Glossary

About-us page and author profiles as E-E-A-T signals — they make accountability and expertise visible and serve as central trust anchors for both quality raters and knowledge graph systems.

Accelerate is an open-source library from Hugging Face that makes PyTorch training code run on any hardware and in distributed setups with minimal changes — including mixed precision and FSDP and DeepSpeed support.

Deliberate split of a Google Ads account by brand, language, country, product line, or funnel stage — prerequisite for clean steering and reporting.

A structural layer in Google Ads between campaign and ad. Bundles keywords, ads, and optionally audiences that belong together thematically and are managed jointly.

One of the three Quality Score components in Google Ads. Rates how closely the ad copy matches the meaning of the booked keyword. Scale: above/average/below average.

A Google Ads rating for Responsive Search Ads and asset groups that measures how varied and complete the supplied assets are. Scale: Incomplete, Poor, Average, Good, Excellent.

An adapter is a small, freshly added network module injected into a frozen language model — only these modules are trained, while the base model itself stays unchanged.

Modular, reusable instruction packages (e.g. Claude Agent Skills) that an AI agent loads from disk on demand to perform specialized tasks — built around a SKILL.md holding metadata and instructions, plus optional scripts and reference files.

Agentic RAG is a RAG architecture in which an agent drives retrieval — it decomposes the question, searches iteratively, checks the hits, and decides itself whether a second or third pass is needed.

AI A/B testing uses artificial intelligence to automatically generate, serve and evaluate test variants of ads, copy or pages. AI identifies the better-converting variant faster and can test many combinations in parallel.

An AI agent combines a language model with tools and works toward a goal across multiple steps — typically in a loop of observing, planning and acting.

An AI content workflow is a structured process for creating content in which AI assists across several steps: briefing, draft, editing, fact-checking and publishing. The human steers, reviews and remains responsible for the result.

AI data analysis means evaluating data (CSV, Excel and more) via natural language: an AI system writes and runs code behind the scenes to clean, calculate and visualise files — e.g. ChatGPT's Advanced Data Analysis.

AI image generation creates images from text prompts using generative models. Tools such as Midjourney, DALL-E, Adobe Firefly and Stable Diffusion are used in marketing for concepts, visuals and image editing.

AI knowledge management makes scattered company knowledge searchable and usable with AI. Documents are turned into vectors so employees get answers from natural-language questions — usually via retrieval-augmented generation.

AI marketing is the use of artificial intelligence and machine learning to analyse marketing data, automate content and delivery, and optimise campaigns in real time. Its main areas are content, personalisation, analytics and generative engine optimization.

AI meeting assistants are software tools that automatically record, transcribe, summarize online meetings and derive action items from them. Well-known examples include Otter, Fireflies, Fathom and Microsoft Copilot in Teams.

AI Optimization (AIO) is the umbrella term for measures that increase the visibility of content in AI-powered answer systems. AIO encompasses variants such as GEO (Generative Engine Optimization) and AEO (Answer Engine Optimization).

AI presentations are slide decks created with the help of generative AI. From a short prompt or an outline, tools like Gamma or Beautiful.ai automatically generate structured slides with text, layout and matching images.

An AI research assistant is an AI system that, given a question, searches the web in multiple steps, reads and cross-checks sources, and synthesises a cited answer or report — e.g. ChatGPT Deep Research, Gemini Deep Research or Perplexity.

AI video generation creates moving images from text or image prompts. Diffusion-transformer models such as Google Veo, Kling, Runway or Sora iteratively denoise latent frames into coherent video, usually as clips of a few seconds.

AI-powered personalization automatically tailors content, offers and recommendations to individual users. Models analyse behaviour and profile data to deliver more relevant experiences in real time — consent-based and privacy-compliant.

Aider is an open-source AI pair programmer for the command line that edits code directly in the Git repository and commits every change automatically.

The visible text of a link — signals the topical context of the target page to search engines. Anchor-text distribution is one of the oldest and still-effective ranking signals for backlinks.

Answer Engine Optimization (AEO) is optimising content so it is delivered as a direct answer — in featured snippets, voice assistants and AI answers. Rather than just ranking pages, the content itself should become the answer.

Computational method in transformer models that weights which parts of the input text are most relevant for predicting the next token — instead of treating every part as equally important.

Attribution models are rules by which the value of a conversion is distributed across the touchpoints involved in a customer journey. They range from simple models such as last-click to data-driven approaches based on machine learning.

Audio normalization is the adaptive linear scaling of an audio signal to a target level — the tool first measures a reference value (peak, RMS or LUFS) and from that calculates the required gain.

Structured markup of a page's author via Schema.org `Person` with a `sameAs` array of profile URLs (LinkedIn, ORCID, Mastodon) — supports E-E-A-T and knowledge graph entity recognition.

A safety layer in Claude Code: a small LLM evaluates every tool call that is not explicitly allowlisted, live, and blocks destructive or unauthorised actions before they execute.

AutoGen is an open-source framework from Microsoft for building multi-agent applications with LLMs. Several AI agents solve tasks together in structured conversations — for example group chats or asynchronous message passing.

AutoGPT is an open-source experiment released in 2023 that runs a language model such as GPT-4 in an autonomous loop — it breaks a goal into subtasks itself, calls tools and works through them with minimal human intervention.

Automatic Speech Recognition (ASR), also called Speech-to-Text (STT), is the automatic conversion of spoken language into written text by an acoustic model.

B2B SEO addresses the specifics of business-customer websites — low search volumes, long funnels, multiple decision-makers, lead forms instead of instant purchase, login areas, and LinkedIn distribution as an amplifier.

B2B shops with login have SEO-specific challenges — most of the catalog and prices sit behind auth, classic crawling logic doesn't apply. SEO concentrates on public funnel pages.

A backlink is a link from an external domain pointing to your own website — one of the oldest and still most important ranking signals for Google.

Asynchronous API mode that collects many requests and processes them at a significant discount — results are typically delivered within 24 hours.

Batch inference is the bundled processing of many prompts in one pass — locally on a GPU or as an asynchronous API job — trading real-time latency for throughput and cost.

An AI benchmark is a standardized task suite that makes language models comparable — same prompts, same scoring, transparent results for reasoning, knowledge, or coding.

Microsoft's search engine with its own index and crawler (Bingbot) — also powers DuckDuckGo, Yahoo, Ecosia, and ChatGPT Search, making it worth more market share than Bing's direct usage suggests.

Microsoft's free tool for monitoring Bing visibility — index status, crawl errors, keyword reports, and URL submission. The Bing-index counterpart to Google Search Console.

Microsoft Bing's official web crawler, which captures pages for the Bing index and is therefore also indirectly relevant for DuckDuckGo, Yahoo, and Bing-based AI answers.

Blog and editorial SEO is the discipline of systematically making editorial content discoverable — topic planning, topic clusters, editorial care, internal linking, and update routines.

SEA strategy of bidding on your own brand name to secure the top SERP position and intercept competitor clicks.

Brand mentions are references to a brand on third-party sites — even without a link. For AI visibility (GEO) they count as a strong signal: language models learn from frequent, credible mentions that a brand is a known entity.

Search engine from the Brave browser company with its own index — privacy-focused and one of the few fully independent indexes alongside Google, Bing, and Yandex.

The widest Google Ads match type. Triggers ads on synonyms, related searches, and adjacent topics — driven by AI signals such as landing page, account history, and prior user searches.

Server-side compression methods that shrink text resources like HTML, CSS and JavaScript before transmission — Brotli usually delivers better ratios than Gzip today.

Browser use refers to AI agents that drive a real web browser — navigating, clicking, filling forms, reading content — to complete web tasks autonomously. The browser serves as both the tool and the perception surface.

A Google Ads asset/ad extension that displays short, non-clickable text snippets under the ad — typically USPs like "Free shipping" or "24/7 support".

The canonical tag (`rel="canonical"`) is an HTML element telling search engines the preferred URL for a page — central to avoiding duplicate-content issues.

Catastrophic forgetting describes the effect that a neural network loses previously acquired knowledge — wholly or partially — when it is trained further on new data; a key problem when fine-tuning language models.

A distributed server network that delivers static — and increasingly dynamic — content from the geographically nearest edge server to the user.

Chain-of-Thought (CoT) is a prompting technique that asks the model to spell out its reasoning in intermediate steps — boosting accuracy on multi-step tasks.

ChatGPT Search is OpenAI's web search integrated into ChatGPT, synthesizing answers from real-time web sources. Conversational search instead of a link list — directly in the ChatGPT interface.

Chroma (ChromaDB) is an open-source vector database for storing and searching embeddings. It is widely used in RAG systems to retrieve relevant text passages via similarity search and feed them to language models.

Chunking is the splitting of longer texts into smaller, self-contained pieces — the first step of any RAG pipeline because embeddings and retrieval work at the chunk level.

The churn rate indicates what proportion of customers is lost within a given period. It is usually calculated as a percentage per period and is a central metric for customer retention and customer value.

Citation optimization covers measures that deliberately shape content so it appears as a cited source in AI answers — for instance in Perplexity or Google AI Overviews. The goal is the numbered citation, not the classic ranking.

Claude Code is Anthropic's official AI coding tool — available as a terminal CLI, VS Code extension, JetBrains plugin and web app, with a strong focus on tool use and configurable workflows.

Claude Mythos (April 2026) is Anthropic's most powerful model to date — a new tier above the Opus line, distributed primarily through the Glasswing program to cybersecurity defenders.

Claude Opus 4.5 is Anthropic's top-tier model from November 2025 — the direct predecessor of Opus 4.6, bringing "Infinite Chats" and substantially reduced token usage.

US law from 2018 (Clarifying Lawful Overseas Use of Data Act). US authorities can compel US providers to hand over data even when it is stored outside the US — for example in the EU. It conflicts with the GDPR.

CLS measures unexpected layout shifts during page load. A Core Web Vital — "good" is ≤ 0.1 (unitless score).

SEA strategy of bidding on search terms containing competitor brand names to capture market share and switching-prone users.

Computer Use is a capability of modern AI models that lets them operate a computer like a human — see the screen, move the mouse, use the keyboard — to perform tasks across arbitrary applications without an API.

Constitutional AI is a training method developed by Anthropic in which an AI model critiques and revises its own behaviour against an explicit list of principles (a "constitution") — rather than relying solely on human ratings.

Content recency as a ranking factor — primarily relevant for time-sensitive topics (news, tech reviews, regulatory changes). Rooted in the "Query Deserves Freshness" concept inside the Google algorithm.

Content gap analysis identifies topics and keywords that competitors rank for but your site doesn't — or doesn't cover well enough. Delivers a content roadmap based on concrete SERP gaps.

Central overview page for a topic area that bundles many detail articles — a variant of the topic cluster model, often with a stronger focus on navigation and filtering rather than a linear pillar page.

Context engineering is the discipline of deliberately curating and maintaining everything inside the LLM context window — system prompt, tool definitions, RAG hits, memory, conversation history.

Context Precision is a RAG evaluation metric (0–1) measuring how well the retriever ranks relevant contexts above irrelevant ones — the focus is ranking quality, not raw hit count.

Context Recall is a RAG evaluation metric (0–1) measuring how completely the retriever supplied the facts needed for the gold answer — every missing fact lowers the score.

The context window is the maximum number of tokens a language model can process at once — input and output combined.

Contextual Retrieval is an Anthropic RAG technique that prepends an LLM-generated context sentence to every chunk before indexing — sharply raising retrieval quality.

Continue.dev is an open-source, model-agnostic AI coding extension for VS Code and JetBrains — configured via YAML, working with commercial as well as local models.

Continued pretraining is the further training of an already pretrained language model on large amounts of domain text — before any classic fine-tuning begins.

Conventional Commits is a convention for commit messages with a fixed format — type, optional scope and description — as the basis for automated changelogs and semantic versioning.

Conversational marketing is a dialogue-based form of customer engagement that responds to prospects in real time via chat, messaging and AI chatbots. The goal is to guide users along the customer journey through conversation rather than rigid forms.

An advertiser-defined user action that counts as a successful outcome of an ad — purchase, lead, call, newsletter signup. The driver of Smart Bidding and the basis for ROAS, CPA, etc.

The configuration object in Google Ads that fully defines a single conversion: name, source, value, counting rule, conversion window, and whether it counts as primary or secondary.

Advertising KPI: share of clicks (or sessions) that result in a conversion. Central efficiency metric for landing page and offer quality.

The measurement setup that lets Google Ads recognize when a click led to a defined conversion. Typically implemented via the Google tag or Google Tag Manager.

A major, publicly announced Google update to the core algorithm — typically broad and topic-agnostic, causing significant visibility shifts across many domains.

Core Web Vitals are three Google metrics (LCP, INP, CLS) measuring loading speed, interactivity and visual stability of a page — part of the Page Experience signals.

Cosine similarity measures how similar two vectors are by the angle between them — the default metric in RAG retrieval for comparing query embeddings against the embeddings stored in a vector index.

Advertising KPI: average cost per conversion. Central efficiency metric for lead gen and performance marketing, and the steering input for tCPA bidding.

Advertising KPI: cost per click. In Google Ads, the primary billing model for Search and the key short-term efficiency lever for bidding.

Advertising KPI: cost per generated lead. Standard efficiency metric for B2B and service marketing where the conversion is an inquiry rather than a purchase.

Advertising KPI: cost per thousand impressions. Standard billing model for Display, Video, and awareness campaigns where visibility is the goal.

Advertising KPI: cost per order. Standard efficiency metric in e-commerce, where an order is the typical conversion.

Advertising KPI: cost per video view. Standard billing for YouTube In-Stream skippable ads — paid only if the user watches at least 30 seconds (or finishes shorter videos).

Crawl budget is the number of URLs Googlebot crawls on a website within a given time frame — determined by server load and the perceived popularity of the content.

The speed at which a search-engine crawler issues requests to a site — measured in requests per second or per day and automatically adjusted to server response time.

An automated program that systematically visits web pages, reads their content, and follows links — the foundation for search-engine indices, monitoring tools, and AI training data.

CrewAI is an open-source framework for orchestrating multiple autonomous AI agents. It assigns agents roles, goals and tools and has them collaborate in teams ("crews") to solve complex tasks through coordinated, multi-step workflows.

The critical rendering path is the sequence the browser runs to turn HTML/CSS/JS into pixels parsing, render tree, layout, paint. Optimizing it shortens first paint and LCP.

CrUX is Google's public field-data dataset of real performance values from Chrome users (opt-in). It's the source of Core Web Vitals scoring in Search Console and PageSpeed Insights — not synthetic.

Advertising KPI: share of impressions that result in a click. Indicator of ad relevance and a component of the Google Ads Quality Score.

Cursor is an AI-first code editor — a fork of Visual Studio Code with AI deeply wired into the editor itself.

Customer Acquisition Cost (CAC) is the average cost of winning one new customer. It is calculated as total marketing and sales spend divided by the number of new customers acquired in that period.

Customer Lifetime Value (CLV, also LTV) is the expected total value of a customer over the entire relationship — usually as contribution margin. Compared with acquisition cost (LTV:CAC ratio) it shows whether winning new customers is profitable.

Processing on behalf occurs when a service provider processes personal data for a controller, bound by instructions. The Data Processing Agreement (DPA) under Art. 28 GDPR governs the mandatory terms of this arrangement.

Data residency refers to the physical storage and processing location of data — the geographic region or country where the servers sit. It says nothing about which legal jurisdiction may access that data.

Datasets is an open-source Python library from Hugging Face for loading, processing and sharing machine-learning datasets. It uses Apache Arrow and memory-mapping to handle even very large amounts of data efficiently.

Three related concepts for digital audio levels: dBFS is the scale with 0 dBFS as the hard ceiling, headroom is the safety margin below the maximum, clipping is the hard cut-off when a signal exceeds it.

Janus-Pro-7B is DeepSeek's open-source multimodal model from January 2025 — unifying image understanding and image generation in a single 7B architecture, MIT-licensed.

DeepSeek V4 (April 2026) is the fourth generation of the Chinese open-weight MoE model — released as Pro (1.6T parameters) and Flash (284B) under MIT license, with a 1M-token context window.

Diffusers is an open-source Python library by Hugging Face for using and training diffusion models for image, video and audio generation. It is built on PyTorch and is structured into pipelines, models and schedulers.

Digital PR is the SEO-oriented form of PR work — editorial mentions and backlinks from journalistic sources via data-driven stories, expert commentary, or original research. Delivers high-quality, natural backlinks.

A Google Ads campaign type that serves image and video ads across the Google Display Network — millions of websites, apps, and Gmail.

Domain Rating (Ahrefs) and Domain Authority (Moz) are third-party metrics that normalize a domain's backlink profile on a 0–100 scale. Not a Google signal — correlated with rankings, not a direct ranking factor.

DPO is a fine-tuning method that aligns language models directly to human preference pairs — without a separate reward model and without reinforcement learning.

DSPy is an open-source Python framework from the Stanford NLP group that lets you program LLM applications instead of hand-crafting prompts. Tasks are described through declarative signatures and modules; optimizers automatically derive effective prompts and examples.

Privacy-focused search engine that draws results primarily from the Bing index and enriches them with its own sources (Wikipedia, Apple Maps, the DuckDuckBot crawler).

E-commerce SEO bundles SEO disciplines specific to online shops — category and product page optimization, facet crawling, Product schema, shopping feed, out-of-stock handling.

E-E-A-T stands for Experience, Expertise, Authoritativeness and Trust — four quality criteria from Google's Search Quality Rater Guidelines used to assess content and websites.

An embedding is a numeric representation (vector) of text, an image, or other data in which semantically similar content sits close together in space.

Ensemble means combining several deliberately varied LLM runs or models whose findings complement each other. Multi-model orchestration drives these runs via orchestrators with sub-agents, so the union of results is larger than any single run.

Entity SEO establishes a brand, person or topic as an unambiguous entity in knowledge graphs such as Google's Knowledge Graph. The goal is for search and AI systems to identify, classify and connect the entity with facts reliably.

Regulation (EU) 2024/1689 is the EU's first comprehensive law on artificial intelligence. It regulates AI systems by risk in four tiers, from prohibited practices to minimal risk, and sets obligations for providers and deployers.

Evals are systematic tests for LLM applications — fixed test cases, automatic scoring of the responses, result as a score. The foundation for objectively comparing prompt or model changes.

The narrowest Google Ads match type. Triggers only when the query carries the same meaning as the keyword. Close variants (plurals, typos, acronyms) are included.

Faceted navigation refers to filter and sort URLs in shops and listings (color, size, price). It quickly generates millions of crawl URLs — a crawl-budget killer without robots.txt, noindex, or parameter handling.

Faithfulness is a RAG evaluation metric (0–1) measuring how many claims in a response can actually be derived from the retrieved contexts — a direct hallucination indicator.

FCP measures when the browser renders the first visible DOM element (text, image, SVG). "Good" at ≤ 1.8 s. The earliest signal that the page is responding — not a Core Web Vital, but an important PageSpeed metric.

Few-shot prompting is a technique that includes a handful of input/output examples in the prompt so the model picks up the desired format and style.

Field data are real performance values from actual users (CrUX, RUM). Lab data are synthetic measurements under controlled conditions (Lighthouse, WebPageTest). Google ranks by field data — lab data only hints at trends.

Continued training of a pretrained model on a smaller, specialized dataset to adapt style, format or domain knowledge in a targeted way.

Flowise is an open-source low-code tool for building LLM applications and AI agents visually via drag-and-drop. It is built on the LangChain ecosystem and connects models, data sources and tools as nodes.

Function calling is a language model's ability to produce a structured function invocation instead of a text reply — the technical foundation for tool use and AI agents.

Strategic model that splits SEA activity by funnel stage: Top (reach), Middle (consideration), Bottom (conversion).

GDPR-compliant LLM use describes the data-protection-compliant use of large language models such as ChatGPT, Claude or Gemini in a company — with a legal basis, data processing agreement, controlled input of personal data, EU hosting and training opt-out.

Gemini 3 Pro (November 2025) was Google's third flagship model with a 1M-token context window and native multimodality — now superseded by Gemini 3.1 Pro.

Gemini 3.1 Pro (February 2026) is Google's current flagship — 1M-token context, 64K output tokens, doubled reasoning performance over Gemini 3 Pro.

Gemma 4 is Google DeepMind's open model family from April 2026 — four sizes (E2B to 31B), 256k context, multimodal, Apache 2.0 licensed.

Three-way segmentation of SEA keywords: generic terms, your own brand terms, competitor brand terms — used for steering and reporting.

GEO is the discipline of shaping content for visibility in AI answer engines — AI Overviews, Perplexity, ChatGPT Search, Claude. The goal is not the classic SERP click, but appearing as a cited source inside the generated answer.

GGUF (GPT-Generated Unified Format) is the standard file format for quantised LLMs on the llama.cpp engine — a single `.gguf` file holds weights, tokenizer and metadata together.

Git is a distributed version control system that records changes to files as a sequence of commits — a baseline requirement for most AI coding workflows.

A branch is an independent line of development in a Git repository — a moving pointer to a commit, allowing parallel work without conflicts with the main code.

Git worktree allows you to have multiple branches of a repo checked out simultaneously in different directories — without stashing or constant branch-switching.

GitHub is the largest hosting platform for Git repositories — with pull requests, code review, issues and CI/CD via GitHub Actions as the standard toolkit for software development.

GitHub Copilot is GitHub's AI coding assistant — an extension for common editors offering tab autocomplete, chat and multi-file edits, with GPT and Claude models selectable.

GitLab is a Git hosting platform with an integrated DevOps stack — repository, CI/CD, issue tracking, container registry and security scans in one tool, optionally self-hosted.

GitUp is an open-source Git client for macOS with a visual live graph of the repository history and snapshot-based rollback.

GLM-4.6 is Zhipu AI's open-weight MoE model from September 2025 — 355B parameters (32B active), 200k context, MIT-licensed, with a focus on coding performance.

GLM-5.1 is Z.ai's open-source flagship from April 2026 — 744B parameters (40B active), 200k context, designed for agentic coding with up to 8 hours of autonomous runtime.

Google's advertising platform for paid ads across Search, Display, Shopping, YouTube, and apps. The central tool for SEA and performance marketing across Google properties and the Display Network.

The current generation of Google's web analytics tool — an event-based data model rather than the session-based Universal Analytics. The sole standard since July 2023, Universal Analytics has been shut down.

GA4 as the analytics backend for SEA — delivers conversions, audiences, and engagement signals that flow into Google Ads for reporting, Smart Bidding, and remarketing.

Google's free tool for managing a business listing on Google Search and Maps. Core asset in Local SEO. Formerly "Google My Business" (GMB).

Google Maps SEO is the practice of optimising a business's visibility in Google Maps and the local search results. The main levers are a well-maintained Google Business Profile, accurate categories, reviews and local relevance signals.

Free Google tool for monitoring a site's visibility in the Google index — performance reports, index coverage, Core Web Vitals data, and URL inspection. A mandatory SEO tool.

Google's tag management system. Tracking codes, pixels, and event logic are deployed centrally via a container snippet — no code changes required.

Googlebot is Google's web crawler that discovers, downloads, and processes pages for the search index. Operates in two main variants — Smartphone and Desktop.

GPT-4o (May 2024) was OpenAI's first natively multimodal model — text, audio and image in a single model, with a real-time voice mode and substantially lower cost than GPT-4 Turbo.

GPT-4o mini (July 2024) is OpenAI's small, low-cost variant of GPT-4o — positioned as the successor to GPT-3.5 Turbo, clearly stronger at a much lower price per token.

GPT-5.2 is OpenAI's model generation from December 2025 — three variants (instant, thinking, Pro), a 400k context window and new state-of-the-art on professional knowledge work.

GPT-5.3 is an iteration of OpenAI's GPT-5 line — the Instant variant launched on March 3, 2026 with a focus on factual accuracy, shorter preambles and a more natural conversational flow.

GPT-5.3-Codex is OpenAI's specialized agentic coding model (February 2026), the successor to GPT-5.2-Codex — 25% faster, with substantial gains on Terminal-Bench and OSWorld.

GPT-5.3-Codex-Spark is OpenAI's coding model from February 2026 — a smaller, real-time-optimized variant of GPT-5.3-Codex delivering over 1000 tokens per second on Cerebras hardware.

GPT-5.4 is OpenAI's reasoning model from March 5, 2026 — the first mainline model with the coding capabilities of GPT-5.3-codex, native computer-use controls and 1M-token context.

GPT-5.5 is OpenAI's flagship from April 23, 2026 — positioned as its "smartest and most intuitive model" and pitched as a step toward AI that works across tools on its own.

GPT4All is an open-source desktop application by Nomic AI for running large language models locally and offline on your own computer. It works without a GPU and without a cloud connection, so data never leaves the device.

GraphRAG is a RAG variant that uses a knowledge graph in addition to vector search — answers come from linked entities and relationship paths, not just similar text.

Grok 4 is xAI's reasoning model from July 2025 — text and image input, 256k context, trained via reinforcement learning on the 200,000-GPU Colossus cluster.

A reference taken to be correct, against which model outputs are measured. In AI code analysis it is a piece of code with deliberately injected, known defects, so you can measure how many real problems a model actually finds.

Guardrails are protective layers around a language model that inspect inputs and outputs to catch unwanted behaviour — off-topic answers, PII leaks, unsafe actions.

A hallucination is a language-model response that sounds plausible but is factually wrong or fabricated — typically caused by statistical guessing without a factual anchor.

Haystack is an open-source AI orchestration framework from deepset for building production-ready LLM applications. Modular components — retrievers, generators, routers, tools — are assembled into explicit pipelines for RAG, agents and semantic search.

HCU recovery is the process of regaining visibility after being hit by a Helpful Content Update — via content audit, cleaning up weak material, and strengthening verifiable expertise.

Google's machine-learning system that classifies content site-wide as "people-first" or not — launched as a standalone classifier, integrated into core ranking in March 2024.

The Helpful Content Update is a series of Google algorithm updates starting August 2022 that prioritize "human, people-first" content and demote thin, search-engine-first material.

Hreflang is an HTML/header attribute telling search engines which language and regional versions of a page exist — crucial for multilingual sites.

Modern versions of the HTTP protocol that deliver parallel requests over a single connection (HTTP/2) or via the faster QUIC transport (HTTP/3).

Hugging Face is the central platform of the open-source AI community — a hub for models, datasets, and demos, and the maintainer of widely used libraries like Transformers, Diffusers, and PEFT.

Central platform for openly shared AI models, datasets and demos — a kind of GitHub for machine-learning artifacts.

Hybrid search combines lexical full-text search (e.g. BM25) with semantic vector search and unites the strengths of both — precise on keywords and robust on meaning.

HyDE (Hypothetical Document Embeddings) is a RAG technique where an LLM first drafts a fake answer to a query — and then uses that answer's embedding to find real documents.

The process by which static HTML rendered on the server is retroactively given JavaScript interactivity in the browser — turning "dead" markup into a reactive app.

Using modern image formats (WebP, AVIF) plus appropriate sizes and compression to dramatically shrink image payloads without visible quality loss.

Indexing is the process by which a search engine analyzes a crawled URL and adds it to its search index. Crawling ≠ indexing — they are separate steps.

Indirect prompt injection is an attack that hides malicious instructions inside external content — web pages, documents, emails — that an LLM then processes and unintentionally executes.

Inference is the act of running an already-trained model to turn an input into a response — production use, not training.

INP measures how quickly a page responds to user interactions. A Core Web Vital since March 2024 (replacing FID) — "good" threshold is ≤ 200 ms.

Tokens you send to an AI model in an API call — your prompt, the context, attached documents. Billed separately from output tokens and usually much cheaper.

Instruction tuning is a variant of supervised fine-tuning in which a language model is trained on a broad mix of instruction-response pairs — so it reliably follows instructions across arbitrary tasks.

Links between pages on the same domain. Internal linking shapes crawl paths, distributes link equity, and signals topical relevance to search engines.

International SEO controls which language and country version of a site ranks in which market. hreflang is the central signal — bidirectional, with an x-default fallback and consistent canonicals per locale.

A jailbreak deliberately bypasses a language model's safety and behaviour rules through cleverly crafted prompts, making the model produce content it would normally refuse.

Jan is an open-source desktop application that lets you run large language models locally and fully offline on your own computer. It uses the llama.cpp engine and is positioned as a privacy-friendly ChatGPT alternative.

Googlebot renders JavaScript pages in two stages — first an HTML crawl, then rendering via the Web Rendering Service (Chromium-based). The delay can hold up indexing of client-rendered content.

A discipline within technical SEO that ensures JavaScript-rendered content is correctly crawled, rendered, and indexed by search engines.

Paid, ad-free search engine that aggregates results from its own crawlers (Teclis, TinyGem) and API sources — targeted at users willing to pay for ad-free, quality-filtered search.

A search term an advertiser targets to trigger ads. In Google Ads a keyword is not a literal user query but a trigger rule paired with a Match Type — Phrase, Exact, or Broad.

A keyword glossary for Speech-to-Text is a short, curated list of project-specific terms, brand and product names handed to an STT model as a context hint before transcription — typically through Whisper's `initial_prompt`.

Free research tool inside Google Ads. Returns search volume, competition, CPC estimates, and new keyword ideas — the data backbone for campaign planning and SEO research.

Systematic identification of the search terms a target audience uses — basis for content planning, site architecture, and prioritization. Delivers keywords with search volume, competition, and search intent.

Kimi K2 is Moonshot AI's open-weight MoE model series with 1T total parameters and 32B active parameters, trained on 15.5T tokens — focused on agentic and coding tasks.

Kimi K2.6 is Moonshot AI's open-weight agent model from April 2026 — 1T parameters (32B active), 262k context, natively multimodal, with agent swarms of up to 300 sub-agents.

Knowledge distillation is a training technique in which a small student model learns from the behaviour of a large teacher model — aiming for comparable quality at significantly lower resource cost.

Advertising KPI: share of ad spend in attributed revenue, expressed as a percentage. Mathematical inverse of ROAS, common in DACH e-commerce reporting.

The KV cache (key-value cache) stores the already computed key and value vectors of a language model's attention layers during text generation. This avoids recomputing the entire context for each new token and significantly speeds up inference.

One of the three Quality Score components in Google Ads. Rates the destination page's content, usability, and speed. Scale: above/average/below average.

Landing-page SEO is the SEO strategy for small sites with ~5–15 pages. Instead of content scaling, it emphasizes maximum quality per page, clear conversion focus, and local/technical presence.

Open-source framework that chains LLMs with data sources, tools and memory into applications — arguably the best-known pipeline library for AI apps.

LangGraph is an open-source framework from LangChain for stateful, graph-based LLM workflows — nodes are functions or agents, edges define the control flow including loops and branches.

Attribution model that gives full conversion credit to the last clicked touchpoint before the conversion — historically the default in Google Ads.

Lazy loading is a technique where images, iframes, or JavaScript modules load only when actually needed — usually when scrolled into the viewport.

LCP measures when the largest visible element in the viewport finishes loading. Part of Core Web Vitals — the "good" threshold is ≤ 2.5 s.

A Google Ads asset/ad extension that opens a pre-filled lead form directly inside the ad — no redirect to a landing page.

Lighthouse is Google's open-source audit tool that synthetically measures a page's performance, accessibility, best practices, and SEO in a controlled environment.

An off-page SEO discipline focused on actively acquiring inbound links. Goal — strengthen authority, visibility, and ranking through external trust signals.

Link velocity describes the growth rate of inbound backlinks over time. Sudden spikes without an apparent trigger may be flagged as unnatural — organic velocity feels gradual and steady.

Llama 3.2 is Meta's model generation from September 2024 — the first with vision support (11B/90B) and lightweight edge models (1B/3B) for mobile and on-device use.

Llama 4 Maverick (April 2025) is Meta's large Mixture-of-Experts variant in the Llama 4 lineup — 17B active out of 400B parameters across 128 experts, 1M-token context.

Llama 4 Scout (April 2025) is Meta's smaller Mixture-of-Experts variant in the Llama 4 lineup — 17B active out of 109B parameters, 10M-token context, multimodal.

Llama-3-SauerkrautLM-70b is a German-language DPO fine-tune of Meta Llama 3 70B, developed by VAGOsolutions and Hyperspace.ai.

llama.cpp is an open-source C++ library for efficient LLM inference on CPU and GPU — the engine many local tools like Ollama and LM Studio are built on.

Llamafile is a Mozilla project that packages a local language model as a single executable file — model weights and inference engine in one. That file runs on several operating systems without installation.

Data framework for LLM applications, specialized in indexing, enriching and retrieving private data — at the core of many RAG architectures.

LLM stands for Large Language Model — a neural network trained on large volumes of text to interpret and generate natural language.

Crawlers from AI providers that gather web content for training or live answers. Examples — GPTBot, ClaudeBot, PerplexityBot, Google-Extended.

LLM-as-a-Judge is the practice of using a language model as an evaluator — it compares responses from other models or scores outputs against given criteria.

llms.txt is a proposed text file in a website's root directory meant to give AI models and LLM crawlers curated, important content in a simple form. As of 2026 it is an unofficial proposal with low practical adoption.

LM Studio is a desktop application for running LLMs locally with a graphical interface — including a model browser, chat interface and OpenAI-compatible server.

Local citations are mentions of a business with its name, address and phone number (NAP) in business directories, map and review services. Consistent citations are considered a supporting signal for local ranking.

Local keywords are search terms with a spatial reference — for example through place or district names or through "near me". They signal a local search intent and are the basis for optimising for regional visibility.

Local landing pages are dedicated subpages of a website optimized for a specific location. They bundle location-specific content such as address, services and local references in order to be found more easily in regional search.

Local link building is the deliberate acquisition of regionally relevant backlinks — for example from associations, local press or partners. Such links strengthen a website's regional authority and support its ranking in local search.

The Local Pack (or Map Pack) is the block in Google's results showing a map and usually three local matches for location-based searches. It is ranked by relevance, distance and prominence — fed mainly from Google Business Profiles.

Share of local search queries answered directly in the SERP — via map pack, knowledge panel or Google Business Profile — without a click to any website. The conversion happens inside Google, not on your site.

Log file analysis evaluates server logs to see which URLs Googlebot actually crawls, how often, and with which status code. Reveals real crawl behavior — unlike Search Console (aggregated) or synthetic crawlers.

Search terms with low volume, high specificity, and usually clear intent. Three or more words, often phrased as a question. Low competition, high conversion.

Low-Rank Adaptation — parameter-efficient fine-tuning that trains only small additional matrices instead of all model weights.

LSI and semantic keywords are thematically related terms around a main keyword. Modern SEO focus isn't on "LSI" in the technical sense but on full topical coverage of a subject.

Make (formerly Integromat) is a cloud-based no-code workflow automation platform. Through a visual interface, you connect apps and APIs into multi-step workflows ("scenarios") without programming — for example to pass data between services.

A manual cost-per-click bidding strategy in Google Ads. Advertisers set bids per keyword or ad group themselves — Google does not automatically adjust them based on conversion likelihood.

Corporate discipline that creates demand, wins customers and maintains relationships — from brand building and content through paid channels and SEO to CRM and loyalty.

The marketing funnel is a model that describes the path of people from first awareness of a brand to purchase and beyond in phases: awareness, consideration, conversion and retention. It serves to plan cross-channel activities.

Marketing mix modeling (MMM) is a statistical method that estimates the contribution of individual marketing channels and factors to business results. It uses aggregated data without user tracking and is seeing a renaissance driven by privacy and AI.

The rule that defines how closely a booked keyword must align with a user query to trigger an ad. Active types in Google Ads: Broad, Phrase, Exact.

An automated Google Ads bidding strategy. Google sets bids so that the daily budget yields as many clicks as possible — without considering conversions.

MCP — Model Context Protocol — is an open standard from Anthropic for connecting AI models to external data sources and tools through a unified interface.

The meta description is an HTML tag in a page's head that contains a short summary of the content — Google often uses it as the snippet text shown in the SERP.

Meta-prompting is the technique of having an LLM write or improve a prompt for a task rather than formulating it yourself — the prompt becomes the output.

Milvus is an open-source, distributed vector database for fast similarity search over large sets of embeddings. It is licensed under Apache 2.0 within the LF AI & Data Foundation; the main developer is the company Zilliz.

MiniMax is an AI lab founded in Shanghai in 2021 whose open-weight M model family (LLMs) targets very long context windows and low running costs via sparse attention.

Mistral Large 3 (December 2025) is Mistral's open-weight MoE flagship — 675B parameters, 256K-token context, natively multimodal, Apache 2.0.

Mistral Small 4 is Mistral AI's open-weight MoE model from March 2026 — merging reasoning, vision and coding into one model, 119B parameters (6.5B active), Apache 2.0.

An indexing mode where Google primarily uses the mobile version of a page for crawling, indexing, and ranking. The default for all sites since July 2024.

A model card is a standardised datasheet for an AI model — documenting purpose, training data, performance, limitations and ethical considerations in one place.

MQL (Marketing Qualified Lead) and SQL (Sales Qualified Lead) are qualification stages of a lead in the funnel. An MQL shows interest through its behaviour; an SQL is assessed by sales as ready to buy and suitable for handover.

NAP consistency means that a business's Name, Address and Phone number are identical across all online directories, profiles and platforms. It is a recognised factor for local search rankings.

An exclusion keyword in Google Ads. Prevents an ad from serving when the user query contains the excluded term. Negative keywords also support match types.

A reusable collection of negative keywords in Google Ads, maintained at the account level and applied to multiple campaigns. Cuts maintenance overhead in large accounts.

News SEO is the SEO discipline for news-oriented sites — speed, freshness, news sitemap, NewsArticle schema, and inclusion in Google News and Top Stories are the levers.

Robots directives that control whether a page gets indexed or its links followed. Set via the meta robots tag or the X-Robots-Tag HTTP header.

Ollama is a tool for running language models locally — Llama, Mistral or Qwen — operated via a CLI, with an OpenAI-compatible HTTP server for applications.

Open WebUI is an open-source chat interface for local and remote LLMs — a ChatGPT-like browser UX that runs as a Docker container and talks to Ollama, OpenAI-compatible APIs and more.

Claude Opus 4.6 is Anthropic's flagship model from February 2026 — top tier of the Claude 4 family with a 1M-token context window, stronger coding and new effort controls.

Claude Opus 4.7 is Anthropic's top-tier model from April 2026 — successor to Opus 4.6 with high-resolution image processing, a new xhigh effort level and task budgets for long agentic runs.

ORPO is a training method that unifies supervised fine-tuning and preference optimisation in a single step — leaner than the classic SFT-plus-RLHF pipeline.

Tokens an AI model produces as its response. Billed separately and usually three to five times more expensive than input tokens because the model has to actively generate them.

PageSpeed Insights is Google's free web tool that combines Lighthouse lab scores with CrUX field data from real Chrome users and outputs concrete optimization suggestions.

Panda (2011) and Penguin (2012) were two formative Google algorithm updates. Panda targeted thin, low-quality content, Penguin manipulative link building. Both are now part of the core algorithm.

Path dependency describes how, during LLM generation, each early token shapes everything that follows — the first verbalized finding steers the entire remaining analysis in one direction.

PEFT bundles methods that adapt a pretrained LLM by training only a small fraction of its weights — typically under 1%, instead of all billions of parameters.

Fully automated Google Ads campaign type. A single campaign serves across all inventory — Search, Shopping, Display, YouTube, Discover, Gmail, Maps — driven by Google's AI toward a conversion goal.

A global setting in Claude Code that controls how tool calls are handled: Plan (planning only), Default (asks for confirmation), Accept Edits (files free), Bypass (allow everything).

Perplexity is an AI-based answer engine that synthesizes web search results with a language model and delivers answers with inline source citations — an alternative to Google, focused on conversational research.

pgvector is an open-source PostgreSQL extension for vector search — adding embedding columns, distance functions, and ANN indexes (HNSW, IVFFlat) directly to an existing Postgres database.

The middle Google Ads match type. Triggers when the query includes the keyword's meaning. Since 2021 the old quotation-mark rules have been replaced by meaning-based logic.

Pinecone is a managed vector database service for RAG and semantic-search workloads — it stores embeddings, answers nearest-neighbour queries, and scales without the user operating the index themselves.

Advertising KPI: ratio of gross profit to ad spend. Margin-aware alternative to ROAS, especially relevant for broad assortments and sale dynamics.

Predictive analytics in marketing uses historical data and machine learning to forecast future customer behaviour. Typical applications are lead scoring (conversion probability), churn prediction (attrition) and forecasting of revenue or campaign performance.

Prefix tuning is a PEFT method that trains a sequence of learned "virtual tokens" prepended to the input rather than any model weights — the base model itself stays frozen.

Resource hints in the HTML head that tell the browser to fetch resources earlier, set up connections in advance, or accelerate DNS lookups.

Classification of conversion actions in Google Ads: Primary drives Smart Bidding, Secondary is observed only and does not feed into bids.

The Product Reviews Update was a series of Google algorithm updates starting April 2021 that distinguished product reviews with real, in-depth hands-on research from shallow affiliate lists.

Scaling strategy where hundreds or hundreds of thousands of pages are generated from structured data via templates — typical for comparison, directory, and travel sites with combinatorial patterns.

Prompt caching is an API feature in which a provider stores recurring prompt prefixes — making subsequent requests cheaper and faster because the cached portion is not reprocessed.

Prompt engineering is the discipline of steering AI models through carefully crafted input prompts — closer to precise writing than to classic programming.

Prompt injection is an attack that hides instructions inside an LLM's input data to make the model ignore or subvert its original instructions.

Prompt leaking is an attack that tricks an LLM into revealing its hidden system prompt or other confidential context contents — a special case of prompt injection.

A prompt library is a central, organized collection of reusable and versioned prompts within a team. It provides proven inputs for AI models so they do not have to be reworded every time they are used.

A prompt template is a reusable prompt skeleton with placeholders that get filled with concrete values at runtime — the basis for reproducible LLM calls in applications.

A pull request is a proposal to merge changes from one branch into another — the standard workflow for code review and collaborative work on a repository.

Qdrant is an open-source vector database written in Rust — it stores embeddings for semantic search, RAG and recommendation systems and runs locally, on your own cluster or as a managed cloud.

Extension of LoRA that quantizes the base model to 4 bits — allowing even very large LLMs to be fine-tuned on a single GPU.

Internal evaluation document Google uses to instruct external quality raters on how to assess search results — basis for the E-E-A-T concept and the source of many algorithm updates.

A Google Ads diagnostic per keyword on a 1–10 scale, indicating how well the ad and landing page match the search query. Components: expected CTR, ad relevance, landing page experience.

Quantization reduces the numeric precision of model weights — e.g. from 16-bit float to 4-bit integer — making language models small and fast enough for consumer hardware, with manageable quality loss.

Query expansion enriches the original query in a RAG pipeline with synonyms, reformulations or hypothetical answer texts in order to lift retrieval recall.

Query types classify search queries by user intent. Four common classes — informational, navigational, transactional, commercial investigation — decide which content type ranks.

Qwen 3.5 (February 2026) is Alibaba's third-generation open model family — from 0.8B to 397B parameters, MoE flagship with 1M-token context, multimodal.

RAG (Retrieval-Augmented Generation) connects a language model to an external knowledge source — relevant passages are retrieved and passed to the model alongside the question.

RAG-friendly content is structured so AI systems can easily split it into meaningful sections, extract individual passages and cite them correctly. Clear structure, unambiguous statements and self-contained sections raise the chance of appearing in AI answers.

RAGAS is an open-source framework for automated evaluation of RAG and agent pipelines — with built-in metrics like faithfulness, context precision, and answer relevancy.

Provider-enforced cap on requests or tokens per time window — it protects infrastructure and ensures fair usage across customers.

ReAct is a prompting pattern that has an LLM alternate between reasoning (thoughts) and action (tool calls) — the foundation behind many agent implementations.

Reasoning Effort is a control parameter on modern reasoning models that sets how much internal step-by-step thinking (thinking tokens) a model spends before answering — higher levels raise quality but also latency and cost.

Recall (hit rate, sensitivity) measures the share of all actually existing relevant cases that a system finds. Formula: relevant cases found divided by all relevant cases that truly exist. Requires a known ground truth.

Re-engaging users who already interacted with a website, app, or ad — across Display, Search (RLSA), Video, or Shopping campaigns.

Render-blocking resources are CSS and JavaScript files the browser must load and parse before it can render. They delay FCP and LCP — minimize them with async/defer, inline critical CSS, and remove unused CSS.

External reputation of a brand or author via reviews, mentions, and third-party sources — explicitly part of the trust assessment in the Quality Rater Guidelines, especially for YMYL topics.

Reranking reorders an already-retrieved list of hits with a more accurate model — typically a cross-encoder that scores query and each candidate together, instead of just comparing vector distances.

The standard ad type in Google Ads Search campaigns. Up to 15 headlines and 4 descriptions are uploaded; Google assembles them into an ad matching the search query.

Review management covers actively gaining, responding to and analysing customer reviews. Reviews serve as a trust signal and feed into local search as a ranking factor — volume, recency and response behaviour matter.

Reviews updates are Google algorithm updates that rank review content by depth, hands-on experience, and added value — shallow affiliate lists get demoted.

Reinforcement Learning from Human Feedback — training procedure that aligns models with helpful and safe behavior using human preference comparisons.

Advertising KPI: ratio of conversion revenue to ad spend. Central steering metric in e-commerce and the input for tROAS bidding.

robots.txt is a text file at a domain's root telling search-engine crawlers which URLs may be crawled and which must not.

Top-level profitability KPI: profit relative to the investment made. Goes beyond pure ad spend and includes all relevant cost components.

Role prompting assigns the LLM a concrete role or persona ("you are an experienced tax lawyer …") to steer style, vocabulary and depth of answers.

S-RAG (Search-optimized RAG) is boostN.ai's take on Retrieval-Augmented Generation. It extends classic RAG with search-engine principles: relevance ranking, recency decay and typed links between chunks for deterministic follow-up retrieval.

Safetensors is a file format developed by Hugging Face for storing model weights — faster, safer and more language-agnostic than the older PyTorch `.pt`/`.bin` format.

Sampling is the weighted drawing of the next token from a language model's probability distribution — governed by temperature, top-p and top-k. It produces the variability between two runs.

SauerkrautLM is a German-language LLM family from the German startup VAGOsolutions — fine-tunes based on Llama, Qwen, Mistral, and other open architectures.

The goal a user is pursuing with a search query. Four common types — informational, navigational, transactional, commercial investigation.

OpenAI's search-engine prototype, originally launched as a standalone product — now largely folded into ChatGPT Search. Delivers AI-generated answers with source citations.

Self-Consistency is a prompting technique where the same question is answered multiple times with chain-of-thought, and the most frequent answer is picked as the final result.

Self-Refine is a prompting technique in which a model critiques its own output and improves it across several iterations — without human feedback.

Semantic search retrieves content by meaning rather than exact word match — query and documents are compared as embeddings, so synonyms and paraphrases also hit.

SERP stands for Search Engine Results Page — the result page Google, Bing and other search engines return for a given query.

SERP features are all display elements in search results beyond the classic blue links — featured snippets, People Also Ask, knowledge panel, local pack, image, video, and shopping boxes.

SFT is the supervised post-training of a pretrained language model on a dataset of input-output pairs — the step that turns a base model into a usable assistant.

Google's AI-generated SERP answers. SGE was the beta name (2023), AI Overviews the production name (May 2024), AI Mode the dedicated chat tab (2025).

Share of model is a metric that measures how often a brand is mentioned in the answers of AI language models, relative to the mentions of all competitors in the same category. It is considered the AI counterpart to the classic share of voice.

A Google Ads campaign type for product ads with image, price, and merchant name. Data comes from the Google Merchant Center feed; targeting is product-attribute-based, not keyword-based.

A Google Ads asset/ad extension that displays additional clickable links under the main ad — pointing to subpages, categories, or promotions.

A sitemap is an XML file listing all relevant URLs of a website — it helps search engines crawl and index large sites systematically.

A structural pattern where each ad group contains exactly one keyword. Historically popular for maximum ad-copy control — largely obsolete given Close Variants and Smart Bidding.

Umbrella term for the conversion- and value-based automated bid strategies in Google Ads: tCPA, tROAS, Maximize Conversions, Maximize Conversion Value. Driven by ML and real-time auction signals.

A soft 404 is a page with no real content (out-of-stock product, empty search) that incorrectly returns HTTP 200 instead of 404/410. Google usually detects it and reports it in Search Console.

Hugging Face Spaces is a hosting platform for deploying and sharing interactive machine learning applications without your own server infrastructure. Apps become usable directly in the browser, for example via Gradio, Streamlit, Docker or static HTML.

Spam updates are regular Google algorithm adjustments that detect manipulative SEO tactics and remove or sharply demote affected pages in search results.

Speech-to-Text refers to the automatic conversion of spoken language into text by an AI model. In an AI workflow, STT replaces the keyboard as the input channel — what matters is model size and domain vocabulary.

Spotlighting is a defense technique against prompt injection in which untrusted inputs are marked so the model treats them as data — not as instructions.

Two opposing render strategies for web pages — either the server generates ready-made HTML (SSR) or the browser assembles the page via JavaScript (CSR).

A structural pattern where each ad group covers a tightly scoped theme with several keyword variants. The modern alternative to SKAG — compatible with Smart Bidding.

Two ways to advertise products on Google: Standard Shopping gives granular control and transparency; Performance Max maximizes reach across all inventory via AI-driven optimization.

Tools that check source code against formal rules without executing it — based on a syntax tree (AST) and control-flow graph. They work deterministically and exhaustively, unlike probabilistic LLMs.

Stop sequences are strings that make an LLM end token generation as soon as they appear. They constrain the output deliberately — to prevent role switches, format markers or overly long answers.

Streaming means transmitting an LLM's response token by token in real time — the user sees the text word by word rather than only after the full generation completes.

Structured data is machine-readable markup in HTML — typically using the Schema.org vocabulary in JSON-LD format — that explains a page's content precisely to search engines.

Structured data for LLMs are machine-readable annotations (usually Schema.org as JSON-LD) that describe content explicitly — author, date, entity, rating. This lets AI systems classify, summarise and cite the content more reliably.

Structured output is a language model's ability to deliver responses in a defined schema (typically JSON), reliable enough to be consumed directly by downstream code.

A Google Ads asset/ad extension that shows a structured list with a fixed header (e.g. "Brands", "Models", "Services") below the ad.

Synthetic data is artificially generated training or evaluation data — typically produced by a language model itself to extend datasets without needing real-world sources.

Synthetic personas are target-audience profiles simulated with AI language models to mimic real respondents. Marketing uses them to test messages, concepts or products quickly against a modelled audience — as a complement to, not a replacement for, real research.

A system prompt is the instruction that shapes a language model's behavior across an entire conversation — in contrast to the user prompt, which varies per message.

A Google Ads Smart Bidding strategy that adjusts bids so the average cost per acquisition lands on the target value the advertiser sets.

A Google Ads Smart Bidding strategy that adjusts bids so the return on ad spend lands on the target value the advertiser sets. Requires tracked conversion values.

Temperature is a sampling parameter that controls how deterministic or creative a language model's responses are — low values stay conservative, high values produce more variety.

TensorRT-LLM is an open-source library from NVIDIA that optimizes LLM inference on NVIDIA GPUs. It compiles models into highly optimized runtime engines and offers in-flight batching, a paged KV cache and quantization for high throughput at low latency.

Text-Generation-WebUI (oobabooga) is an open-source web interface for running LLMs locally. It bundles several inference backends, chat and completion modes and an OpenAI-compatible API under one common UI.

TGI (Text Generation Inference) is an open-source toolkit from Hugging Face for serving large language models efficiently. It provides a production-ready inference server with continuous batching, token streaming and optimised attention methods.

The thinking budget (token budget) is the maximum number of internal reasoning tokens a reasoning model may spend on "thinking" per request — controlling the trade-off between answer depth and cost or latency.

Tiered model line-up from a provider — small fast variants (Mini/Flash/Haiku) at a fraction of the price of the big frontier models. Also: volume tiers with quantity discounts.

The title tag is the HTML element in the head defining a page's title — it appears as the clickable headline in the SERP and in the browser tab.

A token is the smallest unit a language model works with internally — usually a sub-word fragment, occasionally a single character.

A tokenizer is the program that splits text into tokens before a language model can process it — it determines how many tokens a piece of text costs.

An invocation of a tool by an AI model during a conversation — such as reading a file, running a bash command, fetching from the web or calling an MCP tool. The foundation of agentic workflows.

Top-p (nucleus sampling) and top-k are sampling strategies for LLMs that decide which token candidates to draw from at each step — together with temperature they control the creativity of the output.

A content architecture built from one broad pillar page and several specific cluster pages, tied together by topic and internal linking.

Toxic backlinks are unnatural, spammy, or paid inbound links that can trigger a manual action or algorithmic devaluation. The Disavow Tool tells Google to ignore those links when evaluating your site.

TPM (tokens per minute) and RPM (requests per minute) are the two common units in which AI API providers express their rate limits — TPM caps the token volume, RPM caps the number of requests per minute.

Open-source Python library from Hugging Face that gives unified access to thousands of pretrained models across text, vision and audio.

Tree of Thoughts (ToT) is a prompting technique in which an LLM explores multiple solution paths as a branching tree, scores them and follows only the promising branches — a generalisation of Chain-of-Thought.

Trustworthiness signals are visible and technical proofs that a website and its content can be trusted — author credentials, imprint, HTTPS, transparent sources, external reputation signals.

TTFB measures the time from request start to the first byte received. It sums DNS, TCP, TLS, and server processing. Rule of thumb — good < 200 ms, problematic > 600 ms; directly affects LCP and FCP.

A vector database stores embeddings as high-dimensional vectors and searches them by semantic similarity — the core infrastructure for RAG, semantic search, and recommendation systems.

Aggregated visibility score for a domain in the Google index, computed by SEO suites (Sistrix, Searchmetrics, Semrush) from their own crawls — useful for trend comparison, but not a Google signal.

vLLM is an open-source high-throughput inference server for LLMs — known for PagedAttention, continuous batching, and an OpenAI-compatible API endpoint, widely used in production self-hosting setups.

VRAM (Video RAM) is the dedicated memory on a GPU — the single most important hardware figure for running language models locally, because ideally the whole model and its context must fit inside it.

Weaviate is an open-source vector database written in Go — it stores objects together with their embeddings, combines semantic and keyword (BM25) search and offers built-in vectorization, RAG and multi-tenancy.

Whisper is OpenAI's open speech-to-text model from 2022 — a multilingual encoder-decoder transformer that turns audio into text and ships in several sizes under an MIT license.

The `initial_prompt` parameter lets you pass Whisper a keyword list or sample sentence before transcription. The model treats this text as context and recognises the included terms far more reliably.

whisper.cpp is an open-source C++ port of OpenAI's Whisper speech recognition — a compact native binary for local transcription on CPU, CUDA, Metal or Vulkan, with no Python dependency.

Windsurf is an AI IDE built on the VS Code base with a focus on agentic coding workflows — the Cascade agent plans and executes multi-step changes autonomously.

HTTP header that sets robots directives (e.g. noindex, nofollow) at the server level — also works for non-HTML resources like PDFs, images, or JSON responses.

XML tags are named brackets like <context>...</context> used to clearly separate sections of a prompt. They structure the input, split instruction from data and make the model's output more predictable.

YMYL refers to content that can impact users' financial well-being, health, safety, or happiness. Google applies particularly strict quality and trust standards to these topics.

Zapier AI bundles the AI features of the automation platform Zapier — including Copilot for building workflows by language, autonomous agents, chatbots and an MCP server that exposes thousands of Zapier actions to external language models.

Zero-shot describes solving a task with a language model without providing any examples — purely via the task description in the prompt.