Glossary
Terms, explained
Plain-English definitions of the AI, automation and AEO/SEO terms used across our Build Logs and Quick Wins.
111 terms · A–Z
111 terms, A–Z
A
25 terms- Agent LoopThe repeated cycle in which an AI agent observes its current state, chooses an action, receives the result and decides what to do next.
- Agent StateThe structured information an AI agent keeps about the current task, completed actions, intermediate results and next steps.
- Agentic AIAn approach to building AI systems that act with some autonomy: they plan steps, use tools and adjust to results, instead of only answering a single prompt.
- AI AgentA software system that uses an AI model to pursue a goal: it decides what to do next, calls tools, reads the results and repeats until the task is done or it gets stuck.
- AI Agent ContextEverything an AI agent can see when it makes a decision: instructions, conversation, tool results, retrieved documents and any task state placed in the prompt.
- AI Agent EvaluationThe practice of testing an AI agent’s task success, tool choices, reliability, safety and cost against defined scenarios.
- AI Agent FrameworkA software library or platform that provides reusable components for building AI agents, such as tool calling, state, memory and execution loops.
- AI Agent GroundingTying an AI agent's answers and actions to trusted sources, such as your documents, databases or live tool results, so it does not rely on guesses.
- AI Agent MemoryThe information an AI agent can keep and recall beyond a single prompt, such as earlier conversation turns, saved facts, past results or user preferences.
- AI Agent ObservabilityThe ability to inspect an agent’s steps, model calls, tool inputs and outputs, errors, timing and cost during real executions.
- AI Agent OrchestrationThe coordination of an AI agent’s steps, tools, sub-agents and control flow so a task runs in a predictable sequence.
- AI Agent PlanningThe process an AI agent uses to break a goal into steps, choose an order and revise the plan when new information arrives.
- AI AssistantA conversational AI tool that helps a person with tasks such as answering questions, drafting text or finding information, usually by responding to their requests turn by turn.
- AI AutomationUsing AI models inside automated processes so that tasks like sorting, summarizing, drafting or routing happen without a person doing each step by hand.
- AI BenchmarkA defined test or collection of tests used to compare AI models or systems on specified tasks and evaluation criteria.
- AI CitationWhen an AI assistant or AI search tool names or links to your page as a source for something in its answer.
- AI CrawlerAn automated bot run by an AI company that fetches web pages, either to train models, to build an AI search index or to load a page when a user asks about it.
- AI SearchSearch that uses AI models to understand a question and write a direct answer, often with sources, instead of only listing links to pages.
- AI Tool OrchestrationThe routing and sequencing of tool calls made by an AI system, including validation, error handling and result handoffs.
- AI VisibilityHow often and how prominently your brand, site or content shows up in answers from AI assistants and AI search tools.
- AI-Readable ContentContent written and formatted so that both people and AI systems can quickly understand what it says: clear headings, direct answers, plain language and specific facts.
- AI-Ready WebsiteA website that AI crawlers and agents can reach, read and understand reliably: crawlable, fast, well structured and clear about what each page covers.
- Answer Engine OptimizationMaking your content easy for answer engines, such as AI assistants and AI search, to find, understand and use when they give a direct answer to a question.
- APIA defined way for one piece of software to ask another to do something or share data, usually over the web using requests and responses.
- Article SchemaStructured data that describes an article or blog post, including its headline, author, publish date and last-modified date, using the Article type from Schema.org.
C
6 terms- Canonical URLThe preferred address of a page that you tell search engines to treat as the main version when the same or very similar content is reachable at several URLs.
- Constrained DecodingA generation technique that restricts which tokens a model may produce so its output follows a grammar, schema or other formal constraint.
- Context CompressionTechniques for reducing the size of information sent to a model while preserving the details needed for the current task.
- Context WindowThe maximum amount of text, measured in tokens, that a language model can take in and consider at once, including your instructions, the conversation and its own reply.
- Conversation SummarizationThe process of condensing a conversation into a shorter record of its key facts, decisions and unresolved questions.
- CrawlingThe process where a search engine or other bot follows links and fetches web pages so it can discover and read them.
D
2 terms- dateModifiedA Schema.org property that states when a page's content was last meaningfully changed, used in structured data such as Article markup.
- Document ChunkingSplitting long documents into smaller pieces before indexing them, so a retrieval system can return just the relevant part instead of a whole document.
E
2 terms- EmbeddingsLists of numbers that represent the meaning of a piece of text (or an image), so that similar meanings end up close together and can be compared by a computer.
- Error HandlingPlanning for and reacting to things going wrong in software or a workflow, so a failure is caught, recorded and dealt with instead of silently breaking or corrupting data.
F
4 terms- FAQPage SchemaStructured data that marks up a page of questions and their single published answers, using the FAQPage type from Schema.org.
- Few-Shot PromptingA prompting technique that includes a small number of examples to demonstrate the task or response pattern expected from a model.
- Foundation ModelA broadly trained model that can be adapted to many downstream tasks, often through prompting, fine-tuning or additional components.
- Function CallingThe original name for tool calling: the model outputs a structured request to run a named function with specific arguments, and your code executes it.
G
2 terms- Generative AIAI that creates new content, such as text, images, audio, video or code, instead of only classifying or predicting from existing data.
- Generative Engine OptimizationShaping your content and site so that generative AI search tools are more likely to use it and mention it in the answers they write.
H
3 terms- H1The main heading of a page, marked up as the top-level heading in HTML, that tells readers and search engines what the page is about.
- Human-in-the-LoopA design in which a person reviews, approves or corrects an AI system's work at defined points instead of letting it run fully on its own.
- Hybrid SearchA search method that combines lexical matching, such as keyword search, with semantic retrieval based on vector similarity.
I
5 terms- IdempotencyA property of an operation where running it once or many times gives the same end result, so repeating it after a failure or duplicate does no extra harm.
- Image Alt TextA short text description added to an image in HTML that screen readers announce and search engines use to understand what the image shows.
- IndexingWhen a search engine stores and organizes a page it has crawled so that it can be shown in search results.
- Inference LatencyThe time an AI inference request takes, often measured from request submission to the first token or to completion.
- Internal LinkingLinks from one page of your site to another page on the same site, which help visitors navigate and help search engines find pages and understand how they relate.
J
2 terms- JSON ModeAn API or model setting that encourages or constrains a response to be valid JSON, usually without enforcing every field in a specific schema.
- JSON-LDA format for writing structured data as a block of JSON in a page's HTML, and the method most search engines recommend for adding Schema.org markup.
K
2 termsL
7 terms- Large Language ModelA neural network trained on very large amounts of text to predict and generate language. It is the engine behind most chatbots, AI assistants and AI agents.
- LLM EvaluationThe process of measuring a language model’s quality on representative tasks using test examples, scoring criteria and human or automated review.
- LLM HallucinationWhen a language model produces information that sounds confident and plausible but is false, invented or not supported by any source it was given.
- LLM InferenceThe process of running a trained language model to produce a response or prediction from a supplied input.
- LLM TemperatureA sampling setting that changes how strongly a language model favors higher-probability next tokens when generating output.
- llms.txtA proposed plain-text Markdown file at /llms.txt that gives language models a short guide to a site and links to its most useful pages.
- Low-Rank Adaptation (LoRA)A parameter-efficient fine-tuning method that trains small low-rank update matrices while keeping most pretrained model weights frozen.
M
9 terms- MCP ClientAn AI application or connector that communicates with an MCP server to discover and use its tools, resources or prompts.
- MCP ServerA program that exposes tools, resources or prompt templates to compatible AI applications through the Model Context Protocol.
- Meta DescriptionA short summary of a page in its HTML that search engines may show under the title in results, written to help people decide whether to click.
- Model Context Protocol (MCP)An open protocol that standardizes how AI applications connect to external tools, resources and prompts provided by MCP servers.
- Model Fine-TuningThe process of continuing model training on a targeted dataset to adapt a pretrained model to a particular task, style or domain.
- Model QuantizationA technique that represents model weights or activations with lower numerical precision to reduce memory use and potentially improve inference speed.
- Model RoutingThe process of choosing which model should handle a request based on factors such as task type, quality needs, latency, cost or availability.
- Multi-Agent SystemA setup in which multiple AI agents cooperate, divide work or check one another to complete a task that would be harder for one agent alone.
- Multimodal AIAI systems that process or generate more than one type of data, such as text, images, audio or video.
N
1 termO
3 terms- Open-Weight ModelA model whose trained weights are made available for others to download or run, subject to the model’s license and terms.
- Organization SchemaStructured data that describes a company or publisher: its name, logo, website and official profiles, using the Organization type from Schema.org.
- Output ParserA component that converts a model’s response into a structured value and checks whether it matches an expected format.
P
3 terms- Prompt ChainingA technique that splits a complex task into multiple model calls, where the output of one step becomes input to the next.
- Prompt EngineeringThe practice of writing and structuring the instructions, examples and context given to an AI model so it produces reliable, useful output.
- Prompt TemplateA reusable prompt structure with named variables or placeholders that are filled with task-specific information at runtime.
R
6 terms- Race ConditionA bug where the result depends on the timing of two or more processes that run at the same time, so the same code sometimes works and sometimes fails.
- ReAct PromptingA prompting approach that interleaves reasoning steps with actions, such as tool calls, and uses the resulting observations to guide the next step.
- Reranker ModelA model trained or adapted to score how relevant a document or passage is to a particular search query.
- Retrieval-Augmented GenerationA technique that finds relevant passages in your own data and adds them to the prompt, so a language model answers from those sources instead of from memory alone.
- Retry LogicRules for automatically trying a failed operation again, such as how many times, how long to wait between attempts and when to give up.
- robots.txtA plain-text file at the root of a site that tells crawlers which parts of it they may or may not request. It is a request, not a security control.
S
14 terms- Schema.orgA shared vocabulary of types and properties, such as Article, Organization and Person, used to describe web content in structured data that search engines understand.
- Search Engine OptimizationThe practice of improving a website so that search engines can find, understand and rank its pages, and so more of the right people reach them.
- Search IntentThe underlying goal behind a search query: what the person actually wants to learn, find, compare or do when they type it.
- Search RankingThe position a page appears in for a given search query on a search engine results page, which affects how many people see and click it.
- Search RerankingA second-stage retrieval process that reorders an initial set of search results according to how relevant each result is to a query.
- Semantic CachingA caching approach that reuses a previous result when a new request is judged semantically similar to an earlier request.
- Semantic SearchSearch that matches by meaning instead of exact keywords, so a query can find relevant content even when it uses different words.
- Small Language ModelA language model designed with fewer parameters or lower resource requirements than large general-purpose models, often for focused tasks or local deployment.
- Speech RecognitionTechnology that converts spoken audio into a text transcript or another representation of recognized speech.
- Speech-to-TextThe process of converting audio speech into written text using an automatic speech recognition system.
- State MachineA model of a system as a set of defined states and transitions triggered by events or conditions.
- Structured DataCode added to a page that labels its content in a standard format, such as who wrote it, what it is and when it was updated, so machines can read it reliably.
- Structured OutputA model response constrained to a defined format, such as fields in a JSON object, instead of unrestricted prose.
- System PromptA set of instructions given to an AI model before the conversation starts that defines its role, rules, tone and limits for every response that follows.
T
8 terms- Technical SEOThe part of SEO that makes sure search engines can crawl, render and index a site properly: speed, structure, links, sitemaps, redirects and correct signals.
- Text-to-SpeechTechnology that generates spoken audio from written text, often with selectable voices, speaking styles and languages.
- Title TagThe HTML title of a page, shown as the clickable headline in search results and in browser tabs, and a key signal of what the page is about.
- TokenizationThe process of splitting text or other input into smaller units called tokens that a language model can process.
- Tokens per SecondA throughput measure showing how many tokens a model generates or processes each second under a stated workload and measurement method.
- Tool CallingA way for an AI model to request that your software run an action, such as a search, database query or API call, and then use the result in its next response.
- Tool SchemaA formal description of a tool’s name, purpose, parameters and expected input types that helps a model or application call it correctly.
- Top-p SamplingA decoding method that samples from the smallest set of likely next tokens whose cumulative probability reaches a chosen threshold.
V
2 terms- Vector DatabaseA database built to store embeddings and quickly find the ones most similar to a query, which powers semantic search and retrieval for AI applications.
- Vision-Language ModelA model trained to work with both visual input and language, enabling tasks such as image question answering, captioning and visual reasoning.
W
3 terms- WebhookAn automatic HTTP message that one system sends to a URL you provide when something happens, so your software can react in real time instead of polling.
- WorkflowA defined sequence of steps, with triggers and conditions, that moves a task or piece of data from start to finish.
- Workflow AutomationUsing software to run a series of steps automatically when something happens, such as moving data between apps, sending messages or updating records, without manual work.