Search behavior is changing — fast
Gartner forecasts that traditional search engine volume will drop by 25% by 2026 as users shift to generative AI and virtual agents. Other industry trackers put zero-click search behavior — where the user gets the answer without leaving the search engine results page — at roughly 60% of all queries. The blue-link economy is contracting.
What's growing is AI-mediated discovery: ChatGPT, Perplexity, Claude, Gemini, and Google's own AI Overviews each pulling from a small set of sources to synthesize an answer. If your page isn't in that source set, you're invisible no matter how well you rank in classical Google.
AEO, GEO, and how they relate to SEO
These terms get used interchangeably, but there are useful distinctions:
- SEO (Search Engine Optimization) — optimizing for rankings in classical search results pages (Google, Bing).
- AEO (Answer Engine Optimization) — optimizing for direct answer features: featured snippets, People Also Ask, voice answers, AI Overviews.
- GEO (Generative Engine Optimization) — Princeton's term for optimizing for citation by generative LLMs (ChatGPT, Perplexity, Claude). Their 2024 paper showed 41% lift from targeted edits.
All three overlap
A well-structured page that ranks for a query often gets cited as an answer source and pulled into LLM responses. The overlap is large. But the optimization tactics diverge: AEO favors direct, quotable answer paragraphs; GEO rewards statistics, citations, and entity density; classical SEO still cares about link equity and crawlability.
The retrieval + extraction model
When you ask ChatGPT or Perplexity a question, the system runs a retrieval step (which sources are relevant?) and an extraction step (which spans of those sources answer the question?). Both steps reward certain content patterns.
- Direct answer paragraphs — 50 to 100 words near the top of the page, plain prose, no marketing fluff.
- Inline statistics and citations — "Princeton's 2024 GEO paper showed 41% lift" beats "studies show AI search is growing."
- Named entities — people, places, products, organizations with real-world referents. Citation graphs love disambiguated entities.
- FAQ blocks — directly extractable Q&A pairs that match natural user query language.
- Structured data (Schema.org) — JSON-LD that tells the LLM what kind of thing the page is.
- Information Gain — content that adds something not already in the top-cited sources. Repetition gets filtered.
What the research actually says
There's no shortage of AEO speculation. The hard evidence is narrower:
- Princeton GEO (2024) — targeted page edits delivered up to 41% lift in citation visibility across generative engines.
- Local Falcon (May 2025) — measured 40.2% lift from on-page optimization for local AI search visibility.
- Surfer SEO (2024) — YouTube videos appearing in AI Overviews at 23% rate, suggesting multimodal source breadth.
- CXL (2024) — top-30% pages by structural quality capture 55% of AI citations in their query set.
- Vercel / MERJ (2024) — pages requiring JavaScript to render content are systematically excluded from many AI crawlers.
- Reddit citations in AI Overviews jumped 450% (mid-2024 vs early 2024) — community-generated content carries weight.
What actually works on a page
The page-level work that moves the needle clusters around six patterns:
- Lead with an answer paragraph (50-100 words) that names the entity, the topic, and the answer in plain prose.
- Use H2 headings phrased as questions or direct claims, not marketing copy.
- Cite verifiable statistics inline with source name ("Princeton 2024", "Local Falcon May 2025") — never "studies show."
- Add Schema.org JSON-LD (Article, FAQPage, Organization, Person) so the structure is machine-explicit.
- Render content in HTML, not via client-side JavaScript — many AI crawlers don't execute JS.
- Maintain a llms.txt at the site root pointing crawlers to your most citation-worthy URLs.
Anti-patterns that look like AEO but aren't
The space is full of bad advice. A few things to avoid:
- Stuffing your page with FAQ blocks that don't match real user query language. Quality > quantity.
- Generating content with an LLM and shipping it unedited — the citation engines can often detect generic LLM phrasing and downweight it.
- Linking out to dozens of irrelevant sources to appear well-researched. Citation graphs notice link-quality patterns.
- Adding Schema.org markup that contradicts the visible page content. Google penalizes mismatched structured data.
- Buying AEO from agencies that won't tell you what they're actually changing on your pages.
