AI Glossary: 40 AI Terms Explained Simply

Updated: October 2026Reading time: 9 min

AI writing is full of jargon that gets thrown around without explanation. This glossary defines the 40 terms you'll actually run into when choosing, evaluating, or talking about AI tools — grouped by topic, in plain English, with a link to a real tool wherever it helps to see the concept in action.

Models & Training Agents & Automation Generation Business & Tooling

Models & Training

LLM (Large Language Model): A model trained on huge amounts of text to predict and generate language. ChatGPT and Claude are both built on LLMs.
Parameters: The internal numeric values a model adjusts during training. Roughly, more parameters means more capacity to learn patterns — but not always better real-world performance.
Training data: The text, images, or other content a model learns from before it's released. What's in the training data shapes what the model knows and how it writes.
Fine-tuning: Taking an already-trained model and training it further on a narrower dataset, so it specializes in a task or tone without starting from scratch.
Context window: How much text a model can "see" at once — your prompt, chat history, and any documents combined. A bigger window means it can reason over longer documents or conversations.
Token: A chunk of text (roughly 4 characters or three-quarters of a word) that models process one at a time. API pricing and context windows are usually measured in tokens, not words.
Temperature: A setting that controls how random or predictable a model's output is. Low temperature gives safer, more repetitive answers; high temperature gives more varied, sometimes stranger ones.
Hallucination: When a model states something false or made-up with full confidence. The main reason to fact-check AI output before publishing or acting on it.
Multimodal: A model that can work with more than one type of input or output — text, images, audio, or video — instead of just text.
Reasoning model: A model trained to work through problems in explicit steps before answering, trading speed for accuracy on math, logic, and multi-step tasks.

Agents & Automation

AI agent: Software that uses an AI model to plan and carry out multi-step tasks on its own — not just answering one question, but deciding what to do next. See our Getting Started with AI Agents guide.
Autonomous agent: An agent that keeps working toward a goal with minimal human input between steps, like AutoGPT.
Tool use / function calling: A model's ability to call an external tool or API — search the web, run code, query a database — as part of answering a prompt, instead of relying only on what it already knows.
Orchestration: Coordinating multiple AI calls, tools, or agents into one workflow, so the output of one step feeds the next automatically.
Workflow automation: Connecting apps and triggers so a task runs without manual input — what tools like Make.com and Zapier do, increasingly with AI steps built in.
No-code / low-code: Building a tool or automation through a visual interface instead of writing code — the model behind Make.com, n8n, and similar platforms.
Webhook: A URL that one app calls automatically when something happens in another app, commonly the trigger that starts an automation.
API (Application Programming Interface): The way one piece of software talks to another. "Using the ChatGPT API" means calling OpenAI's model programmatically, instead of through the chat website.
MCP (Model Context Protocol): An open standard that lets an AI assistant connect to external tools and data sources (like a CRM or an analytics platform) in a consistent way, instead of needing custom code for each one.
Human in the loop: A workflow design where a person reviews or approves an AI agent's action before it takes effect — the opposite of fully autonomous.

Content Generation

Prompt: The instruction or question you give a model. How you phrase it directly shapes the quality of what comes back — see prompt engineering.
Prompt engineering: Deliberately structuring a prompt — with examples, constraints, or role instructions — to get a more reliable or specific result from a model.
Diffusion model: The technique behind most AI image generators, including Midjourney and Stable Diffusion — it starts from noise and gradually refines it into an image matching the prompt.
Text-to-image: Generating an image directly from a written description, the core function of tools like Midjourney, DALL-E 3, and Leonardo AI.
Text-to-speech (TTS): Converting written text into spoken audio. ElevenLabs is a leading TTS tool for natural-sounding voice.
Voice cloning: Training a model to reproduce a specific person's voice from a short audio sample, so it can speak new text in that voice.
Upscaling: Using AI to increase an image or video's resolution while adding plausible detail, rather than just stretching the pixels.
In-painting / out-painting: Editing part of an AI-generated image (in-painting) or extending it beyond its original borders (out-painting) using the same model.
Style transfer: Applying the visual style of one image or reference to the content of another — part of how many image generators handle "in the style of" prompts.
Watermarking / provenance: Embedding an invisible or visible marker in AI-generated content so it can later be identified as machine-made — increasingly required by platforms and regulation.

Business & Tooling

SaaS (Software as a Service): Software you pay for and access online rather than install — the delivery model behind nearly every tool in our pricing comparison.
Freemium: A pricing model with a genuinely usable free tier, with paid plans unlocking higher limits or advanced features — the most common model among the tools we cover.
Seat-based pricing: Charging per user per month, common in team plans for tools like GitHub Copilot ($19/user/month) versus flat per-account pricing.
Rate limit: A cap on how many requests or messages you can send a model or API in a given period, used to manage cost and server load.
Open source (AI): A model or tool whose code (and sometimes weights) is publicly available to inspect, modify, and self-host — like AutoGPT or Flowise — versus closed, proprietary systems.
Self-hosted: Running software on your own servers instead of a vendor's cloud, trading convenience for control and data privacy — an option for tools like n8n.
SEO (Search Engine Optimization): Shaping content and a site's technical setup so it ranks higher in search results — what tools like Semrush are built to measure and improve.
Keyword difficulty: A score estimating how hard it would be to rank for a given search term, based on the strength of pages already ranking for it.
Affiliate link: A link that credits the referring site when someone signs up or buys through it. CloudAtelier discloses these on every page where they appear — see our privacy policy.
Vendor lock-in: The cost or difficulty of switching away from a tool once your data, workflows, or team are built around it — worth weighing before committing to a platform.

Didn't find a term?

This glossary covers the terms that come up most in our comparisons and guides. If there's one you'd like explained, check our full list of articles — most of these concepts are covered in more depth in a dedicated comparison or guide.