Glossary

AI customer service, defined

Short, honest definitions for the terms that come up when you're evaluating AI support. Each one answers the question first.

AI agent: An AI agent is a system that uses a language model to work towards a goal over multiple steps, typically by deciding which tools or actions to use along the way.
AI customer service: AI customer service is the use of a language model to answer customer questions automatically, usually on a website, in chat, or over email.
AI knowledge base: An AI knowledge base is a collection of an organisation's content indexed by meaning rather than by keyword, so a question can be answered from the relevant passage directly instead of returning a list of documents to read.
Answer engine optimization (AEO): Answer engine optimization is the practice of structuring content so that AI-powered answer engines can understand it, quote it accurately, and attribute it to you.
Chunking: Chunking is the process of splitting a document into smaller passages so that each can be indexed and retrieved independently.
Citation: A citation is a link from a generated answer to the specific source page or passage the answer was drawn from.
Confidence score: A confidence score is a numeric estimate of how well-supported an answer is, most usefully derived from how closely retrieved passages matched the question.
Containment rate: Containment rate is the proportion of customer conversations that are handled entirely within an automated channel, without transferring to a human.
Context window: A context window is the maximum amount of text, measured in tokens, that a language model can take into account when producing a single response.
Conversational AI: Conversational AI is software that holds a natural-language exchange across multiple turns, keeping track of what was already said rather than treating each message independently.
Cosine similarity: Cosine similarity measures how similar two vectors are by the angle between them, producing a score from -1 to 1 where 1 means they point in exactly the same direction.
Crawler: A crawler is a program that fetches web pages automatically and follows the links it finds, building a collection of a site's content.
Customer support automation: Customer support automation is the handling of support work — answering, routing, triaging or resolving — without a person doing it, usually for requests that recur in a predictable form.
Deflection rate: Deflection rate is the proportion of customer questions resolved by self-service or automation without reaching a human agent.
Embedding: An embedding is a numerical representation of text that captures its meaning, allowing passages with similar meaning to be found even when they share no words.
FAQ schema: FAQ schema is structured data using the FAQPage type, marking up a list of questions and their answers so search and answer engines can read them directly.
Fine-tuning: Fine-tuning is further training of an existing model on a specific dataset, adjusting its weights so it behaves more like the examples it was shown.
Grounding: Grounding is the practice of tying every generated answer to specific source text, so the answer reflects that text rather than the model's general knowledge.
Hallucination: A hallucination is an answer a language model produces that is fluent, confident and false — invented rather than drawn from any source.
Handoff: A handoff is the transfer of a conversation from an automated system to a human, along with enough context that the person does not have to start over.
Intent recognition: Intent recognition is the classification of a customer message into one of a predefined set of purposes, so a system can decide how to respond.
Knowledge base: A knowledge base is an organised collection of articles and documents that answers recurring questions about a product, service or organisation.
Knowledge gap: A knowledge gap is a question customers ask that a business's published content does not answer, revealed when an automated system has to refuse.
Large language model (LLM): A large language model is a system trained on very large amounts of text to predict what text should come next, which lets it write fluently, summarise, translate and answer questions.
Lead qualification: Lead qualification is the process of deciding which enquiries are a genuine fit for what you sell, so sales time goes to the ones most likely to convert.
Live chat: Live chat is a channel where a website visitor exchanges messages with a human agent in real time.
llms.txt: llms.txt is a proposed convention: a markdown file at a site's root that lists its most important pages in a form intended for language models rather than browsers.
Prompt injection: Prompt injection is an attack in which instructions are hidden inside content a language model reads, causing it to follow the attacker's directions instead of the operator's.
RAG (retrieval-augmented generation): Retrieval-augmented generation (RAG) is a technique where relevant source passages are retrieved from a document collection and given to a language model, so its answer is grounded in those passages rather than its training data.
RAG chatbot: A RAG chatbot is a chat interface that retrieves relevant passages from a specific document collection and answers only from those passages, rather than from the language model's general training data.
Semantic search: Semantic search finds text by meaning rather than by shared words, using numeric representations of passages so that similar ideas sit close together.
Structured data: Structured data is machine-readable markup added to a page, usually as JSON-LD following schema.org vocabulary, that states explicitly what the page is about.
System prompt: A system prompt is the instruction text placed in front of a conversation, setting a model's role, tone, rules and constraints before any user message arrives.
Token: A token is the unit of text a language model processes — roughly a word or part of a word, produced by splitting text into common sub-word pieces.
Vector database: A vector database stores numeric representations of text or images and finds the entries closest to a query vector, usually using an approximate nearest-neighbour index.
Website chatbot: A website chatbot is a chat interface embedded on a website that answers visitor questions, either from a scripted decision tree or by generating answers from source content.
Zero-click search: A zero-click search is one where the searcher gets what they needed from the results page itself and never visits a website.