SEO, AEO and GEO glossary

Plain definitions of the terms used across this guide. Where a term has an official definition, the primary source is linked.

The disciplines

SEO, search engine optimization. The work of being found, indexed and ranked by search engines. Covers technical health (crawlability, speed, structure), content quality, and links.

AEO, answer engine optimization. The work of being the source an engine quotes when it answers a question directly: in Google's AI Overviews and AI Mode, or in assistants like ChatGPT and Perplexity. An answer cites few sources, so selection is winner-takes-most.

GEO, generative engine optimization. The work of earning citations inside generated answers. The founding research found that quotations, statistics and cited sources lift visibility in generative answers by roughly 41, 30 and 27 percent respectively, while structure alone does not.

How engines work

Crawling. An engine's automated visit to your pages. If a crawler cannot reach a page, nothing downstream can happen to it.

Indexing. The engine storing and organising what it crawled. Google is explicit that a page must be indexed and snippet-eligible before any AI feature in Search can use it.

RAG, retrieval-augmented generation. The technique behind grounded AI answers: the model retrieves relevant pages first, then writes its answer from what it retrieved. Google describes its AI features in Search as RAG grounded in the core ranking systems.

Grounding. Tying generated text to retrieved sources so the answer reflects live web content rather than only what the model memorised in training.

Query fan-out. Google's term for issuing several related queries concurrently behind one question, gathering broader context before generating the answer. One page can be retrieved by a fan-out query the user never typed.

Citation. The link an answer engine attaches to a claim in its generated answer. Citations are the currency of AEO and GEO.

AI Overviews and AI Mode. Google's generative experiences in Search: a generated summary above results, and a fully conversational search mode. Both select sources through the mechanics above.

Content and quality

Helpful content. Google's umbrella term for people-first content: made for readers, demonstrating first-hand experience, satisfying the visit. The opposite of content made primarily to rank.

Thin content. Pages with little substance beyond what already ranks: template text, spun summaries, undifferentiated roundups.

Scaled content abuse. Google's spam policy against producing many pages primarily to manipulate rankings rather than help users. The detected pattern is the cluster fingerprint: templated narratives, shared infrastructure, coordinated publish frequency. Note that a study across 1,000,000 pages found 8.4 to 11.7 percent of top-10 results are substantially AI generated with stable impressions, so AI authorship in itself is not the trigger; the pattern is.

E-E-A-T. Experience, expertise, authoritativeness, trustworthiness: the lens Google's quality rater guidelines use to describe good sources. Guidance for evaluating content, not a direct ranking factor you can inject.

Primary source. The origin of a fact: a standard, a statute, official statistics, peer-reviewed work, vendor documentation, first-party data. Citing primary sources is one of the three measured GEO lifts.

Substance density. Verifiable claims per hundred words. A practical proxy for whether a page carries information or filler.

Technical terms

llms.txt. A proposed file for telling language models about your site. Google states that Google Search ignores llms.txt files; they neither harm nor help there. Other crawlers set their own policies.

Structured data / schema markup. Machine-readable annotations (JSON-LD) describing page content. Useful for classic rich results; Google states it is not required for generative AI search.

Snippet eligibility. Whether a page may be shown with a text snippet in results. A precondition for appearing in Google's AI features.

hreflang. The annotation that maps language and regional versions of the same content to each other.

IndexNow. A protocol for notifying participating search engines about new or changed URLs immediately, instead of waiting for a crawl.

Measurement

Share of voice. How often a brand appears in answers for a prompt set, relative to competitors.

Measurement variance. The instability of AI answers: three identical ChatGPT runs leave roughly 2.2 to 2.3 percent of cited sources stable, and query language accounts for 26.5 to 32.0 percent of variance in brand responses while brand identity accounts for 1.5 percent. A single check is noise; defensible measurement runs the same prompts across languages, models and phrasings, and reports confidence intervals.

Confidence interval. The range a measured value plausibly sits in, given the noise. Visibility reported without one is a guess with a number attached.

Generative AI performance report. The Search Console report where Google shows how your site performs in its AI experiences. The only first-party window; Google notes no third-party tool has access to its internal ranking or AI systems.