AI terms and acronyms, made simple

The world of AI is full of jargon that makes simple ideas sound complicated. Most of it isn’t that hard once someone explains it plainly. This is your cheat sheet — the terms you’ll actually run into, in everyday language, with a quick example for each. Skim it, bookmark it, come back whenever a word trips you up.

The big ones

Start here. These five come up constantly.

AI (Artificial Intelligence)

A catch-all term for software that does things we used to think needed a human — writing, answering questions, recognizing images. When people say “AI” today, they usually mean the chatbot kind.

LLM (Large Language Model)

The engine behind chatbots like ChatGPT, Claude, and Gemini. It’s a system trained on enormous amounts of text that’s very good at predicting what words should come next — which, it turns out, lets it hold a conversation, write, and explain things.

Prompt

The instruction or question you type in. Better prompt, better answer. (We wrote a whole guide on writing good prompts.)

GPT (Generative Pre-trained Transformer)

The name of the technology family behind ChatGPT. You don’t need the full mouthful — just know “GPT” refers to this kind of text-generating model. A “Custom GPT” is a version you set up with your own instructions.

Generative AI

AI that creates new things — text, images, audio, code — rather than just sorting or labeling existing data. Chatbots and image tools like Midjourney are generative AI.

How it works under the hood

Token

The little chunks of text an AI reads and writes — roughly ¾ of a word each. AIs measure length in tokens, and paid tools often charge by them. “This costs $0.01 per 1,000 tokens” just means per ~750 words.

Training

The process of teaching a model by showing it huge amounts of data. It’s done ahead of time — when you chat with an AI, it’s not learning from you in real time; it’s using what it already learned.

Model

A single trained AI “brain.” Companies release different models (bigger, smaller, faster, smarter). “Which model are you using?” just means which version.

Parameters

The internal dials a model adjusts during training. More parameters can mean a more capable model — it’s the number behind phrases like “70-billion-parameter model.” You’ll rarely need this one, but now it won’t throw you.

A few more model words

Multimodal

An AI that handles more than text — images, audio, even video. A multimodal model can look at a photo you upload and describe it, or read a chart. Most leading assistants are now multimodal.

Prompt engineering

The skill of writing instructions that get the best out of an AI. It sounds technical, but it’s mostly clarity and structure — exactly what our prompt guide covers.

Temperature

A setting that controls how creative versus predictable answers are. Low temperature = focused and consistent; high = more varied and surprising. Some tools let you adjust it.

Zero-shot & few-shot

How many examples you include in your prompt. “Zero-shot” means you just ask; “few-shot” means you show a couple of examples of what you want first — which usually improves the result.

Chain-of-thought & reasoning models

When an AI works through a problem step by step instead of blurting an answer. “Reasoning models” are built to do this automatically, which helps with math, logic, and multi-step tasks.

Embeddings

A way of turning words or documents into numbers that capture their meaning, so software can tell which pieces of text are related. It’s the quiet machinery behind search and “chat with your files” tools.

Vector database

A database built to store those embeddings, so an AI can quickly find the most relevant pieces of your information to answer a question. Often paired with RAG.

Diffusion model

The technology behind most AI image generators (like Midjourney or DALL·E). It builds a picture by starting from random noise and gradually refining it into what you described.

Inference

The moment an AI actually produces an answer for you (as opposed to being trained). “Inference cost” just means the cost of running the model to respond.

The foundations

Machine learning (ML)

The broad field behind modern AI: instead of being programmed with fixed rules, the software learns patterns from examples. Everything else here is built on it.

Deep learning

A powerful kind of machine learning that uses layered “neural networks.” It’s what made today’s leap in AI possible.

Neural network

A computing structure loosely inspired by the brain, made of layers of connected “nodes.” It’s the basic building block of deep learning and LLMs.

NLP (Natural Language Processing)

The area of AI focused on understanding and generating human language — the reason a chatbot can read your question and write a reply.

Computer vision

The area of AI that interprets images and video — recognizing objects, reading text in a photo, spotting faces. It’s how a multimodal AI “sees.”

Quirks & things to watch for

Hallucination

When an AI states something false with total confidence — a made-up fact, fake quote, or invented source. This is the single most important term for a beginner to know: always double-check anything factual.

Context window

How much the AI can “hold in mind” at once. Go past it in a long chat and it starts forgetting earlier details. If an AI seems to lose the thread, you’ve likely filled its context window.

Bias

AIs learn from human-made data, so they can absorb human blind spots and stereotypes. Worth remembering when an answer feels one-sided.

Cutoff (knowledge cutoff)

The date after which a model wasn’t trained on new information. Ask about something newer and it may not know — unless the tool can search the web.

Words for doing more

System prompt

Standing instructions that shape how an assistant behaves across a whole conversation — its role, tone, and rules. It’s what you set when building a Custom GPT. Our AI Prompt Builder writes these for you.

Fine-tuning

Taking an existing model and training it a bit more on your own examples so it specializes. Powerful, but usually overkill for beginners — a good system prompt gets you most of the way.

RAG (Retrieval-Augmented Generation)

A method that lets an AI look things up in your documents before answering, so it can cite your actual material instead of guessing. It’s how many “chat with your files” tools work.

Agent

An AI set up to take actions in steps — not just answer, but do (search, click, book, send). A growing area, and the reason you’ll hear “AI agents” everywhere lately.

API (Application Programming Interface)

A way for other software to plug into an AI directly. Mostly a developer thing — as a beginner you can happily ignore it until you need it.

Bigger-picture terms

AGI (Artificial General Intelligence)

A hypothetical future AI that could do essentially any intellectual task a human can, rather than being good at specific things. It doesn’t exist yet, and experts disagree on how close we are.

Frontier model

The most advanced, cutting-edge models available at any given moment, from the leading AI labs. Today’s frontier models are the biggest and most capable.

Open vs proprietary models

Open (or “open-weight”) models can be downloaded and run yourself, like Meta’s Llama. Proprietary/closed models, like GPT and Claude, are used through a company’s app or API. Open offers control; closed is often the most capable.

Distillation

Training a smaller, cheaper model to imitate a bigger one — keeping much of the quality at a fraction of the size and cost. It’s why capable small models keep appearing.

Quantization

A technique that shrinks a model by storing its numbers less precisely, so it runs faster on smaller hardware with only a small quality trade-off. Mostly relevant if you run models yourself.

AI & search (for marketers)

SEO (Search Engine Optimization)

The long-standing practice of making your website rank higher in search results like Google. It’s the reference point the newer terms below are reacting to.

AEO (Answer Engine Optimization)

Optimizing your content so it gets used and cited in AI-generated answers (ChatGPT, Perplexity, Google’s AI Overviews) — not just in the traditional list of links. The goal shifts from “rank #1” to “be the answer.”

GEO (Generative Engine Optimization)

Very close to AEO — optimizing to be included and referenced in generative AI results. The two overlap heavily and are often used interchangeably; usage is still shaking out.

AI Overviews (formerly SGE)

Google’s AI-written summaries at the top of many search results. They grew out of Google’s experimental “Search Generative Experience” (SGE) and are a big reason AEO and GEO suddenly matter.

LLM SEO

An umbrella phrase for getting your brand or content to show up in AI chatbots’ answers. Often used as a catch-all that overlaps with AEO and GEO.

Business & automation

Agentic AI

AI that doesn’t just answer but pursues a goal across multiple steps — making decisions and using tools along the way. An “AI agent” is the thing; “agentic” describes that do-it-for-you quality. (Still a buzzy, loosely used term.)

MCP (Model Context Protocol)

An open standard (introduced by Anthropic) that gives AI assistants a common way to plug into outside tools, apps, and data — think of it as a universal adapter, like USB-C for AI. It’s why assistants can increasingly connect to your files, calendar, and other software.

Copilot / AI assistant

A general term for an AI that helps you inside a task or app — drafting, summarizing, suggesting — while you stay in control. “Copilot” emphasizes that it assists rather than replaces you.

RPA (Robotic Process Automation)

Older automation that follows fixed, rule-based steps (like a macro clicking through software). It isn’t really “AI,” but it’s increasingly combined with AI to handle messier tasks.

Workflow automation

Connecting apps and steps so routine work happens automatically. AI is now being added to these workflows to handle judgment-based steps, not just rigid rules.

Safety & responsibility

Alignment

The effort to make AI systems actually do what people intend and reflect human values — reliably helpful and safe rather than harmful or off the rails.

Guardrails

The rules and limits built around an AI to keep it from doing unwanted things — like refusing harmful requests or staying on approved topics.

Prompt injection

A security trick where hidden instructions (say, buried in a web page or document) hijack an AI into doing something it shouldn’t. Increasingly important as assistants gain access to your tools.

Red-teaming

Deliberately stress-testing an AI by trying to make it misbehave, so its makers can find and fix weaknesses before release.

Grounding

Tying an AI’s answers to real, verifiable sources (your documents, live data) so it’s less likely to make things up. RAG is a common way to ground a model.

Explainability

How well we can understand why an AI gave a particular answer. Big models are hard to fully explain, which matters in high-stakes uses like lending or hiring.

Frequently asked questions