GEO/AEO: getting cited by the answer.

The blue link is not dead, but it has been demoted. Google's AI Overviews now appear on a majority of informational queries in the US and have measurably cut click-through to the underlying pages [1]. ChatGPT and Perplexity together serve billions of queries a month that never resolve to a traditional SERP at all [2]. The unit of distribution is no longer the ranked page. It is the named entity inside a synthesized answer.

That shift is what GEO (Generative Engine Optimization) and AEO (Answer Engine Optimization) exist to address. They are not SEO with a new logo. They optimize for a different surface, with different mechanics, and a different definition of winning.

What actually changed

Three mechanics moved at once.

• Retrieval replaced ranking. LLM-backed answers are assembled from a retrieval step (the model's training corpus, plus a live web fetch for grounded systems) and a synthesis step. Both steps reward structured, quotable, entity-rich passages — not keyword-stuffed pages [3].
• The citation is the impression. When Perplexity or Google AI Overviews name a brand in the answer body, that is the equivalent of a top-three SERP placement. When they only footnote a brand, click-through collapses. Princeton's GEO benchmark paper measures this directly and shows it is reproducible across engines [3].
• Entities outweigh keywords. The models reason over entities — people, products, places, organizations — and their relationships. A brand that is well-described as an entity (in Wikidata, in schema.org markup, in consistent third-party copy) gets pulled into more answers than one that is not [4].

What GEO is, concretely

GEO is the practice of shaping a brand's public footprint so generative engines pick the brand up cleanly and quote it accurately.

1. Entity engineering. Establish a canonical entity for the brand and each named product. That means a structured Wikidata record, schema.org JSON-LD on every owned page, consistent naming across third-party citations, and a public knowledge graph the models can crawl [4].
2. Answer-targeted content. Write to the question, not the keyword. Lead with the answer in the first 60 words. Use claim-style sentences a model can lift verbatim — "X is a Y that does Z" — and surround them with the citations a grounded engine will need to trust the claim [3].
3. Quotable structure. Lists, comparisons, definitions, and short factual tables are over-represented in AI Overview pulls. Long expository prose is under-represented. The structure of the page is part of the optimization [1].
4. Cross-engine presence. ChatGPT, Perplexity, Gemini, Claude, and Google AI Overviews each pull from a different mix of training and live sources. A brand that only optimizes for Google search is invisible inside ChatGPT, where there is no live crawl by default and the model's prior knowledge is the entire SERP [2].

What AEO adds

AEO is the narrower discipline of being the answer to a specific question — voice assistants, featured snippets, AI Overviews' direct-answer band. It rewards:

• Direct, declarative answers in the first paragraph of any page that targets a question intent.
• FAQ and HowTo schema on every supportable page, because both are over-pulled by every major answer engine [4].
• Conversational phrasing that matches how users speak to assistants, not how they type into a search bar.

GEO is the brand-level program. AEO is the page-level tactic. Both have to be in place.

What dies

Keyword density. Backlink farming. Programmatic SEO at scale. None of these survive a synthesis layer that reads for meaning and cites for trust. Google's own March 2024 core update and its 2024–2025 spam policy enforcement removed an estimated 45% of low-quality content from results [5] — and the AI Overviews that replaced those clicks reward the opposite signal: a small number of well-structured, well-cited, entity-clean pages.

What we do for clients

The practical question is not "how do we rank for this term." It is "how do we make sure the answer names us, accurately, in the brand's own language."

We run a four-step program: an LLM citation audit across ChatGPT, Perplexity, and Gemini for the brand's twenty most important prompts; an entity-engineering pass that fixes Wikidata, schema.org, and third-party canonical copy; an answer-targeted content rewrite focused on the brand's defensible claims; and a quarterly re-test against the same prompt set so the program has a measurable line.

The takeaway

The SERP is becoming a synthesis. The winners will be the brands the synthesis names by name. Build for the citation, not the click.