AI
Software that performs tasks such as recognising patterns, generating text or interpreting images. It is not a guarantee of understanding or accuracy.
For example: A tool suggests a first draft of a customer email.
← AI Tool Finder
AI basics / No jargon barrier
A little clarity goes a long way.
The words behind the tools, explained in plain English. Start here, then choose a small, low-risk job to try.
Editorial guide · Nov 7, 2026
18 of 18 terms · Search stays in your browser
Generative AI
AI that produces new text, images, audio or other content from instructions and learned patterns.
For example: You describe a poster and the tool proposes a design.
LLM
A large language model: a system trained on language patterns that can generate and transform text. Its confident wording is not proof that an answer is true.
For example: An assistant turns rough notes into a readable draft.
Prompt
The instructions and context you give an AI tool. Good prompts state the job, supplied facts, audience, boundaries and desired format.
For example: “Rewrite this in plain Canadian English; do not add new facts.”
Token
A chunk of text or other information a model processes. It is not the same as a word; token limits and billing vary by model.
For example: A long document uses more of the available context than a short question.
Context window
The amount of information a model can consider at once. A large limit does not guarantee it will notice every detail.
For example: For a lengthy report, specify the sections that matter and check that nothing was omitted.
Hallucination
An AI output that is invented or incorrect but presented plausibly. Verify important claims against original sources.
For example: A convincing citation points to an article that does not exist.
Agent
An AI system that can use tools and carry out several steps toward a goal. More permissions mean more need for boundaries and human approval.
For example: An assistant drafts a reply and prepares—but does not send—a follow-up.
Automation
A process that runs predefined actions when conditions are met. It may use AI, but does not have to.
For example: A submitted form creates an internal notification.
RAG
Retrieval-augmented generation: finding relevant material first, then giving it to a model to help answer a question. It improves access to sources but does not eliminate mistakes.
For example: A support assistant retrieves an approved policy before drafting a reply.
Grounding
Connecting an answer to specific supplied or retrieved evidence rather than relying only on a model’s learned patterns.
For example: Ask for the exact source paragraph supporting each policy claim.
Fine-tuning
Additional training that adapts a model to particular examples or tasks. It differs from simply attaching a document to a conversation.
For example: A governed training process teaches a consistent response format.
Multimodal
Able to work with more than one kind of input or output, such as text, images and audio. Supported combinations depend on the product.
For example: An assistant explains a chart from an uploaded image.
API
A defined way for one program to request information or actions from another. API pricing and permissions may differ from a normal app subscription.
For example: An approved integration asks a service to create a draft record.
Data residency
Where specified data is stored or processed under a particular contract. Canadian storage is not necessarily Canadian processing, and neither alone guarantees legal compliance.
For example: A plan stores customer content in Canada but uses an overseas service for a connected feature.
Human in the loop
A person reviews or authorises a consequential step rather than leaving the whole process to software.
For example: A manager approves an invoice before it is sent.
Prompt injection
Instructions hidden in a message or document that try to redirect an AI system away from its intended rules. Treat external content as data, not authority.
For example: A document tells an assistant to ignore its rules and reveal customer records.
Model vs product
The model is the underlying AI system; the product adds the interface, tools, limits and terms. A model announcement is not necessarily a feature available on your subscription.
For example: Two apps may offer related models but different privacy settings and usage limits.