AI Glossary
The terms that come up when a company talks about AI, in five groups: foundations, models and technology, working with AI, leadership and law, search and visibility. Each term with a definition, how it works, common questions, an example and the point where it stops carrying weight.
75 terms
No term matches. Reset the filter or try another spelling.
A
- AI agentsAn AI agent is a system in which a language model pursues a goal in steps, decides on its own which tools to call, and checks the results until the task is done.
- AI and copyrightAI and copyright covers three questions: whether models may be trained on protected works, whether AI output is protected, and when an output infringes someone else's rights.
- AI assistantAn AI assistant is software built on a language model that completes a task when a person asks, such as writing a draft, and hands the result back for review.
- AI benchmarkAn AI benchmark is a standardized test set with fixed tasks and uniform scoring that lets you compare AI models on a specific capability.
- AI biasAI bias refers to systematic distortions in an AI system’s results that repeatedly favor or disadvantage certain groups or characteristics.
- AI ethicsAI ethics covers the principles and trade-offs that determine whether AI systems are built and used responsibly, including in areas where no law provides an answer.
- AI guardrailsAI guardrails are technical controls outside the model: checks and limits that define what an AI application may accept, output and execute.
- AI inferenceAI inference is running a fully trained AI model on new input, such as a language model writing an answer; the model's weights do not change.
- AI labeling and disclosureAI labeling and disclosure means indicating that content was wholly or partly generated or altered by AI, either visibly for people or in machine-readable form.
- AI orchestrationAI orchestration is the control layer that connects several AI models, tools, data sources and review steps into one traceable workflow and decides which step may do what.
- AI OverviewsAI Overviews are summaries created with generative AI that Google shows above its search results, with links to the web pages they draw on.
- AI ROIAI ROI weighs the benefit of an AI initiative against its full cost, calculated over a stated period and with disclosed assumptions.
- AI safety and securityAI safety and security covers two risks: an AI system causing harm on its own (safety) and attackers exploiting it (security). Red teaming tests both with deliberate attacks.
- AI searchAI search is a form of web search in which a language model retrieves pages relevant to a question and writes an answer from them, with links to its sources.
- AI TransformationAI transformation is the shift of an entire company to AI: workflows, skills and responsibilities change because AI becomes a fixed part of daily work.
- AI Use CaseAn AI use case describes a specific, recurring task that AI is meant to support, with the starting situation, data needed, desired result, the person who reviews it and quality criteria.
- AI washingAI washing means claims that credit a product, service or company with more artificial intelligence than it actually uses or can be shown to deliver.
- Artificial intelligence (AI)Artificial intelligence (AI) refers to computer systems that infer from data or encoded knowledge how to generate outputs such as predictions, content, recommendations or decisions.
C
- Chain of thought (CoT)Chain of thought (CoT) is a prompting technique in which a language model writes out the intermediate steps of a task before it answers. It often improves multi-step tasks.
- ChatbotA chatbot is a program that takes input in natural language, typed or spoken, and responds in a dialogue, either by following fixed rules or by using a language model.
- Chunking and rerankingChunking splits documents into smaller passages so an AI search can find the right one; reranking then reorders the retrieved passages with a more accurate model.
- Context engineeringContext engineering is the deliberate design of everything a language model sees in a call: instructions, retrieved knowledge, conversation history, tools and intermediate results.
- Context WindowThe context window is the maximum number of tokens a language model can take into account in a single call, including instructions, history, documents and, depending on the model, the answer.
D
- Data governanceData governance is the framework of roles, rules and decision paths that sets who in a company decides about which data, how it is maintained and who may use it.
- Data qualityData quality describes how well data fits its specific purpose, measured by dimensions such as accuracy, completeness, consistency, uniqueness, validity and timeliness.
- Deep learningDeep learning is a branch of machine learning in which neural networks with many layers learn the relevant features on their own from raw data such as images, speech or text.
- DeepfakeA deepfake is AI-generated or manipulated image, audio or video content that resembles real people, places or events and would falsely appear authentic.
E
- E-E-A-TE-E-A-T stands for experience, expertise, authoritativeness and trustworthiness: four qualities Google's quality raters use to assess how reliable a piece of content and its source are.
- EmbeddingsEmbeddings are numeric vectors that a model computes from text or other content so that similarity in meaning can be measured by calculation.
- Evals (LLM evaluation)Evals are repeatable tests that use fixed test cases and scoring criteria to measure whether an AI system handles a specific task well enough, including after every change.
- Explainable AIExplainable AI (XAI) covers methods that show why an AI model reached a result, either by using a model that is readable on its own or by explaining a complex model after the fact.
F
- Few-shot promptingFew-shot prompting means giving a language model a few examples of input and desired output in the prompt so it can infer the task, format, and tone without being retrained.
- Fine-TuningFine-tuning is the additional training of a pretrained AI model on your own reviewed examples so that it handles a specific task with a set format, style or behavior.
- Foundation modelA foundation model is an AI model pretrained on very broad data that can be adapted to many different tasks; large language models are the best-known example.
- Function calling (tool use)Function calling is an API feature in which a language model proposes a call to a described function, with parameters, and the application runs that call.
G
- General-purpose AI (GPAI)GPAI (general-purpose AI) is the EU AI Act’s legal term for general-purpose AI models: broadly usable models whose providers carry their own set of obligations.
- Generative AIGenerative AI is artificial intelligence that uses learned patterns to create new content such as text, images, audio, video or code, rather than only classifying existing data.
- Google AI ModeAI Mode is a dedicated part of Google Search where a Gemini model answers complex questions conversationally and backs the answer with links to web pages.
- Grounding (AI)Grounding ties a language model’s answer to verifiable sources available at the time of the request, such as documents, search results or databases, instead of training data alone.
L
- LLM gatewayAn LLM gateway is a central layer between applications and AI model providers that every model call passes through, with one point of access, permissions, cost tracking, logging and swappable models.
- LLM: Large Language ModelAn LLM (large language model) is an AI model that processes and generates text based on learned language patterns, for example to summarize, draft or sort information.
- LLMOps and MLOpsLLMOps is the disciplined operation of applications built on large language models: versioned prompts and models, testing, monitoring, cost control, logging, and fallback paths.
- llms.txtllms.txt is an open proposal for a Markdown file that offers AI agents and language models a website's most important content as selected links with short explanations.
- Local and on-premise AILocal AI means AI models that run on hardware the company itself controls, such as servers in its own data center or an end device, rather than at a provider in the cloud.
M
- Machine learningMachine learning is a branch of artificial intelligence in which software learns its rules from example data instead of having them programmed step by step.
- MCP: Model Context ProtocolThe Model Context Protocol (MCP) is an open standard through which AI applications, acting as clients, use tools and data sources that a server provides through one shared interface.
- Multimodal AIMultimodal AI means AI models that process or generate several kinds of data, such as text, images, audio, video, and documents, and relate their content to each other.
N
- Natural language processing (NLP)Natural language processing (NLP) is the branch of computer science and artificial intelligence in which computers analyze and generate human language.
- Neural networkA neural network is a computational model made of many connected artificial neurons that learns its task from examples by adjusting the weights between neurons during training.
P
- Pilot GraveyardPilot graveyard describes AI pilots that, after the trial, are neither moved into supported routine operation nor deliberately shut down, and simply sit unused.
- Prompt cachingPrompt caching is an AI API feature in which the vendor briefly stores the computed state of an unchanged prompt prefix and reuses it for later requests.
- Prompt EngineeringPrompt engineering is the systematic design and testing of instructions given to an AI model so that the same instruction produces usable results across many different inputs.
- Prompt injectionPrompt injection is an attack in which text reaches a language model’s input and the model follows it as an instruction, even though it was only meant as content.
R
- RAG: Retrieval-Augmented GenerationRAG (retrieval-augmented generation) is a method in which a system searches a knowledge base for relevant content before a language model answers and passes that content to the model as context.
- Reasoning modelA reasoning model is a language model that generates a longer internal chain of thought before it answers, spending more compute on every request than a standard model does.
- Reinforcement learning (RLHF)Reinforcement learning is a machine learning method in which an agent learns by trial and reward which actions lead to its goal over the long run.
- RPA: Robotic Process AutomationRPA (robotic process automation) is software that performs repetitive, rule-based steps in the user interface of existing applications the way a person would by hand.
S
- Semantic searchSemantic search finds content by meaning rather than matching words, usually by comparing the query and the texts as numeric vectors (embeddings).
- Shadow AIShadow AI refers to AI tools that employees use for work without the company having provided, reviewed or approved them.
- Single source of truthA single source of truth (SSOT) is the designated place where a given piece of information is maintained and whose version wins when copies in other systems disagree.
- Small language model (SLM)A small language model (SLM) is a language model with relatively few parameters that runs on a single server or device and suits narrowly defined text tasks.
- Sovereign AISovereign AI means a company can make its own decisions about its AI models, data, and providers: where processing happens, which law applies, and whether the model or provider can be replaced.
- SycophancySycophancy is the tendency of an AI model to tailor its answers to the user’s apparent opinion, even when that makes them inaccurate or one-sided.
- Synthetic dataSynthetic data is artificially generated data that reproduces the structure and statistical properties of real data without containing real individual records.
- System PromptA system prompt is the top-level instruction that the operator of an AI application places before every request to a language model, setting its role, rules and answer format.
T
- TokenA token is the unit a language model splits text into, such as a word, part of a word or a punctuation mark; providers measure both cost and capacity in tokens.
- Topical authorityTopical authority is an SEO term for the assumption that a site covering a subject area thoroughly and with expertise has an advantage in searches on that subject.
V
- Vector DatabaseA vector database stores numeric vectors (embeddings) along with references to the source data and quickly finds the entries most similar to a query.
- Vibe codingVibe coding is a way of working in which a person has an AI write software from plain-language instructions and accepts the code without reading or reviewing it.