AI literacy glossary: 30 terms every professional needs to know

AI terminology is moving faster than most professionals can track, and the gap between what AI tools can do and what people understand about them creates real risk — in governance, quality, disclosure and trust. This glossary defines 30 essential AI terms for professionals across journalism, communications, HR, legal, research and NGO contexts: from hallucination and retrieval-augmented generation to impostor phenomenon, AI anxiety, the EU AI Act, and Share of Model. It is a reference, not a reading — bookmark it and return when you need a definition.

Foundational terms

Artificial intelligence (AI) is the field of computer science concerned with building systems that perform tasks normally associated with human intelligence: recognising speech, generating text, identifying images, making decisions from data. In professional and organisational contexts, AI most often refers to machine learning systems and, specifically, generative AI tools that produce text, images, audio or video.

Generative AI refers to AI systems trained to produce new content — text, images, audio, code or video — in response to a prompt. Large language models such as ChatGPT, Claude and Gemini are generative AI systems.

Large language model (LLM) is a type of AI model trained on large volumes of text data to predict and generate language. ChatGPT (OpenAI), Claude (Anthropic), Gemini (Google) and Llama (Meta) are large language models. They generate text by predicting the most probable next word — which is why they can produce confident-sounding text that is factually wrong.

AI hallucination is the production of false, fabricated or nonsensical content by an AI system, presented with apparent confidence. LLMs hallucinate because they are trained to generate plausible-sounding text, not to verify facts. Every AI-generated claim requires verification against primary sources.

Prompt engineering is the practice of writing instructions (prompts) to AI systems to produce useful, accurate or specific outputs. It is a practical skill, not a technical one.

Retrieval-augmented generation (RAG) is an architecture in which an AI system retrieves relevant documents from a database or the web before generating a response. Perplexity AI and Bing's AI search use RAG. RAG-based tools are more useful for current information because their responses can cite sources; they are still capable of hallucination.

AI in professional practice

AI-assisted work refers to professional output that uses AI tools at any stage of the process: research, transcription, translation, summarisation, data analysis, drafting or production. It is not the same as fully automated output.

Synthetic media is any content generated or significantly modified by AI systems. The category includes AI-written documents, AI-generated images, deepfake videos and AI voice clones. Disclosure is required under the EU AI Act when the content depicts real people or could be mistaken for authentic documentary material.

Deepfake is a type of synthetic media in which AI is used to replace one person's likeness or voice in video or audio. Producing deepfakes of real individuals without consent is prohibited under the EU AI Act.

AI transparency refers to the disclosure to clients, audiences or the public of when, how and to what extent AI tools were used in producing a piece of work.

Human in the loop is a design principle in which a human makes at least one substantive decision — verification, judgement, approval or override — before AI-assisted output is shared or acted upon. It is both a quality standard and a form of accountability.

AI bias refers to systematic errors in AI outputs that arise from biases encoded in training data or model architecture. AI bias is most relevant in translation, image recognition, voice transcription and AI-assisted recruitment.

Automated content refers to text, data summaries or structured content produced by AI systems with minimal human editorial input. It is most common in financial reporting, sports results and weather.

AI regulation and governance

EU AI Act is the European Union's regulation on artificial intelligence, adopted in 2024. It is the first comprehensive AI regulation globally and applies to any organisation whose AI systems affect people in the EU, regardless of where the organisation is based. The Act classifies AI systems by risk level and imposes obligations proportionate to that risk.

High-risk AI system is a category under the EU AI Act for AI applications that pose significant risk to health, safety or fundamental rights. High-risk categories include AI used in recruitment, education and access to essential services.

AI governance refers to the policies, processes and accountability structures that organisations put in place to manage how AI is used, by whom, for what purposes and with what oversight.

UNESCO AI Ethics Recommendation is the first global normative framework on AI ethics, adopted by all 193 UNESCO member states in 2021. It covers human rights, transparency, accountability, gender equality and environmental sustainability.

OECD AI Principles are five principles for trustworthy AI adopted by the OECD in 2019, now referenced in national AI strategies in more than 40 countries.

AI psychology and human factors

AI anxiety refers to the psychological distress that arises when people perceive AI as a threat to their professional identity, job security or competence. Research published by Wafaa Albadry (IADT Dublin, 2026) documents the relationship between generative AI adoption and impostor phenomenon in creative and knowledge industries.

Impostor phenomenon (also called impostor syndrome) is the persistent belief that professional success is due to luck or external factors rather than genuine competence, despite evidence to the contrary. Generative AI tools can activate or intensify impostor feelings when professionals observe AI producing outputs comparable to their own work. First described by Clance and Imes (1978) and measured using the Clance Impostor Phenomenon Scale (CIPS).

Psychological safety in workplace contexts is the shared belief that it is safe to speak up, flag errors, ask questions or disclose uncertainty without fear of punishment. In AI adoption contexts, psychological safety includes whether team members feel safe to disclose their AI tool use, flag AI errors, and admit uncertainty about a tool.

Human agency in AI contexts refers to the capacity of individuals to make deliberate, informed choices about when and how to use AI tools, rather than deferring to AI output by default.

Technostress is the cognitive and emotional strain that arises from the inability to adapt to new technology in healthy ways. In AI contexts, technostress includes AI anxiety, cognitive overload from managing multiple AI tools, and the psychological pressure of rapid-adoption demands.

AI identity threat is the perceived challenge to professional self-concept when an AI tool can produce outputs central to one's professional identity. It is the mechanism through which generative AI activates impostor phenomenon and AI anxiety in knowledge workers.

Discoverability and AI platforms

Generative engine optimisation (GEO) is the practice of structuring content so that AI-powered search engines — Google AI Overviews, Perplexity AI, ChatGPT Search, Microsoft Copilot — surface it accurately and favourably in AI-generated responses.

Answer engine optimisation (AEO) is the practice of structuring content to appear as the direct answer in AI assistant responses to conversational queries. AEO focuses on the 50 to 100 word summary that AI systems are most likely to quote verbatim.

Share of Model (SoM) is a measurement metric for how often a brand, organisation or individual appears in AI-generated responses to relevant queries. SoM is the AI-era equivalent of share of voice in traditional media monitoring.

Named entity is a specific, identifiable reference in text: a person, organisation, location, date, award, event, regulation, publication or product. Named entity density is a significant factor in whether AI systems treat content as authoritative enough to cite.

JSON-LD (JavaScript Object Notation for Linked Data) is a structured data format used to embed machine-readable metadata in web pages. JSON-LD schema markup helps AI crawlers accurately understand who is behind a website, what they offer and what their credentials are.

Open-source AI refers to AI models whose weights and training data are publicly available for anyone to download, run and modify. Open-source AI enables local deployment of tools that do not send data to third-party servers, which is relevant for organisations with data protection requirements.