Jargon Buster

The buying and selling of online ad inventory through automated methods rather than human actions, often using AI. This includes but is not limited to Real-Time Bidding (RTB).

A security risk where someone deliberately gives an AI misleading or malicious instructions to manipulate its output or behaviour. 

This can cause the AI to ignore its original instructions, reveal sensitive information, or produce unsafe or incorrect responses. It’s especially important in customer-facing AI tools like chatbots or ad systems. 

A user tries to trick a brand chatbot into ignoring its rules and exposing internal pricing rules or generating unauthorised content. 

Process of providing instructions, questions, context or examples to an AI model to guide the output it generates. The clarity and quality of a prompt strongly influences the relevance, accuracy, tone and format of the response. 

It acts as the parent concept for techniques such as zero-shot prompting, few-shot prompting and chain-of-thought prompting.

 A marketer might prompt an AI to “create three ad headlines for a luxury travel brand targeting high-income audiences,” specifying audience, tone, and output format to shape the result.

 A technique where AI looks up relevant external information before generating an answer, so its responses are more accurate and up to date. 

Instead of relying only on its training data, the AI retrieves information from sources like documents, databases, or websites to improve the quality and relevance of its responses. RAG is one way to achieve grounding. See more here.

A brand chatbot checks product catalogues and FAQs before answering questions about pricing, availability, or features, ensuring the information is current. 

 A way of transacting media that allows an individual ad impression to be put up for bid in real time. This is done through a programmatic on-the spot auction, which is similar to how financial markets operate. RTB allows for addressable advertising; the ability to serve ads to consumers directly based on their demographic, psychographic, or behavioural attributes. AI plays a key role in bid adjustments and targeting. 

A way of transacting media that allows an individual ad impression to be put up for bid in real time. This is done through a programmatic on-the spot auction, which is similar to how financial markets operate. RTB allows for addressable advertising; the ability to serve ads to consumers directly based on their demographic, psychographic, or behavioural attributes. AI plays a key role in bid adjustments and targeting.

A theoretical idea in AI where a system could improve its own abilities over time by repeatedly updating and enhancing itself. 

Today, AI can help humans improve models and code, but it cannot fully and independently redesign or upgrade itself without human involvement.

Isolated environments where AI systems can be tested safely without affecting live systems or real data. 
 
A marketing team tests a new AI-driven campaign optimisation tool in a sandbox environment before applying it to live campaigns, ensuring it performs as expected without risking budget or results.  
 
Sandboxes allow advertisers to experiment with AI tools safely, reducing risk, catching errors early, and ensuring systems are ready before going live.

Sovereign AI refers to artificial intelligence systems that are developed, hosted, and controlled within a specific country or organisation’s own infrastructure, rather than relying on external providers or foreign-owned platforms. 

The goal is to maintain control over data, models, and computing resources so that sensitive information stays within a defined jurisdiction or governance framework.