Glossary

72 terms from AI at Work Basics, in plain words, each with the chapter that teaches it.

Search the glossary

#

[CHECK]: A marker you ask the AI to put on anything it is unsure of, such as an FAQ answer, so you know what to verify. Type it exactly as shown.
[UNCLEAR]: A marker you ask the AI to put on text it can't read in an image, such as a whiteboard photo, so you can fill the gaps yourself.

A

AI (artificial intelligence): Computer programs that do tasks we would normally expect to need a person, such as writing, summarizing or answering questions.
AI agent: An assistant that carries out multi-step tasks for you, such as researching, filling in forms or organizing files, rather than just replying to one message.
AI assistant: A program you talk to in everyday language by typing or speaking, which writes back in seconds. ChatGPT, Claude, Gemini and Microsoft Copilot are examples.
AI policy: A workplace's written rules about which AI tools employees may use, for what tasks, and with which kinds of information. Where none exists, a simple one-page policy can be drafted together with a manager.
Alt text: A short description of an image that screen readers read aloud for people who can't see it. You can ask an AI to write it for a product photo.
Anonymizing: Removing the details that identify a person before pasting text into an AI, for example changing a customer's name and address to "Customer A". The AI can still write a good reply.
Attach: Adding a file such as a PDF or spreadsheet to your message with the paperclip or + button so the assistant can read it alongside your request.
Auto: A model setting where the app decides for you whether to answer quickly or think step by step.

B

Bias: Unfair assumptions about groups of people that the AI picked up from the human-written text it learned from. Read anything it writes about people for them, and don't let it decide about people on its own.
Brainstorming: Asking the AI for many ideas first (for example 20, not 5), then narrowing them down in a separate step.
Business plan: A version of an AI assistant set up by an employer, with stricter privacy. It normally does not use your conversations to train future models.

C

Cell: One box in a spreadsheet grid, named by its column letter and row number, such as B2.
Chart types (line, bar, pie): Use a line chart for change over time and a bar chart for comparing items such as products. For parts of a whole, a bar chart is usually clearer than a pie chart.
Chat: One conversation with an assistant. It sees everything said in that chat so far, but not your other chats.
Chat history: The saved list of your past chats, usually shown down the side of the screen. You can rename or delete chats there.
ChatGPT: An AI assistant made by OpenAI that you chat with in everyday language. It is one of the best-known assistants, alongside Claude, Gemini and Microsoft Copilot, and is the most widely used.
Citation (source): A link to the web page or document a fact came from. When an assistant searches the web it usually lists them so you can open each one and check the claim.
Claude: An AI assistant made by Anthropic, described as strong at writing and long documents.
Connector: A feature that links an assistant to another service such as Google Drive, Gmail or a calendar, so it can find things like last month's supplier invoice. Connect only what you need.
Context: The background facts and situation you give the AI in a prompt, such as who you are, who the reply is for and the facts it must use. It is one of the five parts of the prompt recipe.
Context window: The limit on how much text an assistant can take in at once. In a very long chat, early messages can fall out or get less attention, which is why you start a new chat for each task.
Copilot: Microsoft's AI assistant, built into Word, Excel, PowerPoint, Outlook and Teams at many workplaces.
CSV: A simple spreadsheet file format (comma-separated values): plain text with one row per line and commas between the values. Every spreadsheet program and AI assistant can read it.
Custom instructions: A few lines in Settings about you and how you like replies, applied to every chat so you don't repeat the background each time.

D

Deep research: A mode in which the AI makes a research plan, reads many web pages over several minutes and writes a report with citations. Treat it as a good starting point, not the final word.

E

Edit (a prompt): Changing a message you already sent so the chat continues from that point, as if the old version had never been sent. It is tidier than piling corrections onto a muddled request.
Example: The part of the prompt recipe where you show what good looks like, such as a past product description. The AI copies length and style from it more reliably than from a description.

F

Fast mode: A model setting that replies almost instantly. It suits everyday writing, rewording and quick questions.
Five-part recipe: A checklist for writing a clear prompt: goal, context, example, format and tone. When a reply disappoints, the missing part is usually the reason.
Follow-up: A short message replying to the AI's answer to change or improve it, such as "shorter and warmer". Often quicker than starting over.
Format: The part of the prompt recipe that says what shape the answer should take: length, bullet points, a table or headings.
Formula: A calculation typed into a spreadsheet cell, always starting with =. For example =B2*C2 multiplies two cells, and =SUM(D2:D50) adds a range.

G

Gemini: Google's AI assistant, which also appears inside Gmail, Docs, Sheets, Slides and Meet.
Gems: Gemini's closest feature to Projects: custom versions of Gemini with your own instructions.
Generate (images): Having the AI create a picture from a written description. Use generated images for mood and illustration, never to show what a customer will really receive.
Goal: The part of the prompt recipe that says what you want made or done, such as "Write a reply to this review."
Google Workspace: Google's set of work apps: Gmail, Docs, Sheets, Slides and Meet, where Gemini is built in.

H

Hallucination: When the AI makes something up, such as a statistic, quote, law or book that doesn't exist, and states it as confidently as a true fact.

K

Knowledge cutoff: The date after which the AI learned nothing. It doesn't know later events unless it searches the web.

L

LLM (large language model): The program behind an assistant, trained on an enormous amount of text so it learned the patterns of language. It writes replies one small piece at a time, choosing a likely next word.

M

Memory: A feature in some assistants that keeps a few notes about you between chats. You can usually view, edit and delete what it remembers in Settings.
Microsoft 365: Microsoft's work apps: Word, Excel, PowerPoint, Outlook and Teams, where Copilot is built in.
Model: One version of the AI behind an assistant. Many apps let you choose, usually between fast, thinking and auto.

P

Placeholder: A stand-in such as "Customer A" or [order number] used in place of real details. In prompts, brackets mark what you replace with your own text.
Pour-over: A way of making coffee by pouring hot water slowly over ground coffee in a filter.
Profit margin: Profit divided by revenue (or price), showing how much of each sale you keep. The AI can calculate it from your costs, but check one figure yourself.
Project: A folder of related chats with its own instructions and uploaded files that every chat in it can use. ChatGPT and Claude call these Projects.
Prompt: The message you type to the assistant. The assistant's answer is its reply or response.
Prompt library: Your own saved collection of prompts that work, grouped by job, with the parts to fill in marked in brackets, so you can reuse them.

R

Range: A block of spreadsheet cells written with a colon, such as D2:D50, meaning everything from D2 to D50.
Reply (response): The assistant's answer to your prompt.
Research brief: A short note giving a research question, why you need it, its scope, preferred sources and the output you want, like a note to a colleague doing research for you.
Role: A part you give the AI at the start of a prompt, such as "You are an experienced customer service manager", to shape its point of view. It doesn't add knowledge the AI lacks.
Role-play: Acting out a situation with the AI playing the other person, so you can practice a hard conversation, such as asking a supplier for a discount, before the real one.

S

Screenshot: A picture of what's on your screen. You can attach one and ask the AI to explain an error message or a chart.
Spreadsheet: A grid of cells for numbers and calculations, made in a program such as Excel or Google Sheets.
Spreadsheet functions (SUM, IF, AND): SUM adds up numbers, such as =SUM(D2:D200). IF gives one result if a condition is true and another if not (for example free shipping when C2 is 50 or more), and AND is true only when every condition inside it is true.
Stop: The button that halts a reply while the AI is still writing, useful when it's heading the wrong way. You can then edit your prompt.

T

Template: A reusable piece of writing with gaps to fill in, such as a reply to "where's my order" with brackets for the order number and tracking link.
Thinking mode: A model setting where the AI works through a problem step by step before answering, sometimes for a minute or more. Good for planning, analysis and math.
Thread: Another name for a chat or conversation.
Tone: The part of the prompt recipe that says how the answer should sound, such as friendly, formal, plain or playful.
Traffic-light rule: A way to sort text before pasting it into an AI tool: green (public information) is fine anywhere, yellow (internal business information) belongs only in an approved assistant, and red (personal or highly sensitive information) stays out unless your workplace rules allow it.
Training setting: A switch in an AI assistant's settings that controls whether your conversations may be used to train future AI models. It is worth checking, and usually turning off, before your first real chat on a personal account.
Transcript: A written record of everything said in a meeting or call. Paste it into the assistant to get decisions and actions, but check names and numbers because transcripts mishear them.
Try again (regenerate): Asking for a fresh reply to the same prompt. Because the AI chooses words with a little randomness, each attempt comes out different.

U

Usage limits: Caps on how much you can use an assistant on your plan, such as messages per few hours or file uploads. When you hit one, wait, switch assistants or upgrade.

V

Voice mode: Talking to an assistant out loud, by tapping the microphone or headphones icon, instead of typing.

W

Web search: When the assistant looks up pages online before answering, so its reply is more current and shows sources. Without it, it answers from what it learned in training.
Weight: How important a criterion is when scoring options, for example 1 to 3. Multiplying each 1-to-5 score by its weight and totaling gives a scored decision table.