Search is no longer a list of links. More and more often the user gets a ready answer from AI and never reaches the site — and that changes the rules for marketing. Google launched AI Overviews in 2024; the llms.txt standard was proposed the same year.
The takeaway is simple. To land in answers, ranking is not enough: you need to understand how engines decide whom to quote. These 32 terms are the working minimum for speaking the same language as your team and your vendors.
Where to start
- Start with the basics: GEO, AEO and AI Visibility Score explain what is being measured at all.
- Get the mechanics: RAG and grounding show how your content lands in an AI answer.
- Close the technical gaps: llms.txt, structured data and AI crawlers are the parts you set up on the site.
Group 01
AI search basics
GEO — Generative Engine Optimization
Optimizing your content so generative AI engines cite it: ChatGPT, Perplexity, Gemini, Grok. The goal is the citation, not the ranking. Be quoted, not just listed.
AEO — Answer Engine Optimization
Preparing content to become the direct answer to a user’s question. It is the broader term — voice search and AI answers both count. GEO is a subset of AEO.
Generative search
The engine composes an answer from several sources instead of returning a list of links. ChatGPT Search, Perplexity and Google AI Mode work this way. One answer, no link list.
Answer engine
A system that answers with text, not links. Perplexity, ChatGPT and AI Overviews are answer engines. The user gets a finished answer and goes no further.
AI Overviews
The AI-generated answer block at the top of Google, launched in 2024. The user reads a ready answer right in search and does not click through. Click-through on the first organic position drops.
SGE — Search Generative Experience
Google’s early name for its generative-answer experiments in search. It grew into AI Overviews in 2024 and AI Mode in 2025. You still meet the term in older articles.
Zero-click search
The user gets an answer right in search and visits no site at all. AI Overviews and AI answers have sharply increased the share of these queries. No click — but the demand is still there.
Conversational search
People search through dialogue, refining a query over several turns as in a chat. ChatGPT and Gemini are built for it. The phrasing is longer and more natural than short Google queries.
Group 02
Metrics and measurement
AI Visibility Score
A single score for how present your brand is across AI answers. It blends three things: how often you are mentioned, in what context, and which competitors stand next to you.
Citation
A mention of or link to your brand inside an AI answer. This is what GEO is aiming for. Your content becomes a source the AI leans on.
Brand mention
Any naming of a brand in text, even without a link. Engines weigh mentions on authoritative sites when they decide whom to name. A link is not required.
Citability
How easy a page is to quote: a direct answer, facts, numbers, a clear structure. The higher the citability, the better the odds of landing in an AI answer.
Sentiment in AI answers
How an engine frames a brand: positively, neutrally or with caveats. Appearing in the answer is not enough. Being framed favourably is what matters.
Hallucination
A confident but false statement an engine invents. Clear sources, structured data and verifiable facts on your site cut the risk that AI gets your brand wrong.
Group 03
How AI engines work
LLM — large language model
A model trained on vast amounts of text that predicts the next word. ChatGPT, Claude and Gemini are built on LLMs. The LLM is the part that writes the answer.
RAG — Retrieval-Augmented Generation
Before answering, the engine retrieves fresh documents and grounds the answer in them. Perplexity and ChatGPT Search work this way. This is why your content can land in an answer at all.
Grounding
Tying an answer to specific sources rather than the model’s memory. A grounded answer carries links. That is exactly where GEO works to place your brand.
Knowledge graph
A database of linked facts about entities — companies, people, products — and the ties between them. AI engines and search use graphs to understand who you are and what you relate to.
Entity
An object an AI recognizes as a distinct thing: a brand, product, person or place. First the AI has to recognize you with confidence. Without that, there is no citation.
Embedding
Turning text into a set of numbers a model uses to measure meaning. Thanks to vectors, AI finds your content by the sense of a query, not an exact word match.
Context window
How much text a model holds in mind at once. The clearer and more compact the answer on your page, the better the chance it fits whole and gets quoted.
Token
The chunk of text a model works in — roughly 3–4 characters, or part of a word. Models measure length in tokens, not words.
Prompt injection
A hidden instruction inside text that tries to make an engine break its rules. It is a security risk. It touches any product that runs third-party content through an AI.
Group 04
Technical optimization
llms.txt
A guide file for AI engines on your site, proposed in 2024 — robots.txt’s counterpart, but for LLMs. It explains who you are, what you do, and which content to trust.
AI crawler
A bot that collects pages for an AI engine: GPTBot, ClaudeBot, PerplexityBot, Google-Extended. Block them and the AI never sees your content.
robots.txt
A file of rules for search and AI bots: which pages to crawl, which to skip. For GEO, the trick is not to accidentally block the crawlers you want to be visible to.
Structured data (Schema.org)
Markup that tells machines what a page means: an article, an organization, a product, an FAQ. The Schema.org vocabulary is backed by Google, Microsoft and Yandex. It helps AI understand and cite your content accurately.
Speakable
A Schema.org property that marks passages fit to be read aloud. It is a signal to voice assistants and AI. It tells them which part of a page can be spoken as the answer.
E-E-A-T
Experience, Expertise, Authoritativeness, Trustworthiness. By these four criteria both Google and AI engines judge whether you can be trusted as a source.
Topical authority
A site’s reputation as an expert on one topic, earned through deep coverage. The more fully you cover a topic, the more often AI picks you as its source on it.
Fine-tuning vs RAG
Two ways to give a model new knowledge. Fine-tuning bakes it into the model itself; RAG feeds fresh documents at answer time. For getting cited, RAG matters more — it works with live site content.
Questions
Frequently asked questions
How is GEO different from SEO?
SEO lifts a page up Google’s rankings. GEO gets an AI engine to cite you in its answer. Some practices overlap — authoritative content and markup matter for both. Some diverge: what moves a Google ranking does not always affect mentions in AI.
Which matters more, GEO or AEO?
AEO is broader: preparing content to be the direct answer, voice search included. GEO is the subset for generative engines. Start with the AEO mindset and measure with GEO metrics like Share of Model.
Do I need llms.txt if I already have robots.txt?
They are different files. robots.txt controls bot access; llms.txt tells engines who you are and which content to trust. They complement each other rather than replace.
Which AI engines should I track first?
The ones your audience uses. For most that is ChatGPT, Perplexity, Gemini and Google AI Overviews. For the Russian market, add Yandex Neuro and Alice.