Glossary
The vocabulary of AI, agents and GEO — explained plainly, without needless jargon.
AEO (Answer Engine Optimization)
AEO optimizes content for answer engines and featured snippets: a direct answer at the top of a section, questions as headings, structured FAQ, schema.org data. The goal is the clearest, most extractable answer to a specific intent.
AI agent
An AI agent doesn't just answer — it acts. Where a chatbot returns text, an agent chains actions — opening a ticket, booking a meeting, updating a CRM — using tools and guardrails to reach a goal.
CRO (conversion rate optimization)
CRO covers the methods for increasing the share of visitors who take the intended action — buy, book, sign up. It relies on speed, message clarity, friction removal and testing, rather than on more traffic.
Deterministic orchestration
Deterministic orchestration means the critical logic runs on predictable, tested code (rules, state machines), and the language model is only called at fuzzy decision points. You keep control and reliability where correctness is non-negotiable.
Embeddings
An embedding represents a piece of text (or image) as a vector of numbers, where closeness between vectors reflects closeness in meaning. It's what lets a machine 'understand' that two different phrasings mean the same thing.
Fine-tuning
Fine-tuning means retraining an existing language model on your own data to specialize its behavior or style. It's powerful but often less relevant than RAG: use it only when prompting and context are no longer enough.
GEO (Generative Engine Optimization)
GEO is optimizing content to be cited by AI answer engines (ChatGPT, Perplexity, Google's AI Overviews), not just to rank in blue links. Information is structured as direct, sourced, extractable answers.
Hallucination (AI)
A hallucination is a language-model answer that sounds plausible but is false or fabricated. The model generates statistically likely text with no guarantee of truth, which becomes dangerous whenever a fact or rule must be exact.
LLM (large language model)
An LLM is an AI model trained on massive text corpora to predict the next word. That makes it an excellent natural-language interpreter — summarizing, classifying, writing, conversing — but a poor guarantor of exact facts without verification.
MCP (Model Context Protocol)
MCP is an open protocol that standardizes how a language model connects to external tools, data and services. Instead of a bespoke integration per tool, it offers a common interface for giving an agent capabilities.
MVP (minimum viable product)
An MVP is the simplest version of a product that already delivers value and lets you test a hypothesis with real users. It's not a rushed product but a scope tightened around the core of the proposition.
Prompt engineering
Prompt engineering is the craft of shaping the instructions given to a language model to get reliable, useful outputs. Far more than a trick, it's a discipline: structuring the request, giving the right context, constraining the format and anticipating edge cases.
RAG (Retrieval-Augmented Generation)
RAG is a technique where the language model answers using documents retrieved in real time, rather than its memory alone. Relevant passages are searched, injected into the context, and the model generates an answer grounded in those sources.
ROAS (return on ad spend)
ROAS measures the revenue generated for each euro spent on advertising. A ROAS of 4 means €4 of revenue per €1 invested. It's the central metric for judging campaign profitability — but read it alongside margin, never alone.
Voice agent (voicebot)
A voice agent is an AI system that holds a real-time phone conversation: it understands speech, decides, and replies with a synthetic voice. Modern speech-to-speech architectures cut latency for a natural exchange.