AI glossary
AI terms, explained simply.
The terms that come up most often in my conversations with managing directors of small and mid-sized companies – briefly explained, with what they mean for your day-to-day work.
Jump to: Generative AI · Large Language Model (LLM) · Prompt and prompting · Hallucination · Shadow AI · AI policy · AI literacy (Art. 4 AI Act) · EU AI Act · Data protection with AI tools · AI agent · RAG (retrieval augmented generation) · AI maturity
As of October 2026 · Author: Stefan Quisdorf, AI Professional (IHK)
Basics
Generative AI
Generative AI creates new content – text, images, tables or program code – based on patterns it has learned from large amounts of data. Well-known examples are ChatGPT, Microsoft Copilot, Google Gemini and Claude.
In practice: In small and mid-sized companies, generative AI helps above all with drafts, summaries and routine texts. The decision and the review remain with people.
Basics
Large Language Model (LLM)
A large language model is a language model trained on very large amounts of text. It predicts, word by word, what is most likely to come next – which is why the result sounds fluent but is not automatically correct.
In practice: What matters in practice: an LLM only “knows” what was in its training data or what you give it. It does not know current prices or internal processes on its own.
Application
Prompt and prompting
A prompt is the instruction you give an AI. Prompting means phrasing that instruction so that a useful result comes out: with a goal, context, examples and the desired format.
In practice: Good prompts are not secret knowledge. If you say who the text is for, how long it should be and what must not be included, you get noticeably better results.
More: AI workshop for teams
Risks
Hallucination
Hallucination is when an AI invents false facts, figures or sources and presents them convincingly. This is not a rare glitch but a consequence of how language models work.
In practice: That is why results must be checked before they reach the customer – especially figures, legal questions and quotes.
More: AI literacy in the team
Risks
Shadow AI
Shadow AI is the use of AI tools without the company’s knowledge or approval – often with private accounts. It usually does not stem from bad intent but from the lack of rules and approved tools.
In practice: Bans rarely help. Approved tools, simple ground rules and training that shows how to do it right work better.
More: AI literacy package · AI maturity check
Rules
AI policy
An AI policy defines which AI tools are allowed in the company, which data may go into them, who checks results and who the contact person is. One page is often enough to start with.
In practice: The policy is not an end in itself: it gives the team confidence and turns many individual experiments into a shared approach.
Rules
AI literacy (Art. 4 AI Act)
AI literacy means the knowledge and skills to use AI systems competently and to assess their opportunities and risks. Since February 2025, Article 4 of the EU AI Act has required companies to take care of this; since the “Digital Omnibus” of 27 July 2026, they must promote their staff’s AI literacy but no longer guarantee a specific level.
In practice: In practical terms: suitable training, simple ground rules and a short record of who took part and when. This is not legal advice – if in doubt, ask your legal adviser.
Rules
EU AI Act
The EU AI Act is the European regulation on artificial intelligence. It classifies AI systems by risk and sets strict requirements for high-risk applications – for example in recruitment or lending.
In practice: For most office applications in small and mid-sized companies, such as texts, summaries or research, the AI literacy obligation is what matters most – no reason to panic.
More: What the AI Act requires
Rules
Data protection with AI tools
As soon as personal data or trade secrets go into an AI tool, the GDPR and confidentiality apply. What matters is whether the provider uses the inputs for training, where the data is processed and whether a data processing agreement (DPA) is in place.
In practice: Rule of thumb for everyday work: use business versions with a DPA, no customer data in private accounts, and anonymise if in doubt.
Application
AI agent
An AI agent does not just give a single answer but carries out several steps on its own – such as researching, filling in a table and drafting an email. To do so, it is given access to tools like calendars, files or shop systems.
In practice: Agents save a lot of time but need clear limits: what may the agent do on its own, and what does a person have to approve?
Application
RAG (retrieval augmented generation)
With RAG, the AI searches your own documents – such as manuals, price lists or FAQs – before answering and bases its answer on them. It answers from your knowledge rather than only from its training data.
In practice: Typical uses are internal knowledge bases and customer service. The quality depends heavily on how up to date and clean the documents are.
Strategy
AI maturity
AI maturity describes how far a company has come in using AI. I distinguish four stages: experimenting, rules, applying and embedding.
In practice: Most small and mid-sized companies are somewhere between the first and second stage – a good starting point for a roadmap.
A term is missing – or you want to know what it means for your company?
30 minutes, free of charge and without obligation. You tell me where you stand – I tell you honestly whether and how I can help.