Glossary · 35 terms
AI Glossary for Alberta Business Owners
The AI words you will actually run into, in plain English, each with one Alberta example. The examples are illustrations of how the term shows up in a business like yours; they are not real companies. Where there is a real Alberta figure or a Canadian rule behind a term, it is shown with its source.
Reviewed by Haruun Ali, President & Managing Partner, AltaMaticLast updated
- AI agent
- AI assistant
- AI Overview
- API
- Artificial intelligence (AI)
- Automation
- Bias
- Context
- Context window
- Custom assistant
- Data analytics with AI
- Data residency
- Deepfake
- Fine-tuning
- Generative AI
- Grounding
- Hallucination
- Human in the loop
- Instructor-led training
- Knowledge cutoff
- Large language model (LLM)
- Machine learning
- Model
- Multimodal
- Natural language processing (NLP)
- Personal information
- Prompt
- Prompt injection
- Retrieval-augmented generation (RAG)
- Speech recognition
- Token
- Training data
- Training on your data
- Voice agent
- Voice cloning
AI agent
Also called: agentic AI
An AI tool that carries out a task in several steps and uses other software to do it, such as reading an email, checking a calendar and booking the appointment. An assistant answers; an agent acts.
Alberta example: An agent for a landscaper in Sherwood Park reads a new web inquiry, checks the crew schedule and replies with two open times, then waits for the owner to approve the booking.
Free lesson: Chain the Tools: Multi-Step AI WorkflowsRelated: Automation, Human in the loop, Prompt injection
AI assistant
Also called: chatbot, virtual agent
A program you talk to in plain language, by typing or speaking, that answers with the help of a language model. It can be a general one you open in a browser or one built into software you already use.
Alberta example: A property manager in Lethbridge keeps an assistant open through the day to draft tenant notices and tidy up maintenance notes.
In Alberta: 26.7% of businesses that use AI report using virtual agents or chatbots, by Statistics Canada's estimate for spring 2026 (quality C, good). See the table
Free lesson: Which AI Tool Should You Actually UseRelated: Custom assistant, AI agent
AI Overview
Also called: AI answer, answer engine
A search result in which AI writes out the answer at the top of the page, with links to where it found it. A customer can read it and never click through to a website.
Alberta example: Someone searching for an emergency plumber in Lloydminster reads a short summary that names two businesses, and calls one of them.
How to show up when customers ask an assistantRelated: Grounding
API
Also called: application programming interface
A way for one piece of software to use another directly, with no person in a chat window. It is how AI gets built into a booking system, a CRM or a website.
Alberta example: A developer in Calgary connects a towing company’s dispatch system to an AI model through its API, so every call note is summarized automatically.
Related: Automation, Token
Artificial intelligence (AI)
Software that does work we used to think needed a person’s judgment: reading, writing, recognizing what is in a picture, predicting what comes next. In a small business today it mostly means tools that read and write text for you.
Alberta example: A plumbing company in Edmonton has AI turn a technician’s spoken notes from the truck into a written quote the owner then checks.
Free lesson: What AI Actually Is (In Plain English)How many Alberta businesses use AIRelated: Generative AI, Machine learning
Automation
Also called: workflow
Software that runs a fixed series of steps when something happens, with nobody clicking. AI can be one of the steps, but an automation follows the path you set; it does not decide the path.
Alberta example: When a quote request arrives on a contractor’s website in Leduc, a workflow saves it to a spreadsheet, drafts a reply with AI and texts the owner.
Free lesson: Connect Your Tools Without CodeRelated: AI agent, API
Bias
When an AI system’s output is skewed against some people, or toward some answers, because of patterns in the data it learned from.
Alberta example: A tool screening applicants for a fabrication shop in Fort Saskatchewan could rank down a strong welder whose résumé reads differently from the ones it learned on. A person should make the hiring decision.
Free lesson: Screen Applicants in Half the TimeRelated: Training data, Human in the loop
Context
The background you give with a request: who the customer is, what the job was, your prices, the tone you want. Adding the right context improves most answers more than anything else you can do.
Alberta example: A welding shop in Fort McMurray gets a usable quote draft only once the prompt includes the material, the shop rate and the travel time to site.
Free lesson: Giving Good Context — The Biggest LeverRelated: Prompt, Context window
Context window
How much text a model can keep in view at once, counting your request, any documents you paste and its own replies. Go past it and the earliest material drops out of view.
Alberta example: After an hour-long chat about a bid, an estimator in Red Deer notices the assistant has forgotten the site address given at the start. A fresh chat with a short summary fixes it.
Custom assistant
Also called: custom GPT, project, gem
An assistant you set up once with standing instructions and reference files, so you do not have to explain your business every time. Each vendor has its own name for the feature.
Alberta example: A realtor in Medicine Hat builds one that knows her listing style, the neighbourhoods she covers and the disclaimers her brokerage requires.
Free lesson: Set Up Your Assistant the Right WayRelated: Retrieval-augmented generation (RAG), AI agent
Data analytics with AI
Using AI to find what a set of numbers is saying: which jobs made money, which months run short, which customers are slipping away.
Alberta example: A restaurant owner in Banff uploads a year of sales exports and asks which menu items sell well and earn little.
In Alberta: 33.1% of businesses that use AI report using AI for data analytics, by Statistics Canada's estimate for spring 2026 (quality C, good). See the table
Free lesson: Understand Your Numbers in Plain EnglishRelated: Machine learning
Data residency
Where your data is physically stored and processed. It matters when a law, a contract or a regulator expects certain records to stay in Canada.
Alberta example: A contractor in Edmonton who keeps the only copies of his receipts in an app hosted outside Canada asks his accountant whether that meets the tax rules.
The Canada Revenue Agency says business records must be kept in Canada unless it gives written permission, and that records kept outside Canada and accessed electronically from Canada are not considered to be records kept in Canada.
Source: Canada Revenue Agency, Where to keep your records, for how long and how to request the permission to destroy them early. Read .
Related: Personal information
Deepfake
A fake image, audio clip or video of a real person, made with AI and good enough to pass as real at a glance.
Alberta example: A video appears to show a well-known Alberta business owner endorsing an investment they have never heard of.
The Canadian Centre for Cyber Security advises that audio or video alone should not be considered proof of identity, especially for sensitive or unusual requests.
Source: Canadian Centre for Cyber Security, Social engineering (ITSAP.00.166). Read .
Related: Voice cloning
Fine-tuning
Further training of an existing model on your own examples, so it leans toward your style or your task. Most small businesses never need it: clear instructions and a few good examples in the prompt do the same job.
Alberta example: An agency in Calgary with thousands of approved ad lines might fine-tune a model on them. A three-person shop gets a similar result by pasting in five lines it likes.
Related: Custom assistant, Training data
Generative AI
Also called: gen AI
AI that produces something new from a request: text, images, audio or video. Older AI mostly sorted or scored things; generative AI writes and draws.
Alberta example: A café in Canmore asks for three versions of a patio-season post, picks the one that sounds like the owner and edits the price.
Generative AI use by Alberta workersRelated: Large language model (LLM), Prompt
Grounding
Also called: web search, browsing
When a tool looks something up, on the web or in files you gave it, and bases its answer on what it found, usually with links. A grounded answer can be checked; an ungrounded one has to be taken on trust.
Alberta example: An owner in Leduc asks, “What does alberta.ca say today about who can be trained under the Productivity Grant? Give me the link.” The link is what makes the answer usable.
Free lesson: Size It Up: Research a Competitor, Supplier or Price in MinutesRelated: Hallucination, Retrieval-augmented generation (RAG)
Hallucination
When an AI tool states something false as if it were fact: an invented price, a regulation that does not exist, a source it made up. The model is predicting likely words. Unless it is connected to a search tool or your documents, it is not looking anything up.
Alberta example: Ask for the permit fee for a deck in Airdrie and you may get a confident number that is wrong. The city’s own fee schedule is the answer.
Free lesson: Spotting When AI Is WrongRelated: Grounding, Human in the loop
Human in the loop
A person checks the AI’s work before it goes out or takes effect. It is the simplest safeguard there is, and the right default for anything a customer will read.
Alberta example: A dental office in St. Albert lets AI draft recall reminders, and the office manager reads each batch before it is sent.
Related: Hallucination, AI agent
Instructor-led training
Training with a qualified instructor the learner can interact with, as opposed to a course you work through alone at your own pace. In Alberta the distinction decides whether a training grant can apply.
Alberta example: A two-day AI workshop taught live by an outside provider is instructor-led. A self-paced online course, this school included, is not.
The Canada-Alberta Productivity Grant funds instructor-led training from a third-party provider, and says eligible training must include interaction with a qualified instructor.
Source: Government of Alberta, Canada-Alberta Productivity Grant. Read .
Knowledge cutoff
The date a model’s training data ends. It does not know what happened after that date unless it can search the web or you tell it.
Alberta example: Alberta changed the terms of its training grant on October 8, 2026. A model trained before then may still describe the old terms with complete confidence.
Large language model (LLM)
Also called: language model
The kind of AI behind assistants such as ChatGPT, Claude and Gemini. It learned from a very large amount of text, and it writes by predicting which words are likely to come next. That is why it is fluent, and also why it can be wrong.
Alberta example: A bookkeeper in Calgary pastes in a long letter from the tax office and asks for a plain-English summary. The model does the reading and the rewriting; she checks the dates and amounts.
In Alberta: 29.1% of businesses that use AI report using large language models, by Statistics Canada's estimate for spring 2026 (quality C, good). See the table
Free lesson: What AI Actually Is (In Plain English)Related: Hallucination, Training data, Model
Machine learning
The method behind modern AI. Instead of being given rules, the software is shown many examples and finds the patterns in them.
Alberta example: A feed mill near Lacombe uses software that has learned from years of orders to forecast how much of each product to make next week.
In Alberta: 20.8% of businesses that use AI report using machine learning, by Statistics Canada's estimate for spring 2026 (quality C, good). See the table
Related: Artificial intelligence (AI), Training data
Model
The trained program doing the work inside an AI tool. Tools come in several models and versions. A larger or newer model is usually more capable, and often slower or more costly to run.
Alberta example: Two people at a dealership in Grande Prairie get different answers to the same question because one is using the free plan’s model and the other a paid one.
Free lesson: Free vs Paid: What's Worth ItRelated: Large language model (LLM), Knowledge cutoff
Multimodal
Able to take in or produce more than text: photos, voice, video and files as well.
Alberta example: A roofer in Okotoks photographs hail damage and asks the assistant to describe it for the estimate, then corrects the description on site.
Free lesson: Snap a Photo, Get the Work DoneRelated: Speech recognition
Natural language processing (NLP)
The branch of AI that deals with human language: understanding it, summarizing it, translating it and answering in it. Language models are its best-known result.
Alberta example: An insurance brokerage in Lloydminster has software read incoming emails and sort them into claims, renewals and new quotes.
In Alberta: 25.1% of businesses that use AI report using natural language processing, by Statistics Canada's estimate for spring 2026 (quality C, good). See the table
Related: Large language model (LLM)
Personal information
Information about an identifiable individual, such as a customer’s name together with their address, phone number or health details. The business that collected it stays responsible for it, including when it is pasted into an AI tool.
Alberta example: Before asking AI to help answer a complaint, a clinic in Edmonton removes the patient’s name and anything about their treatment.
Alberta has its own private-sector privacy law, the Personal Information Protection Act (PIPA). The federal law, PIPEDA, applies when personal information crosses provincial or national borders in the course of commercial activity.
Source: Office of the Privacy Commissioner of Canada, PIPEDA requirements in brief. Read .
Free lesson: What to Never Paste Into AIRelated: Training on your data, Data residency
Prompt
What you type or say to an AI tool: the request itself, plus whatever background you give it. A clear prompt names the task, the audience, the details and the format you want back.
Alberta example: “Write a text to a customer confirming her furnace tune-up Thursday between 8 and 10. Friendly, under 40 words, signed with my first name” is a complete prompt.
Free lesson: The Prompt Recipe: Role + Task + Context + FormatRelated: Context, Context window
Prompt injection
Instructions hidden in a web page, an email or a file that try to make an AI tool do something its user never asked for. It matters most for agents, because an agent can act on what it reads.
Alberta example: An AI inbox assistant at a firm in Calgary opens an email containing hidden text that says to forward recent invoices to an outside address. A well-built tool treats that as content, not as an order.
Related: AI agent
Retrieval-augmented generation (RAG)
A setup in which the tool first finds the relevant passages in your own documents and then writes its answer from them. It is how an assistant can answer from your price list or your policy manual instead of from general knowledge.
Alberta example: A heating company in Red Deer loads its service manuals so a technician can ask how to reset a particular furnace and be shown the page the answer came from.
Free lesson: Build Your Own AI Assistant That Knows Your BusinessRelated: Grounding, Custom assistant
Speech recognition
Also called: speech-to-text, transcription, dictation
Turning spoken words into written text. It is what lets you dictate a note instead of typing it, and what lets a tool write up a recorded call.
Alberta example: An oilfield service supervisor near Drayton Valley dictates the day’s ticket notes on the drive back instead of typing them at night.
In Alberta: 26.6% of businesses that use AI report using speech or voice recognition, by Statistics Canada's estimate for spring 2026 (quality C, good). See the table
Free lesson: Talk, Don't Type: Turn a Voice Note Into a Clean QuoteRelated: Voice agent, Multimodal
Token
The unit a model reads and writes in. A token can be a character, part of a word, a whole word or a punctuation mark. Context windows, usage limits and prices for AI tools are counted in tokens, not words.
Alberta example: A software firm in Calgary building a quoting tool for a client estimates its monthly AI cost from the number of tokens each quote uses.
OpenAI notes that a token count is not the same as a word count, and that the same text can produce different counts depending on the model and the language.
Source: OpenAI, Understanding and counting tokens. Read .
Related: Context window, API
Training data
The text, images or audio a model learned from before you ever used it. What it learned from shapes what it knows well and what it gets wrong.
Alberta example: A model has read far more about Canada as a whole than about any one Alberta town, so its answer about a local bylaw deserves more checking than its answer about a general topic.
Related: Knowledge cutoff, Bias, Training on your data
Training on your data
Also called: model training opt-out
Some AI tools use what you type to improve their models unless you switch that off. Plans sold to businesses usually do not. Check the setting and the terms before anything confidential goes in.
Alberta example: An accounting firm in Calgary moves its staff to a business plan before any client file goes near an AI tool.
OpenAI, for example, says it may use content from its services for individuals to train its models unless the user opts out, and that by default it does not train on inputs or outputs from its business products.
Source: OpenAI, How your data is used to improve model performance. Read .
Free lesson: What to Never Paste Into AIIs a free AI plan safe for client files?Related: Personal information
Voice agent
Also called: AI receptionist, AI phone answering
An AI system that answers the phone, speaks with the caller and takes a message or books a job. It needs rules for emergencies, for the area you serve and for handing the call to a person.
Alberta example: A furnace company in Spruce Grove uses one after hours, set to transfer straight to the on-call technician the moment a caller says “no heat”.
Should I let AI answer my phone?Related: AI agent, Speech recognition
Voice cloning
Using AI to copy a person’s voice from a recording, so that new words can be spoken in it.
Alberta example: A bookkeeper in Camrose gets a call that sounds like the owner asking for an urgent e-transfer. The rule that stops it: hang up and call back on the number already on file.
The Canadian Centre for Cyber Security says the technique replicates a person’s voice using small audio samples, and that any request involving a sensitive action should be confirmed through a second, independent form of communication.
Source: Canadian Centre for Cyber Security, Social engineering (ITSAP.00.166). Read .
Adapted from Statistics Canada, Table 33-10-1167-01, Use of artificial intelligence by businesses and organizations in producing goods or delivering services over the last 12 months, second quarter of 2026, released May 27, 2026. This does not constitute an endorsement by Statistics Canada of this product.
Related: answers to questions Alberta owners ask, free AI courses compared, Alberta AI statistics and the funding guide.