📌 The full version (with 4 SVG figures and a related-article network) lives on my blog:

👉 https://okikusan-public.dev/seo-aeo-geo-context-ops.en

This dev.to post is the condensed version. The visualisations live on the canonical page.

Introduction

Marketers are sitting in the middle of the quietest, most destructive tectonic shift in twenty years.

The "type a keyword, click one of the ten blue links" routine is collapsing fast. Google's AI Overviews, Perplexity, ChatGPT Search — these AI search engines crawl the web on the fly and present "a single answer" in response to whatever complex question the user typed in.

The short version: "You don't need to throw SEO out. But on its own, it no longer works." From here on, AEO and GEO have to be stacked on top of SEO.

Five prerequisite terms

This article assumes the following background. Skim here if any are unfamiliar.

• SEO (Search Engine Optimization) — getting found in the first place. Crawlability, semantic markup, speed.
• AEO (Answer Engine Optimization) — getting picked as the direct answer by summarising AIs. TL;DR / answer-first / FAQ / comparison tables.
• GEO (Generative Engine Optimization) — getting cited as the source for generative AI answers. First-hand experience, proof data, author attribution.
• AI Overviews / Perplexity / ChatGPT Search — examples of AI search engines that crawl the web and summarise it into a single answer. They always attribute citations.
• Context Ops — a concept growing out of LLMOps. Output quality is decided more by "what context you hand the model" than by "which model."

TL;DR

• With AI search (Google AIO / Perplexity / ChatGPT Search) on the rise, the "ten blue links" SEO era is over.
• The new paradigm is a three-layer stack: SEO (entrance) + AEO (the answer) + GEO (the citation). SEO is the foundation, AEO is the pick, GEO is the real battlefield.
• AI crawlers measure cosine similarity in vector space. Commodity round-ups score zero information gain and get filtered out. Only first-hand experience and proof data survive as "unique nodes."
• The essence of web marketing shifts to Context Ops — handing the right context to AI agents. Humans produce the tacit; AI cleans it into the explicit.

Chapter 1 — The Three-Layer Stack: SEO / AEO / GEO

The stack looks like this:

┌──────────────────────────────────────────────────────────┐
│  GEO (Generative Engine Optimization)                    │
│    cited as the source by generative AI answers          │
├──────────────────────────────────────────────────────────┤
│  AEO (Answer Engine Optimization)                        │
│    picked as the direct answer by summarising AIs        │
├──────────────────────────────────────────────────────────┤
│  SEO (Search Engine Optimization)                        │
│    the entrance — get found at all                       │
└──────────────────────────────────────────────────────────┘

SEO is not dead. Even an AI search crawler has to discover and index your page before anything else can happen. Technically clean, semantic markup, fast loading — these remain the foundation layer on which AEO and GEO are built.

Chapter 2 — Vector search and the cosine trap

Why are the round-up sites and curation articles that used to dominate SEO getting wiped out of AI search citations?

Because of a cold algorithmic fact: AI search engines run on vector search and cosine similarity.

[ AI crawler's decision flow ]
   Web content
      ↓
   project into a multi-dimensional vector space
      ↓
   measure cosine similarity vs pre-trained data & other pages
      ↓
   ┌────────────────────────┐    ┌────────────────────────┐
   │ HIGH similarity (~99%) │    │ LOW similarity         │
   │ "already known"        │    │ "unique node"          │
   │ info gain ≈ 0          │    │ worth citing           │
   │ → DROPPED              │    │ → CITED                │
   └────────────────────────┘    └────────────────────────┘

A round-up of stuff that lives on the internet gets scored at "99% similarity (= the AI already knows this perfectly)." No reason to load it as RAG context. Information gain is zero.

By contrast, content like the following sits far from the existing learned data in vector space, recognised as a "unique node":

• Raw before/after logs from actually using your own product
• The failure in a specific project, and the judgement you arrived at because of it
• The tacit knowledge you scraped together on the ground, finally put into words

💡 This lines up perfectly with Google adding Experience to "E-A-T" in its search quality guidelines. In an era when AIs with no experience mass-produce commodity round-ups, content with "lived, first-hand proof data from someone who actually did it" gets prioritised. That's the official signal.

Chapter 3 — Marketing shifts to Context Ops

Getting your information "accurately understood by AI search and autonomous agents, and chosen as the basis for the user's decision" — at its core, that is Context Ops against the AI system.

Old web marketing:
└ hacks for the things humans see (display / traffic)

New web marketing (Context Ops):
└ hand the right context to AI agents and shape their reasoning

The role-confusion trap with reasoning models

Here's the trap many companies and marketers fall into: "have the big reasoning model write the long blog post or the rambling explainer." It's a confusion of roles.

No matter how huge or clever the LLM, the documents it produces are statistically the average of its training data — "polite round-ups (commodity)." Outsourcing the actual creation lands you with both a hallucination risk and immediate removal from the AI search index by the cosine trap. Two collapse risks in one move.

The human-AI standardisation boundary

┌─────────────────────────────┐       ┌─────────────────────────────┐
│ Human (primary author)      │  →    │ AI (assistant)              │
│ Creating tacit knowledge    │       │ Cleaning into explicit form │
│  · raw field experience     │       │  · structuring & whiteboard │
│  · messy struggle           │       │  · semantic organisation    │
│  · unique judgement         │       │  · parser-friendly shaping  │
│  · measured evidence        │       │  · AEO/GEO optimisation     │
└──────────────┬──────────────┘       └──────────────┬──────────────┘
               │                                     │
               └──────────────┬──────────────────────┘
                              ↓
              ┌──────────────────────────────┐
              │ Clean knowledge on the web   │
              │ cited by AI / chosen by you  │
              └──────────────────────────────┘

Ask the reasoning model to do the upstream work — brainstorming, structuring complex information, breaking tacit knowledge apart. For the final step of putting it into words, let the human (or a fast lightweight LLM) inject lived experience and emotional heat.

Chapter 4 — A template for the next-gen page

The rule for content design that wins AEO/GEO is one page, one point — narrow and deep, not wide and shallow.

Stack these blocks top-down and you get a skeleton that catches SEO, AEO, and GEO all at once.

Closing

SEO is the entrance. AEO is the answer. GEO is the citation. And your unique perspective is the biggest citation asset of the AI Mode era.

Web marketing from here on is not a side-show of tactics for hacking search slots. It's a deeply intelligent, creative practice: extract the living tacit knowledge you earned on the ground, structure it into explicit context both AI and humans can read at a glance, and plug it into the global knowledge graph called the internet.

📌 The full version with 4 SVG figures and the related-article network is on the original:

👉 https://okikusan-public.dev/seo-aeo-geo-context-ops.en

Related (the author's adjacent threads)

• The gap between spec and Source of Truth = context: https://okikusan-public.dev/context-is-the-gap.en
• Tacit knowledge × tacit thought = the territory AI descends into: https://okikusan-public.dev/longtail-tacit-agent.en
• Make Obsidian the AI's memory itself: https://okikusan-public.dev/obsidian-as-ai-memory.en

Sources (official references)

• Google AI Overviews announcement: https://blog.google/products/search/generative-ai-search/
• OpenAI ChatGPT Search: https://openai.com/index/introducing-chatgpt-search/
• Perplexity: https://www.perplexity.ai/
• Google E-E-A-T update (adding Experience): https://developers.google.com/search/blog/2022/12/google-raters-guidelines-e-e-a-t

How is the cosine trap showing up in your stack? Would love to hear which "unique node" you're trying to publish next. 🦄 💬