AI Glossary for Business: Plain-English Definitions

AI glossary

Acceptable use policy (AI)

A short company document that says which AI tools staff may use, what information they can put into them, and who checks AI output. Why it matters: It’s the cheapest protection against data leaks. Create one in five minutes.

AI agent

AI that can take actions toward a goal, not just answer questions. It can search, fill in forms, update records or send messages, often across several steps. Why it matters: Agents can save real time on repetitive work, but they need clear limits and human approval for anything important. See AI automation.

AI assistant

A general-purpose chat tool such as ChatGPT, Claude, Gemini or Microsoft Copilot that helps with writing, research, analysis and brainstorming. Why it matters: Usually the first AI tool a business adopts. Compare the major assistants.

Algorithm

A set of step-by-step rules a computer follows to solve a problem or make a decision. AI systems learn many of their rules from data rather than having them written by hand.

API

Application programming interface: a way for one piece of software to talk to another. AI vendors offer APIs so developers can build AI into their own apps and workflows. Why it matters: API use is usually billed by usage (tokens), separately from chat subscriptions.

Bias

When AI produces systematically unfair or skewed results, often because of patterns in its training data. Why it matters: It’s especially important in hiring, lending and other decisions about people, where several laws now apply. See AI regulations.

Chatbot

Software that holds a conversation with users. Modern AI chatbots understand natural language and can answer from your own content. Why it matters: One of the fastest wins for customer service. Chatbot buyer’s guide.

Context window

How much text, measured in tokens, an AI model can consider at once, including your prompt, any documents you attach and its reply. Why it matters: Larger windows let you analyze longer documents in one go.

Copilot

Microsoft’s family of AI assistants, built into Windows, Edge and Microsoft 365 apps like Word, Excel, Outlook and Teams. The word is also used generically for any AI that works alongside you. Why it matters: If you’re on Microsoft 365, it’s often the simplest assistant to roll out.

Data classification

Sorting your information into sensitivity levels, such as public, internal and restricted, so people know what can go into which tools. Why it matters: It’s the core of a good AI policy. See Is your data ready for AI?

Deepfake

AI-generated or AI-altered video, audio or images that convincingly imitate a real person. Why it matters: Criminals use deepfake calls to impersonate executives and request payments. How to protect your business.

Embedding

A way of turning text into a list of numbers that captures its meaning, so software can find similar content even when the wording differs. It powers AI search over your documents.

Fine-tuning

Further training an existing AI model on your own examples so it adopts a particular style or skill. Why it matters: Most small businesses don’t need it. Good prompts and RAG usually get you there cheaper.

Generative AI

AI that creates new content, such as text, images, audio, video or code, rather than only analyzing existing data. ChatGPT and image generators are examples.

Grounding

Connecting an AI’s answers to specific, trusted sources, such as your documents or the web, so responses are based on real information and can cite it. Why it matters: Grounding reduces hallucinations in customer-facing tools.

Guardrails

Rules and technical limits that keep AI within safe boundaries, for example blocking certain topics, requiring citations or routing questions to a person.

Hallucination

When AI states something false with confidence: a made-up fact, statistic, quote or citation. Why it matters: Always verify AI output before it reaches customers. See our AI writing workflow.

Inference

The act of an AI model producing an answer from your input, as opposed to training, which is how the model learned in the first place.

Large language model (LLM)

The type of AI behind chat assistants: a model trained on huge amounts of text to understand and generate language. GPT, Claude and Gemini are LLMs.

Machine learning

The broader field of AI in which computers learn patterns from data instead of following hand-written rules. It powers fraud detection, recommendations and forecasting.

MCP (Model Context Protocol)

An open standard that lets AI assistants connect to other tools and data sources, such as your CRM, files or calendar, in a consistent way. Why it matters: It’s how many AI agents reach your business systems, so treat each connection like any other app permission.

Model

The trained AI system that does the work. Vendors offer several models with different speed, cost and capability, for example a fast model and a more capable “reasoning” model.

Multimodal

AI that can work with more than one type of input or output, such as text, images, audio and video. You can show it a photo of a receipt and ask questions about it.

Natural language processing (NLP)

The branch of AI that deals with understanding and generating human language. It’s behind chatbots, transcription, translation and sentiment analysis.

Open-weight model

An AI model whose trained weights are published, so it can be downloaded and run on your own servers. Some are fully open source. Why it matters: They give you more control over data, but you need technical skills to run them securely.

Parameters

The internal numbers a model learns during training. More parameters often, but not always, means a more capable model.

PII (personal information)

Personally identifiable information: anything that identifies a person, such as names, addresses, ID numbers, account numbers or health details. Why it matters: Keep PII out of AI tools unless they’re approved for it.

Prompt

The instructions or question you give an AI. Clear prompts with context, a goal and the format you want get far better results. Prompt templates.

Prompt injection

An attack where hidden instructions in a web page, email or document trick an AI into doing something it shouldn’t, like revealing data or taking an action. Why it matters: It’s a key risk for AI agents and tools connected to your email or files.

RAG (retrieval-augmented generation)

A technique where the AI first looks up relevant information in your documents, then answers using it, often with citations. Why it matters: It’s how “chat with your documents” and customer-service bots stay accurate. See our CustomGPT guide.

Reasoning model

An AI model that works through a problem step by step before answering. It’s slower and costs more, but it’s better at math, planning, analysis and complex questions.

Shadow AI

Employees using AI tools the company hasn’t approved, often free personal accounts, for work tasks. Why it matters: It’s the most common way business data leaks into AI. A clear policy and an approved alternative fix it.

Synthetic media

Any AI-generated image, video, audio or text. Some laws and ad platforms now require labeling it in certain situations.

Temperature

A setting that controls how creative or random an AI’s responses are. Low is more predictable; high is more varied.

Token

The unit AI models use to measure text, roughly three-quarters of a word. Usage limits and API pricing are usually counted in tokens.

Training data

The information a model learned from. Why it matters: Check whether a vendor uses your data to train its models. Business plans usually don’t. See free vs paid AI tools.

Transformer

The neural network design, introduced in 2017, that today’s language models are built on. It’s the “T” in GPT.

Vector database

A database that stores embeddings so software can quickly find content with similar meaning. It’s the search engine behind many RAG systems.

Voice cloning

Creating a synthetic copy of someone’s voice from a recording sample. It’s used legitimately for narration, and by scammers to impersonate people. Responsible AI voice use.

Zero data retention

A vendor commitment not to store your prompts and outputs after processing. It’s often available on enterprise or API plans for sensitive uses.