The GEO glossary: 15 terms every marketer should know

GEO borrowed half its vocabulary from SEO, invented the other half from scratch, and then let vendors name the same concept three different ways. Rather than cover everything, here are the 15 terms that actually come up in day-to-day GEO work.

GEO (generative engine optimization)

Structuring content and brand presence so AI engines like ChatGPT, Perplexity and Gemini retrieve, synthesize and cite it in their answers, rather than optimizing for a ranked position in a list of links. A coffee subscription brand doing this might rewrite its FAQ with a direct answer to "how much coffee does a subscription include," so ChatGPT can lift that answer straight from the page when someone asks for dark-roast recommendations.

AEO (answer engine optimization)

The narrower practice of becoming the direct, extracted answer itself, whether that's a featured snippet, a voice assistant response, or an AI Overview panel, rather than just being mentioned somewhere in one. A recipe site doing this well might restructure its ingredient list into a clean block near the top of the page, which is exactly the block Google's AI Overview starts pulling when someone searches "how long to boil an egg."

Large language model (LLM)

The type of AI model behind ChatGPT, Claude and Gemini, trained on large volumes of text to generate human-like responses; nearly every GEO concept exists to influence what one of these models says about a brand. When a customer asks Claude to compare accounting software for freelancers, Claude is the LLM doing the comparing, drawing on whatever it can retrieve and reason about.

RAG (retrieval-augmented generation)

The technique most AI search products use to answer questions: retrieve a handful of relevant documents, then generate a response grounded in what was found, rather than relying only on training data. If someone asks Perplexity about a retailer's return policy, it retrieves the retailer's actual policy page and a review or two before writing its answer, which is also why an outdated policy page leads to an outdated answer.

Citation

A specific instance of an AI response naming a source, whether as a clickable footnote, an entry in a source panel, or just a mention in the running text; it's the closest thing GEO has to a ranking position. Asked whether a hiking backpack holds up on multi-day trips, ChatGPT might answer by naming a specific outdoor gear forum thread as its source, and that named thread is the citation.

Share of voice (citation share)

The percentage of citations for a topic that go to your brand versus competitors, GEO's version of the traditional marketing metric of the same name. A bank tracking 20 prompts about "best savings account Denmark" and getting cited in 6 of the resulting AI answers has a 30 percent share of voice for that set, a number worth watching against competitors over time.

Entity

Anything an AI model can identify as a distinct, nameable thing, most relevantly a brand, product or person; models increasingly reason about entities and how they relate to each other rather than just matching keywords. A model doesn't just see the string "Novo Nordisk" in a sentence, it recognizes Novo Nordisk as a specific company, connected to its headquarters, products and executives as related entities.

E-E-A-T

Shorthand for experience, expertise, authoritativeness and trustworthiness, a framework Google built for search quality that's now used across GEO for the qualities AI models seem to reward when deciding what to trust. An article on diabetes management written by a named endocrinologist with a stated hospital affiliation carries a stronger E-E-A-T signal than an anonymous blog post making the same claims, and is more likely to be treated as trustworthy.

Schema markup (structured data)

Code added to a page, usually in JSON-LD format, that explicitly labels what content is, a product, a review, an organization, so machines can read it without inferring it from prose. An e-commerce page with schema stating exact price and stock status lets a model answer "is this in stock right now" directly, instead of trying to guess from marketing copy that may not mention stock at all.

AI Overviews

Google's AI-generated summary appearing above traditional results for many queries, pulling from a small set of sources it displays as citations. Searching "is Copenhagen expensive to visit" triggers an Overview summarizing typical costs with two or three travel sites listed as sources, before the normal results appear further down.

Query fan-out

How a model breaks one user question into several related sub-queries before generating an answer, retrieving sources for each separately. Someone asking "best running shoes for flat feet" might trigger separate retrievals behind the scenes for overpronation, specific shoe models, and podiatrist advice, meaning your content can get pulled in for a sub-question it never directly answers.

Hallucination

When a model states something confidently that isn't true, often by filling a gap in its sources with a plausible-sounding guess. A newly rebranded company with little information online yet might get its founding date stated three years wrong by an AI model that had nothing reliable to draw on and didn't flag the uncertainty.

Zero-click search

An AI interaction that resolves entirely inside the answer itself, with the user never clicking through to a website. Someone asking Gemini for a store's holiday opening hours gets the answer directly in the chat window and never visits the store's site, even though that site was the actual source.

AI crawler

A bot AI companies use to fetch and index web content, such as OpenAI's GPTBot, Anthropic's ClaudeBot, or PerplexityBot, controlled through a site's robots.txt the same way search crawlers always have been. A publisher blocking GPTBot to keep articles out of training data also blocks ChatGPT from retrieving and citing those same pages in live answers, since one file governs both.

Knowledge graph

A structured database of entities and their relationships, used to answer factual questions without reading a page in real time. Searching a company's name and seeing its founding date, headquarters and executives appear instantly in a panel is Google's Knowledge Graph at work, not any single article being read on the spot.

Using this list

These 15 terms cover the fundamentals: what GEO is trying to achieve, how models actually retrieve and generate answers, and how to talk about measuring it. If you want to see a few of these ideas play out in real citation data rather than in the abstract, our piece on what AI actually cites breaks down the specific page types behind real citations across 100+ brands.