AI terms in plain English

You can use AI without collecting a second vocabulary. Look up the term you need, then get back to the task.

These explanations are deliberately short. Product names and feature labels vary, and vendors sometimes use the same word differently.

What you'll find

33 short explanations of common AI terms, with examples of what each means at work.

How to use it

look up one term using the contents list or your browser’s Find command, then return to the guide or task you were working on.

The tool you are using

AI

Artificial intelligence is a broad name for computer systems that perform tasks such as recognising patterns, making predictions or generating content. It describes a family of technologies, not one product.

At work: ask what a particular system can do, what information it uses and how you will check it.

Generative AI

AI that produces new content, such as text, images, audio or code, from patterns learned during training and the information supplied to it.

At work: a fluent new paragraph may still contain an incorrect fact or an unapproved promise.

LLM

A large language model is a model trained on large amounts of data to work with language. It generates responses using learned patterns and the context available to it. Many current systems can also work with images or other kinds of input.

At work: it can help draft and organise material, but it does not automatically know your current prices, policies or agreements. Background on model development.

Model

The trained system doing the processing. An app can offer several models with different capabilities, speed, costs and limits.

At work: ChatGPT is an app. The model selected inside it is one part of the service. Changing the model does not necessarily change your account’s data policy.

Multimodal

Able to work with more than one kind of information, such as text and images. Exactly which inputs and outputs are supported depends on the model and product.

At work: a tool that accepts a PDF may still handle its text differently from its charts. Check the feature rather than assuming “multimodal” means it sees everything.

What you give it

Prompt

The request and material you give the AI to guide its answer. It can include instructions, source text, examples, images or other supported inputs.

At work: “Turn these approved notes into six numbered steps” is a prompt. A good one explains the task, source, constraints and intended output. Prompting basics.

Prompt engineering

Designing and improving prompts to get more useful, consistent results.

At work: give a clear brief, inspect the answer and change the instruction that caused the problem. You do not need a grand title for doing this carefully.

Context

The information available to the model for the current response. It can include instructions, conversation history, supplied documents and results from tools.

At work: important facts need to be available in the right place. Do not assume the model has seen a document because it exists somewhere in your business.

Context window

The limit on how much information a model can work with in one request, measured in tokens. Products may manage longer work using summaries or retrieval. The usable space depends on the model and how the product assembles the request.

At work: a large context window is not a guarantee that every relevant detail will be noticed or used correctly.

Token

A unit a model uses to process information. For text, it might be a word, part of a word, punctuation or a character. Token counts are not word counts and vary by model and language.

At work: tokens affect context limits and often API costs. You rarely need to count them for a short ordinary chat. Tokens and limits.

Project

A product feature for keeping related chats, reference material and instructions together. The exact memory and sharing behaviour varies by service and settings.

At work: a project can help with recurring tasks. It does not, by itself, give a personal account business-plan protections. Example: ChatGPT projects.

Memory

A feature that can carry selected information or preferences between conversations. It may operate differently inside and outside projects.

At work: check what is remembered and where it may be used. Memory settings and model-training settings address different things. OpenAI’s data-control explanation.

What comes back

Output

What the system produces: an answer, image, document, code or other result.

At work: an output is a draft until the relevant checks and approvals are complete.

Hallucination

A plausible-looking but incorrect or unsupported output. This can include made-up facts, quotes, citations or explanations.

At work: a calm tone and an exact-looking number do not prove accuracy. Check important claims against the original source. Reliability limitations.

Grounding

Basing an answer on relevant evidence supplied or retrieved for the task.

At work: ask the tool to use the approved policy and identify the supporting section. Grounding can help; the answer still needs checking.

Citation

A reference telling you where a claim supposedly came from.

At work: open it and check that it supports the statement. The existence of a link is not proof that the source says what the answer claims.

Inference

In AI, inference means using a trained model to produce a prediction or response from new input. Training prepares the model; inference is the model being used. Google’s technical definition.

The word also has an everyday meaning: a conclusion drawn from evidence rather than something directly stated in it.

At work: an “inference cost” usually refers to running the model. An “inference in this report” may mean an interpretation you should check. “The shipment was delayed” may be recorded; “the supplier is unreliable” needs more evidence.

Evaluation

A planned way to judge whether a tool or workflow meets your requirements, using examples and clear checks.

At work: test whether the tool preserves dates, flags missing information and produces a usable format. One attractive answer does not establish reliability.

Synthetic or illustrative example

Material created to demonstrate something rather than report a real event or result. A fictional customer, sample dataset or mock answer belongs in this category.

At work: label it clearly. Do not present an editorial example as an actual response from a named product or as evidence that a workflow was tested.

Files, actions and automation

Retrieval

Finding relevant material from a source such as a document collection, search index or connected system.

At work: the system may retrieve selected passages rather than place your entire library into every response.

RAG

Retrieval-augmented generation. The system finds relevant information and supplies it to the model as context for an answer.

At work: a company-policy assistant might retrieve passages from a handbook before answering. Missing or outdated source material can still produce a poor answer. RAG does not mean the underlying model has been retrained on your files. Anthropic’s explanation.

Tool

A capability the AI can call, such as a calculator, search service or file editor.

At work: some tools only read information; others can change it. Ask what the tool is allowed to do before granting access.

Connector or integration

A link between the AI app and another service, such as cloud storage, email or a calendar. Products use different names, including apps and plugins.

At work: check the permissions and the account being connected. A useful connection can expose far more information than a single uploaded document.

Agent

An AI system that can work towards a goal over multiple steps, often deciding which tools to use and how to continue. The word is used loosely, so check the actual behaviour.

At work: define the goal, boundaries, permissions and stopping point. A tool that can send messages needs more oversight than one that only drafts them. Anthropic’s distinction between workflows and agents.

Automation

A process that runs when a trigger or schedule is met. It may use fixed rules, AI or a mixture of both.

At work: test it on a small, reversible task. Decide who reviews failures and how it can be stopped before making it recurring.

API

Application programming interface. A way for software to ask another system to do something or return information.

At work: an application might send a document to a model through an API. Ordinary chat does not require you to set one up. Chat subscriptions and developer API access can have separate billing. Anthropic’s product distinction.

API key

A credential used to authenticate software access to a service. Depending on its permissions, someone holding it may be able to use services or incur charges.

At work: keep it in approved secret storage. Never paste it into a prompt, public document or screenshot.

Prompt injection

Instructions placed in material the AI encounters, such as a webpage or document, that try to redirect what it does. They may conflict with the task you gave it.

At work: treat source content as evidence to inspect. Avoid granting unnecessary action-taking permissions, and do not assume a warning in your prompt can block every attack. Prompt-injection guidance.

Accounts and information handling

Workspace

An area in an AI service used by an individual or organisation. A managed organisation workspace may have central membership, billing and administrative policies.

At work: confirm which workspace is active before adding company material. A workspace name alone tells you nothing about what data has been approved for it.

SSO

Single sign-on. Signing in to a service through an organisation’s identity system.

At work: use the route your administrator gives you. An existing personal account and a company-managed sign-in are not automatically the same arrangement.

MFA

Multi-factor authentication. An additional verification step for signing in, beyond a single credential.

At work: enable the approved methods available to your account and plan how you would recover access. Example: OpenAI MFA.

Model training

The process of adjusting a model using data. This is different from giving an existing model a document as context for one task.

At work: check the provider’s policy for your exact account type and your settings. Do not infer the answer from whether you pay for the app.

Retention

How long a service keeps information, subject to its policies, settings and applicable exceptions.

At work: “not used for training” and “not stored” are different claims. Check both, alongside access and sharing.

Keep this: When a term affects your data, cost or permissions, check what it means in the product you are actually using.

Sources and status

Official references checked on 4 October 2026. Definitions are plain-English editorial summaries rather than contractual or exhaustive technical definitions. For a practical starting point, follow the beginner learning path.