- Agent / custom agent
- A tailored AI assistant configured for a specific task, typically built on top of a general-purpose model such as ChatGPT or Claude by feeding it instructions and reference documents, rather than by writing code.
- API (Application Programming Interface)
- A way for one piece of software to send requests to another, commonly used to pass data to an AI model for processing.
- ChatGPT
- A consumer AI product from OpenAI, often used as the base model for a custom agent built through plain-language instructions rather than code.
- Claude
- An AI model from Anthropic, sometimes used as an alternative or complement to ChatGPT.
- Claude Code
- A coding-focused AI tool used to support the technical side of an AI build.
- CMS (Content Management System)
- The publishing software a newsroom uses to write, edit and publish articles.
- Data minimisation
- A privacy principle of collecting only the data a task strictly requires, often paired with role-based access controls limiting who can view, edit or export it.
- Disclosure
- A statement to the audience that AI was used in producing a piece of content. Newsrooms vary on whether disclosure happens every time or selectively, generically or by naming the tool, and on air, in text, or via caption or watermark.
- False positive / false negative
- In moderation, a false positive is benign content wrongly flagged as harmful, such as a comment saying a show "was a bomb" (meaning excellent) being flagged as terrorism; a false negative is genuinely harmful content that passes unflagged, such as abusive slang terms a generic tool doesn't recognise.
- Google AI Studio
- A free environment for building and testing AI prototypes without first standing up cloud infrastructure.
- Google Discover / meta title
- Google Discover is a feed that surfaces articles to users based on search-engine signals, including an article's meta title, a short title distinct from the on-page headline that search engines display.
- Guardrails
- Rules, restrictions or technical controls built into an AI workflow to limit risk, such as barring AI from generating editorial imagery or requiring human sign-off before publication.
- Hallucination
- When an AI model generates false or fabricated information presented as fact, a persistent problem in large language models.
- House style / house format
- A newsroom's established tone, structure and conventions. Tools are often trained on an outlet's own past articles so their output reads as the newsroom's own rather than generic AI text.
- Human-in-the-loop
- A design principle in which a human must review, approve or intervene in an AI-driven process before it takes effect, rather than the AI acting autonomously.
- Keyword library
- A list of terms, compiled from how a specific community actually speaks, used to flag content for moderation or fact-checking.
- LLM (Large Language Model)
- A type of AI model trained on large volumes of text to generate human-like language, prone to occasional hallucination.
- Meta dashboard
- Facebook/Instagram's own analytics and moderation interface, sometimes integrated alongside a custom moderation tool rather than replaced by it.
- MVP (Minimum Viable Product)
- A basic, working version of a tool built to test a concept before investing in a fuller build.
- n8n
- A workflow-automation tool used to connect other software together, such as linking a messaging platform to an AI model and a database. Free if self-hosted, or available as a paid licence.
- NapoleonCat
- A third-party social media management tool.
- NotebookLM
- A Google AI tool used for research, summarisation and content generation, including converting written text into narrated video or audio.
- OpenAI Platform
- The API service (distinct from the consumer ChatGPT app) used to send data to an AI model for processing.
- Opus Clip
- An off-the-shelf AI tool that cuts long video or audio recordings into short clips suitable for social platforms like TikTok and Instagram.
- Package
- A broadcast journalism term for a fully produced, narrated report combining voice, sound and interviews.
- PostgreSQL
- An open-source database system used to store structured data.
- Prompt / prompt-writing
- The instructions a user gives an AI model to produce a desired output, frequently identified as the main operational challenge in running a custom agent.
- Risk assessment
- A formal check of the risks of a given AI use, such as data exposure or source confidentiality, required before certain tools or data types are approved.
- Sentiment register
- A category of emotional tone used to classify audience comments.
- Shadow AI
- The informal, undisclosed use of AI tools by individual staff members outside of any newsroom-sanctioned process.
- Sheng
- Urban Swahili-English slang, which global moderation tools tend to miss.
- Slack
- A workplace messaging platform frequently used as the human-review interface in an AI workflow, rather than building review functions inside the AI tool itself.
- Trello
- A task-management tool.
- Vox pop
- Short for vox populi; brief, informal audience or street interviews used as a story format.
- Whitelist / approved-tools list
- A list of AI tools formally approved for staff use, typically reviewed periodically and paired with rules on what data or tasks are off-limits to unapproved platforms.
