reading a 500-page book and just reading its summary.
• Become the Professor of a Single Subject: Don’t be a jack-of-all-trades. Build deep, comprehensive content hubs that prove you are an undeniable authority on one specific topic. This is what marketers call E-E-A-T (Experience, Expertise, Authoritativeness, Trustworthiness). You want the AI to think of you as the undisputed “Professor of Tacos.”

ProTip: Go to Schema.org right now. Look up the schema type for your product, service, or content (e.g., “Recipe,” “Product,” “LocalBusiness”). Implementing even the most basic schema can make you instantly more legible to AI engines.

Preparing for Your Next Customer: The Autonomous AI Agent (AAIO)

Okay, so we’ve made our shopfront legible to the AI concierge. That’s defense. That’s survival. Now, let’s talk offense. Let’s talk about thriving.

While GEO and LLMO are about reacting to the AI on the search page, Agentic AI Optimization (AAIO) is about preparing for a completely new kind of visitor.

Forget the human tourist. Forget the AI concierge. I want you to picture a new entity in our futuristic city: a swarm of autonomous delivery drones.

These aren’t browsing. They aren’t “surfing the web.” They are AI agents sent by users with a specific, complex job to do. A user won’t Google “best CRM for small business” anymore. They’ll tell their personal AI assistant:

ProTip: Research the top three CRM platforms under $50/month that integrate with QuickBooks. Summarize their core features, find the best pricing tier for a team of five, and schedule demos with their sales teams for next Tuesday.

That AI agent is now your visitor. It’s a programmatic customer. It doesn’t care about your clever branding, your beautiful hero image, or your witty blog post. It has a checklist, and it needs to execute a task.

If your website is just a pretty storefront, that drone is going to fly right past. To win in the agentic web, your website needs to be a warehouse with a damn good loading dock.

Your new customers don’t use the front door. Build them a loading dock.

This is what an agent-ready website looks like:

1. Machine-Readable Everything: Structured data was the start. Now, make everything programmatically accessible. Think price lists, feature tables, availability — all in formats that a script can parse instantly.
2. Build a Loading Dock (APIs): An API (Application Programming Interface) is a dedicated service window for AIs. Instead of the agent drone having to fly into your store, awkwardly navigate the aisles, and read product labels (i.e., parse your visual HTML), it can go directly to the API loading dock and place its order. “Give me pricing for five users.” “Book a demo for Tuesday at 2 PM.” This is the language of agents.
3. Digital Bouncers (Permissions): As these agents start acting on behalf of users — accessing their data, spending their money — you need robust, secure protocols for authentication and authorization. You need to be able to know, with certainty, that this agent is who it says it is and that it has permission to book that demo.

Trivia: The idea of autonomous agents isn’t new! In the 90s, the concept of “Software Agents” or “softbots” was a hot topic in AI research, predicting a future where programs would act as our personal delegates on the internet. It just took 30 years for the hardware and algorithms to catch up.

From Optimizing for AI to Optimizing with AI

This is where it gets really wild. The final evolution isn’t just about preparing your website for other people’s AIs. It’s about deploying your own fleet of autonomous agents to run your business better than any human team ever could.

This isn’t sci-fi. The academic research is already proving this out in the real world.

The ultimate goal: an autonomous AI fleet running your business operations with superhuman efficiency.

• The AI Web Designer: Researchers built a reinforcement learning (RL) agent that could redesign a webpage in real-time for every single visitor. It learned to map a user’s data to the perfect layout, achieving a 15% boost in conversions over traditional methods (Chen et al., 2020). Imagine a shop that redecorates itself instantly for every person who walks in the door.
• The AI Media Buyer: Another team built an RL agent to manage a massive digital ad budget. It learned a sophisticated bidding strategy that blew human-designed rules out of the water, intelligently pacing its spending and winning the most valuable ad slots without overpaying (Wu et al., 2018).
• The AI Ecosystem Steward: Going even bigger, researchers have designed multi-agent systems to manage entire content platforms (like YouTube or Netflix). The AI’s goal wasn’t just to get you to click the next video, but to optimize for long-term health, content diversity, and fairness to new creators (Wang et al., 2020). It learned to make small sacrifices in immediate engagement to build a healthier, more resilient community over time.

Simple Explanation: Reinforcement Learning (RL) How does this work? Think of training a puppy. The agent (the puppy) tries an action (like personalizing a headline). If that action leads to a reward (the user clicks the ‘buy’ button!), it gets a treat and learns to do that action more often. Now, imagine a puppy that can do this millions of times per second. It quickly becomes a super-genius at its one specific task.

The Double-Edged Sword: Power, Peril, and Responsibility

Okay, let’s pause. I see the look in your eyes. This is exciting. It’s powerful. It’s also terrifying.

As someone who’s spent years in both AI and cybersecurity, and who gets punched in the face for fun (it’s called kickboxing, and it teaches you a lot about control), I can tell you this: with great power comes the absolute necessity for even greater control. The same agentic systems that can optimize your sales funnel can also, if you’re not careful, become brand-destroying nightmares.

The same technology that offers unprecedented personalization can create unaccountable discrimination.

• Digital Redlining: An agent optimizing for conversions might learn that people in a certain zip code are less likely to buy. So, it simply stops showing them your best offers. Without any malicious intent, it has just invented a new form of algorithmic discrimination (Datta et al., 2015).
• Manipulation vs. Personalization: Where’s the line? An agent could learn that a user is more likely to make an impulse purchase when they’re feeling sad, and start showing them specific ads when their social media sentiment dips. Is that smart marketing, or is it predatory?
• The Accountability Black Box: When your AI agent makes a catastrophically bad decision — say, it bids your entire marketing budget on a single ad — who’s at fault? The deep learning models that power these agents are often so complex that even their creators can’t fully explain their reasoning. This creates a terrifying crisis of accountability.

Ignoring these issues isn’t just bad ethics; it’s a suicidal business strategy. Responsible AI isn’t a charity project. It’s a core component of risk management and long-term brand survival.

“We are the first generation to be armed with the knowledge of how our discoveries will be used. We have a responsibility to use that knowledge.” — J. Robert Oppenheimer

Your AEO Roadmap: From Survival to Dominance

So, the world has changed. The old maps are useless. What do you do right now, sitting here with your now-cold cup of tea?

It’s not about panic. It’s about a pivot.

For C-Suite Leaders: This is not a “marketing thing.” This is a fundamental shift in your business infrastructure. You need to start investing in your data architecture, your technical talent, and a culture that is ready for an agent-first world.

For Marketing & Digital Leaders: Here is your battle plan.

Your AEO battle plan: Defend your position, proactively build for agents, then go on the offensive.

1. Act Now (The Next 3–6 Months): The Defensive Stance.
   — Audit your entire content library for GEO/LLMO readiness.
   — Implement structured data (Schema.org) across your entire site. Yesterday.
   — Start creating direct, answer-focused content that makes you the most citable source in your niche.
2. Plan Ahead (The Next 12–18 Months): The Proactive Strategy.
   — Develop a roadmap for AAIO. What are the key tasks a user (or their agent) needs to accomplish on your site?
   — Start planning the API and data infrastructure you’ll need to allow agents to perform those tasks programmatically.
3. Start Experimenting (Today): The Offensive Drill.
   — You don’t need to build a god-like AI tomorrow. But you can start exploring small-scale agentic tools. Use AI for ad bidding, for simple personalization on your landing pages, or for internal process automation. Build your team’s muscle memory.

The Post-Credits Scene

The ground beneath our feet has shifted. Trying to win the next decade of the internet with the old SEO playbook is like showing up to a Formula 1 race with a horse and buggy. You might be the best horseman in the world, but the nature of the contest has changed.

AI Engine Optimization (AEO) is the new vehicle. It requires a defensive strategy to get cited by the new AI gatekeepers (GEO/LLMO), a proactive strategy to build a business that speaks the language of autonomous agents (AAIO), and an offensive strategy of deploying your own agents to out-maneuver the competition.

The businesses that see this shift, that embrace this agent-first reality, and that master these new disciplines will not just survive. They will own the next era of digital commerce and connection.

The agentic web is here. Your move.

References

Foundations of Agentic Systems (Reinforcement Learning)

• Chen, J., Li, H., Zhang, Y., & Wang, J. (2020). Personalized Webpage Optimization via Deep Reinforcement Learning. Proceedings of The Web Conference 2020, 1983–1994. https://doi.org/10.1145/3366423.3380254
• Sutton, R. S., & Barto, A. G. (2018). Reinforcement learning: An introduction. MIT press.
• Wang, X., Zhang, Y., & Agarwal, D. (2020). A Multi-Agent Reinforcement Learning Framework for Content Recommendation. Proceedings of the 13th International Conference on Web Search and Data Mining, 649–657. https://doi.org/10.1145/3336191.3336217
• Wu, T., Ren, Y., Zhang, W., & Yu, Y. (2018). Budget-Constrained Bidding by Model-free Reinforcement Learning in Display Advertising. Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, 2525–2534. https://doi.org/10.1145/3219819.3219832

The SEO-to-AEO Paradigm Shift (GEO & AAIO)

• Authoritas. (2025). The State of AI Overviews — User Intent Research. Authoritas Blog. https://www.authoritas.com/blog/ai-overview-study-how-user-intent-drives-aio-appearance-rates/
• Google. (2025, May). AI in Search: Going beyond information to intelligence. The Keyword. https://blog.google/products/search/ai-in-search-io-2025/
• Indig, K., & van Buskirk, E. (2025). The first-ever UX Study of Google’s AI Overviews: The Data We’ve All Been Waiting For. GrowthMemo. https://www.kevin-indig.com/the-first-ever-ux-study-of-googles-ai-overviews-the-data-weve-all-been-waiting-for/
• Jaffrelot Inizan, T., et al. (2025). System of Agentic AI for the Discovery of Metal-Organic Frameworks. arXiv preprint arXiv:2504.14110. http://arxiv.org/abs/2504.14110
• Pew Research Center. (2025). Google users are less likely to click on links when an AI summary appears in the results. Pew Research Center.
• Razorfish. (n.d.). How Agentic AI Will Reshape Search. Razorfish.
• Surfer SEO. (n.d.). 7 Large Language Model Optimization Strategies. Surfer SEO Blog.

Ethical Considerations & Responsible AI

• Ali, M., Sapiezynski, P., Le, V., T., Nguyen, A., & Mislove, A. (2019). Investigating Ad Transparency Mechanisms in Social Media: A Case Study of Facebook’s Ad Library. Proceedings of The Web Conference 2019, 1–12. https://doi.org/10.1145/3308558.3313631
• Bashir, A., Arshad, S., Robertson, W., & Wilson, C. (2019). Automated Experiments on Ad Privacy Settings. 2019 IEEE Symposium on Security and Privacy (SP), 115–132. https://doi.org/10.1109/SP.2019.00037
• Datta, A., Tschantz, M. C., & Datta, A. (2015). Discrimination in Online Advertising: A Multistakeholder Study. Proceedings of the 22nd ACM SIGSAC Conference on Computer and Communications Security, 1008–1021. https://doi.org/10.1145/2810103.2813651
• Yao, M., Chen, Y. C., & Wang, F. D. S. (2019). When Recommendations Treat You Unfairly: The User’s Perspective on Algorithmic Unfairness. Proceedings of the 15th ACM Conference on Recommender Systems, 25–35. https://doi.org/10.1145/3298689.3346988

Disclaimer: The views and opinions expressed in this article are my own and do not represent those of any past, present, or future employer. AI assistance was used in researching, drafting, and generating images for this article. This work is licensed under a Creative Commons Attribution-NoDerivatives 4.0 International License.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI