AI glossary for beginners: ten terms in plain English

Large Language Model (LLM)

An AI system trained on huge amounts of text to understand requests and write human-like answers. ChatGPT, Claude and Gemini all run on LLMs.

A large language model is the engine behind AI chat tools. It learned from an enormous amount of written text and, at its core, does one thing: it predicts which words are most likely to come next. That simple skill is enough to write, summarize, translate and hold a conversation. On its own, though, it doesn’t look anything up — it works from patterns it absorbed during training, which is why it can sound sure and still be wrong.

Prompt

Whatever you type to an AI — a question, a task or instructions. The clearer and more specific it is, the better the answer.

A prompt is your message to the AI. It can be one short question or a detailed brief. The model only knows what the prompt tells it, so missing details — who it’s for, how long, what tone — get filled in with guesses. That’s why most disappointing answers come from a vague prompt, not a weak model.

In AskClear: You don’t have to write the perfect prompt: you type a rough request, answer a few short questions, and see the improved prompt next to your original before anything is sent.

Token

The small chunk of text — a word or part of a word — that AI models read and write in. Length limits and usage are counted in tokens.

AI models don’t read letter by letter or word by word; they split text into tokens. A token is often a short word, a piece of a longer word or a punctuation mark. As a rule of thumb, 100 tokens are about 75 English words; languages such as German or Russian usually need more tokens for the same text. Both a model’s limits and its running costs are measured in tokens.

In AskClear: You never have to count tokens: your plan usage is shown as a percentage, and any extra credit as an amount of money.

Context Window

How much text an AI can keep in view at once — your messages, any documents and its own reply together. It’s measured in tokens.

The context window is the model’s short-term working space. Everything it considers for the next reply has to fit inside it: the instructions, the conversation so far, any documents you added and the answer it is writing. When a conversation or file is too long to fit, something has to be left out — that’s why a long chat can seem to “forget” what you said at the start.

In AskClear: If the sources you add in Facts mode are too long to fit, the passages that best match your question are kept, instead of the document simply being cut off at the end.

Hallucination

When an AI states something false or made up — a fake figure, quote or source — in the same confident tone as a real fact.

A hallucination is an answer that sounds fluent and certain but isn’t true. The model isn’t lying on purpose: it produces text that sounds plausible, and sometimes the plausible-sounding thing is wrong. Invented statistics, quotes nobody said, books that don’t exist and totals that don’t add up are all typical. Nothing in the tone warns you, so anything important is worth checking.

In AskClear: Facts mode is designed to make this visible: quotes are checked against your sources and marked “Unverified” if they can’t be found there, and the model has an exact calculator available for arithmetic instead of estimating.

Grounding

Making an AI answer from specific sources you provide — your document, pasted text or web results — instead of from memory.

Grounding means giving the model the actual material and asking it to answer from that material only. Without it, the model answers from a blurry memory of everything it once read. With it, the answer is tied to text you can see and check. And if the answer isn’t in the sources, a well-grounded assistant should say so rather than guess.

In AskClear: This is what “From my docs” (Facts mode) does: you attach a file or paste text, and the answer is based on those sources. If they don’t have the answer, the AI can also search the web.

Retrieval-Augmented Generation (RAG)

A technique where the system first finds the passages that match your question, then hands them to the AI to answer from.

Retrieval-augmented generation, or RAG, works in two steps. First, retrieval: the system searches a set of sources — documents, a company knowledge base or the web — for the parts that relate to your question. Then, generation: those passages are added to the prompt, and the model writes its answer from them. RAG is how an AI tool can answer questions about your own files, or about recent news it was never trained on. It is one common way to achieve grounding.

Citation

A reference that shows exactly where in a source a statement comes from, usually with the quoted words, so you can check it yourself.

A citation links a statement in an answer to the place it came from — a document, a pasted text or a web page — and ideally quotes the exact words. It turns “trust me” into “see for yourself”. But a citation is only as good as its check: a chatbot can produce a quote that looks real and yet isn’t in the source at all.

In AskClear: In Facts mode, every quote is matched against your sources: if the exact words are found, it is shown as verified; if not, it is clearly marked “Unverified”.

System Prompt

Hidden background instructions that tell an AI its role, tone and rules before your conversation starts.

The system prompt is a set of instructions the model receives before you type anything. You usually don’t see it, but it shapes every reply: what role the assistant plays, how formal it is, what it should avoid and how it should handle uncertainty. Your own message — the prompt — says what you want right now; the system prompt sets the ground rules for the whole chat.

AI Mediator

A step between you and the AI that clarifies your rough request and rewrites it into a clear prompt before the AI answers.

An AI mediator sits between you and the language model. Instead of passing your first, half-formed request straight through, it works out what’s missing — by asking you a few short questions — and then rewrites the request into a clear, specific prompt. You see the result before it is sent, so nothing happens behind your back. You get a useful answer without having to learn how to write prompts.

In AskClear: The whole product is built around this: up to three short questions per round, at most three rounds, and you can skip any question or jump straight to the result.