# AI terms for business owners: a glossary URL: https://goudbeek.com/en/notes/ai-glossary-for-business-owners Language: en Last updated: 2026-09-24 Provider: Goudbeek — Almere, the Netherlands — samgoudbeek2007@gmail.com > AI terms for business owners in plain language: from AI agent and RAG to prompt injection, with recognisable examples, so you ask suppliers better questions. 24 September 2026 8 min read [Sam Goudbeek](https://goudbeek.com/en/about) Glossary AI terms for business owners in plain language: from AI agent and RAG to prompt injection, with recognisable examples, so you ask suppliers better questions. AI terms are the jargon you run into as soon as you want to use AI in your business, such as AI agent, language model, RAG, token and prompt injection. This glossary explains them in plain language, with recognisable examples. Goudbeek in Almere wrote it for business owners who want to improve their business processes with AI and software. You don't need to know them all to get started. But if you know what a context window or a system prompt is, you ask a supplier better questions and spot sooner when a proposal sounds too good. The examples below are fictional and serve only as illustrations. ## AI agent An AI agent is software that uses a language model to work towards a goal by itself. It reads information, decides on the next step and uses tools to take it, such as a search function, a calendar or a customer system. Unlike a chatbot, an agent doesn't just talk; it acts. Take an installation company: an agent reads fault reports, looks up the customer and prepares a scheduling proposal for a planner to approve. What the agent may do on its own is decided up front. More on [AI agents for businesses](https://goudbeek.com/en/solutions/agents) and in [what an AI agent is and does](https://goudbeek.com/en/notes/what-is-an-ai-agent). ## Large language model (LLM) A large language model, or LLM, is an AI model trained on vast amounts of text, which lets it process and write text. It keeps predicting which piece of text is likely to come next. That is enough to summarise, translate, classify and draft. A language model does not know your business. It only knows your current prices, customers and agreements once you give it that information. It is the engine under many [AI solutions for businesses](https://goudbeek.com/en/solutions), not the solution itself. ## Chatbot A chatbot is a program that answers questions in a conversation, usually in a window on your website or in a messaging app. Modern chatbots often run on a language model, and they only become truly useful when they answer from your own information. Think of a web shop where a chatbot answers questions about delivery and returns and brings in a colleague whenever it is not sure. In essence, the EU AI Act (Article 50) asks that people are told they are talking to an AI system, unless that is already obvious. The official text sets out what applies to you. See also [a chatbot for your business](https://goudbeek.com/en/solutions/chatbots). ## RAG (answers from your own documents) RAG stands for retrieval-augmented generation. Before the language model answers, the system first looks up the relevant passages in your own documents and passes them along with the question. The answer is then grounded in your source, and the system can show where it came from. Picture an accounting firm with an internal assistant that answers questions from the firm's own work instructions. When an instruction changes, you update the document, not the model. ## Embedding An embedding is a series of numbers that captures the meaning of a piece of text. Texts that mean the same thing get numbers that sit close together, even when they use different words. That lets a system search by meaning rather than by exact wording. If a customer asks 'when will my parcel arrive?', embeddings help the system find the page on delivery times, even though the word 'parcel' appears nowhere on it. ## Vector database A vector database stores embeddings and quickly finds the pieces of text whose meaning is closest to a question. It is often the archive behind a RAG system: your documents are cut into pieces, turned into embeddings and kept there. Not every project needs one. With a handful of documents, a simpler search can be enough, and then that is the better choice. ## Context window The context window is how much text a language model can take in at once: your question, the instructions, any documents supplied and the conversation so far. It is measured in tokens. Whatever no longer fits drops out of view or has to be shortened. That is why you don't paste every document you own into one question. A good system picks the pieces that matter, which is exactly what RAG does. ## Prompt and system prompt A prompt is the instruction or question you give a language model. A system prompt is a standing set of instructions the builder supplies in advance and the user usually does not see: the system's role, its tone and what it should stay away from. An accounting firm might put in its system prompt that tax questions always go to an adviser. Note that a system prompt steers behaviour but is not a lock on the door. Real boundaries come from what the system can technically reach. ## Token A token is the unit in which a language model reads and writes text. Often it is a word or part of a word; long or compound words are split into several tokens. If you use a language model through an API, you usually pay per token. For you, this mostly means that long documents and long conversations cost more and fit less easily in the context window. ## Fine-tuning Fine-tuning means training an existing language model further on your own examples, so it handles a particular style, format or task better. Think of answers that must always follow a fixed structure, or jargon the model otherwise keeps getting wrong. For knowledge that changes, such as prices, stock or procedures, fine-tuning is usually not the most practical route. Keep that information current in your documents and retrieve it with RAG. ## Hallucination A hallucination is an answer that sounds convincing but is wrong: an invented source, a wrong date, a rule that exists nowhere. It happens because a language model predicts likely text; it does not check facts. You cannot rule it out entirely, but you can keep it small and visible. Have answers come from your own sources, let the system say 'I don't know' when the source is silent, and have a person check wherever a mistake costs something. ## Prompt injection Prompt injection is an attack in which outside text poses as an instruction. An email, web page or PDF then contains instructions for the AI system, sometimes invisible to a person, for instance to forward data. The model cannot reliably tell content from instruction, because to the model it is all text. Filtering helps only so much. The real defence is in the design: limit what the system can reach and make actions with consequences wait for a person. Read more in [prompt injection, explained without the hype](https://goudbeek.com/en/notes/prompt-injection-without-the-hype) and about [AI security and access rules](https://goudbeek.com/en/security). ## API and webhook An API is an agreed way for one piece of software to talk to another: one system asks a question or gives an instruction, the other responds. A webhook works the other way round: a system sends a message by itself as soon as something happens, such as a new order. An example: a web shop sends a webhook for every order, after which a connection uses the accounting package's API to prepare the invoice. The systems talk to each other and nobody retypes anything. ## Process automation [Process automation](https://goudbeek.com/en/solutions/workflows) means that recurring steps in a business process are done by software instead of people. That can involve AI, but often something simpler will do: a rule, a form or a link between two systems. AI mainly adds something where text has to be read or judged, such as extracting data from work orders or invoices: [document processing with AI](https://goudbeek.com/en/solutions/documents). So start with the process, not the technology. Which processes lend themselves to it is covered in [which business processes you can automate](https://goudbeek.com/en/notes/which-processes-to-automate). ## How does Goudbeek approach AI in your business? We start with the work, not with the terms above. First we map how a process runs today and where it gets stuck. Then we pick the simplest thing that works, and that is not always AI. Where AI does fit, we build one working prototype on your real data, at a fixed price agreed up front after a short intake. We put in writing what the system may see and do, and what waits for a person. Then it goes live on managed hosting from €39 per month. The code and content stay yours. Read [how we put AI to work in business processes](https://goudbeek.com/en/solutions), [request a quote](https://goudbeek.com/en/pricing) or [get in touch](https://goudbeek.com/en/contact). ## Sources 1. [EUR-Lex — Regulation (EU) 2024/1689 (AI Act) eur-lex.europa.eu](https://eur-lex.europa.eu/eli/reg/2024/1689/oj) 2. [OWASP Top 10 for Large Language Model Applications owasp.org](https://owasp.org/www-project-top-10-for-large-language-model-applications/) ## Frequently asked questions ### What is the difference between a chatbot and an AI agent? A chatbot answers questions in a conversation; a good business chatbot does so based on your own information. An AI agent goes a step further: it uses tools and carries out actions, such as looking up a customer, preparing a draft email or proposing an appointment. That is why an agent needs clear boundaries agreed in advance. Read more about [chatbots for businesses](https://goudbeek.com/en/solutions/chatbots) and [AI agents that carry out tasks](https://goudbeek.com/en/solutions/agents). ### Do I need fine-tuning to make AI work with my own information? You usually do not need fine-tuning to make AI work with your own information. If you want a system to answer from your own manuals, terms or procedures, RAG is generally the better route: the system first looks up the right passages and bases its answer on them. When your information changes, you only update the document. Fine-tuning is mainly useful for steering style, format or a specific task. ### How do you stop an AI system from making things up? You cannot entirely stop an AI system from making things up, because a language model predicts likely text and does not check facts. You can make the risk small and visible, though. Have answers come from your own sources, show where each answer came from, let the system say honestly when it does not know, and have a person check wherever a mistake has consequences. ### Is prompt injection a risk for a small business too? Yes, prompt injection is a risk for a small business too, as soon as an AI system reads outside text, such as emails, websites or uploaded files. The size of your business makes little difference. What matters is what the system can reach if an attack succeeds. Limit access to what the task needs and make actions with consequences wait for a person. More on this on our page about [AI security](https://goudbeek.com/en/security). ### Do I need to know these terms before I start with AI? No, you do not need to know AI terms before you start with AI. It does help to recognise them, so you can judge a proposal and ask the right questions. But the most important question is not technical: which work takes too much effort now, or often goes wrong? Start there, because the technology follows from the process. Our page on [AI solutions for businesses](https://goudbeek.com/en/solutions) shows what that looks like. ### What does it cost to use AI in my business? What it costs to use AI in your business depends on the process, the systems that need connecting and how much a person needs to keep checking. At Goudbeek we agree a fixed price after a short intake, so you know where you stand up front. Once the system is running, managed hosting starts from €39 per month. The AI model itself is billed separately by usage; the quote shows what to expect. [Request a quote](https://goudbeek.com/en/pricing). About the author: Sam Goudbeek Sam Goudbeek is the founder of Goudbeek in Almere, which improves business processes with AI and software. [More about Sam and Goudbeek](https://goudbeek.com/en/about) ## Further reading Where this article meets the work. - [AI solutions for businesses](https://goudbeek.com/en/solutions): Agents, chatbots, document processing and dashboards, on your own data. - [AI agents](https://goudbeek.com/en/solutions/agents): Software that works a queue and knows when to stop. [All articles](https://goudbeek.com/en/notes) ## More from the knowledge base ### [What is an AI agent? An explanation for businesses](https://goudbeek.com/en/notes/what-is-an-ai-agent) An AI agent works through a task on its own, within limits you set. How it differs from a chatbot, where it fits in a business, and how to start small. 24 Sept 2026 6 min read ### [Chatbot or AI agent: which one does your business need?](https://goudbeek.com/en/notes/chatbot-or-ai-agent) A chatbot answers questions; an AI agent does work. How to tell which one your business needs, what both require, and when it makes sense to combine them. 24 Sept 2026 6 min read ### [Implementing AI in a small business: where do you start?](https://goudbeek.com/en/notes/implementing-ai-small-business) Implementing AI in a small business starts with one recurring job, not a tool. Here is how to pick the process, set boundaries and test on real data. 24 Sept 2026 5 min read