← back

Concepts

A2UI (Agent-to-User Interface)volume_upvideo_camera_back

An open web standard that lets AI agents generate rich, interactive interfaces as declarative data, rendered through a client app's own trusted components instead of raw code.

AEO (Answer Engine Optimization)volume_upvideo_camera_back

Answer Engine Optimization. A name for optimising to appear in AI answer engines.

AI Agentvolume_upvideo_camera_back

An AI system that plans and executes multi-step tasks autonomously — researching, calling tools, and acting toward a goal rather than just answering.

AI Assistantvolume_upvideo_camera_back

A conversational AI like ChatGPT, Gemini or Claude used for short, varied tasks — usually one question and one answer, not search-style queries.

AI Content Detectionvolume_upvideo_camera_back

Classifying whether text was written by AI; as models improve, detection needs constant fine-tuning and hybrid deep-learning-plus-heuristic methods.

AI Crawlersvolume_upvideo_camera_back

The bots AI systems use — training-data scrapers versus agentic assistants — that site owners must decide whether to allow or block.

AI Modevolume_upvideo_camera_back

Google's AI-powered search experience — essentially Gemini with search, calculator, time, location and Python tools — answering queries conversationally.

AI Overviewsvolume_upvideo_camera_back

Google's AI-generated answer summaries atop search results, which leaked signals suggest are built on its Dialogflow agentic framework.

AI Rankvolume_upvideo_camera_back

Our AI visibility and rank-tracking framework that probes language models to map how they perceive and associate brands.

AI SEOvolume_upvideo_camera_back

A common name for AI-centric SEO: the practice of working toward AI Visibility.

AI Visibilityvolume_upvideo_camera_back

The broad outcome of AI-centric SEO: being seen, cited, and recommended by AI systems when they answer questions.

AIO (AI Optimization)volume_upvideo_camera_back

AI Optimization. Another name for the practice of improving AI Visibility.

AP2 (Agent Payments Protocol)volume_upvideo_camera_back

A payment-agnostic open standard that lets autonomous AI agents authorize and complete transactions securely via cryptographically signed "mandates," removing manual checkout.

Activation Steeringvolume_upvideo_camera_back

Injecting a computed direction vector into a model's residual stream at inference time to bias its output toward a target behaviour — without any weight updates or retraining.

Agentic Harnessvolume_upvideo_camera_back

The orchestration layer around a language model that equips it with tools, memory, and control flow — turning a text predictor into an agent that can plan, act, and loop.

Annotated Page Content (APC)volume_upvideo_camera_back

The structured, machine-readable representation Chrome builds from a page's rendering tree when a tab is shared with Gemini. It captures every visible element — text, links, images, forms, tables — as a tree of content nodes, each tagged with geometry, styling, interaction data, and a unique node ID

Associative Embeddednessvolume_upvideo_camera_back

A measure of how deeply a brand is woven into a model's memory, from how often and how early the model recalls it across many runs.

Attentionvolume_upvideo_camera_back

The transformer mechanism that lets a model weigh how much every token relates to every other, in parallel — the core of modern LLMs.

Binary Classificationvolume_upvideo_camera_back

A model task with exactly two possible output labels — yes/no, spam/not-spam, AI/human — the simplest and most common form of supervised classification.

Binary Vector Embeddingsvolume_upvideo_camera_back

Compressing float embeddings to one bit per dimension, reducing storage by ~32× with surprisingly small quality loss — useful when speed and scale matter more than marginal precision.

BlockRankvolume_upvideo_camera_back

A method that makes LLM in-context ranking scale to large candidate sets by exploiting the block-sparse structure of attention.

Brand Association Networkvolume_upvideo_camera_back

The graph of entities, concepts, sentiments, and competitor brands a language model associates with a given brand, surfaced through systematic probing and tracked over time.

Brand Authorityvolume_upvideo_camera_back

How much cognitive space a brand occupies in a model's memory — how readily and prominently it recalls the brand unprompted.

CAPSvolume_upvideo_camera_back

Our proposed Content Attribution Payment Scheme — a framework to fairly attribute and pay creators when AI systems use their content.

Checkpointvolume_upvideo_camera_back

A saved snapshot of a model's weights at a point during training, enabling recovery, comparison across training stages, and selection of the best-performing version.

Chrome History Embeddingsvolume_upvideo_camera_back

Chrome's local system that turns pages you visit into vectors, enabling natural-language search of your own browsing history on-device.

Citation Miningvolume_upvideo_camera_back

Systematically running many prompts across AI platforms to record which domains and URLs get cited — mapping the AI citation landscape for a topic.

Class Imbalancevolume_upvideo_camera_back

When one class in a training dataset vastly outnumbers another, causing a naive model to ignore the minority class — requires deliberate handling through weighted loss, resampling, or specialised loss functions.

Classifiervolume_upvideo_camera_back

A model that assigns one or more labels to an input — the fundamental pattern behind query intent models, spam detectors, sentiment models, and content detection.

Content Fetchervolume_upvideo_camera_back

One of AI Mode's two page-reading tools — it retrieves a batch of structured sources at once, versus browsing a single URL.

Content Nodevolume_upvideo_camera_back

A single element in Chrome's page representation. Each node carries a type (one of 21, from headings and links to form controls and dialogs), bounding box coordinates, text styling, accessibility metadata, and an ID that the AI uses to reference or interact with that specific element.

Content Substance Classificationvolume_upvideo_camera_back

Detecting whether content actually says something or just sounds like it does — separating genuine substance from eloquent fluff.

Context Engineeringvolume_upvideo_camera_back

Deliberately constructing the semantic environment of a prompt to activate specific representational circuits within a model — moving beyond keyword targeting toward architecture-aware content design.

Context Windowvolume_upvideo_camera_back

The maximum number of tokens a model can read at once — the total working memory available for a prompt, grounding documents, conversation history, and the generated response combined.

Cosine Similarityvolume_upvideo_camera_back

A measure of how alike two vectors are by the angle between them; for normalised vectors it's identical to a dot product.

Document Groundingvolume_upvideo_camera_back

Grounding an AI answer in the content of a specific web page you supply, rather than in results from an open web search.

DocumentChunkervolume_upvideo_camera_back

Chrome's internal Blink algorithm that recursively walks a page's DOM layout tree to group text nodes into semantically cohesive passages of up to 200 words.

DomDistillervolume_upvideo_camera_back

Chrome's Reader Mode engine, whose content-extraction heuristics are a public proxy for how machines separate main content from boilerplate.

Dynamic Retrievalvolume_upvideo_camera_back

Google's mechanism for deciding, per query, whether Gemini runs a live search — governed by a default grounding threshold of 0.3.

Early Stoppingvolume_upvideo_camera_back

Halting training when validation performance stops improving rather than running for a fixed number of epochs — the primary defence against overfitting in practice.

EmbeddingGemmavolume_upvideo_camera_back

Google's compact open embedding model, a miniaturised relative of Gemini, that turns text into meaning-carrying vectors for search understanding.

Epochvolume_upvideo_camera_back

One complete pass through the entire training dataset — the basic unit of training progress, used to track how many times the model has seen each example.

F1 Scorevolume_upvideo_camera_back

The harmonic mean of precision and recall — the standard single-number summary of classifier performance when both false positives and false negatives matter.

Fine-tuningvolume_upvideo_camera_back

Continuing training of a pre-trained model on a smaller, task-specific dataset to specialise its behaviour — the standard way DEJAN adapts base models into production classifiers.

Focal Lossvolume_upvideo_camera_back

A modified cross-entropy loss that down-weights easy, well-classified examples so training focuses on hard cases — the standard solution when positive examples are rare.

Frequencyvolume_upvideo_camera_back

How often your citations and mentions occur across AI queries and over time.

GEO (Generative Engine Optimization)volume_upvideo_camera_back

Generative Engine Optimization. A name for optimising to appear in generative AI engines.

Generate-then-Groundvolume_upvideo_camera_back

Google's approach of drafting an answer first, then searching for sources to attribute it — which we argue is more fragile than retrieving first.

Generative self-retrievalvolume_upvideo_camera_back

A reasoning model surfacing facts from its own parameters by generating them as text, where those generated facts then condition and improve its final answer.

Groundingvolume_upvideo_camera_back

The process by which an AI system answers with current information: it runs a search, retrieves pages, reads short extracts, and writes its answer from them rather than from memory.

Grounding Biasvolume_upvideo_camera_back

A form of secondary bias: how much an AI model defers to, or discounts, retrieved sources once they are grounded into the context, rather than leaning on what it already believed.

Grounding Chunkvolume_upvideo_camera_back

The slice of a page's content an AI system actually uses to support a sentence, drawn from a fixed per-query word budget.

Grounding Classifiervolume_upvideo_camera_back

Our model that predicts whether a query "deserves grounding" — whether an AI system will run a live web search to answer it.

Grounding Snippetvolume_upvideo_camera_back

A short, extractive selection of sentences pulled from a page and supplied to an AI model as the evidence it grounds an answer on; the atomic unit of visibility in AI search.

Hyperparametersvolume_upvideo_camera_back

Settings configured before training begins that control how a model learns — learning rate, batch size, epoch count, dropout rate — as distinct from the weights the model learns itself.

Implicit Queriesvolume_upvideo_camera_back

Searches an AI system formulates and runs on your behalf without you typing them, based on your on-device context and profile.

In-Context Rankingvolume_upvideo_camera_back

Re-ranking candidate documents for a query by feeding them all into an LLM's context and letting it pick the most relevant.

Inferencevolume_upvideo_camera_back

Running a trained model on new inputs to produce predictions — the production mode of a model, as distinct from the training phase where weights are learned.

LLM Search Volume Predictionvolume_upvideo_camera_back

Using a language model to estimate a query's monthly search volume — directionally useful for sizing topics, but not precise.

Language Modelvolume_upvideo_camera_back

A model trained to understand and generate text by learning the statistical patterns of language — the foundational class of model underpinning search, AI assistants, and every DEJAN classifier.

Large Language Modelvolume_upvideo_camera_back

A language model at sufficient scale — billions of parameters, trained on web-scale text — to exhibit emergent capabilities like reasoning, instruction following, and open-ended generation. Gemini, GPT, Claude, and Llama are all LLMs.

Latent Entitiesvolume_upvideo_camera_back

A critical piece of information a title deliberately withholds — the subject, reason, process or outcome — to force a click.

LoRAvolume_upvideo_camera_back

Low-Rank Adaptation — a parameter-efficient fine-tuning technique that trains small adapter matrices instead of updating a model's full weights, dramatically reducing memory and compute requirements.

LongFormContextResultvolume_upvideo_camera_back

The data wrapper Chrome uses to deliver extracted page content to Gemini's context window. It separates external source material from conversation, ensuring the model grounds its answers in the page rather than its training data. Each block carries a session-wide index used for citation mapping.

Loss Functionvolume_upvideo_camera_back

The mathematical measure of how wrong a model's predictions are — the quantity training minimises, and the choice of which shapes everything about how the model learns.

MUVERAvolume_upvideo_camera_back

A Google method for multi-vector retrieval that compresses many per-token vectors into a single fixed-dimensional encoding for fast search.

Manifold Hypothesisvolume_upvideo_camera_back

The principle that natural data lies on a low-dimensional manifold within high-dimensional space — the theoretical foundation for why vector embeddings capture meaning rather than noise.

Matryoshka Representation Learningvolume_upvideo_camera_back

A training technique that nests coarse-to-fine information inside one embedding, so it can be truncated to fewer dimensions with little loss.

Mechanistic Interpretabilityvolume_upvideo_camera_back

Techniques for tracing what happens inside a model — logging activations and patching them — to see how and where it forms an output.

Mixture of Expertsvolume_upvideo_camera_back

An architecture where only a subset of specialised sub-networks (experts) activate for each token, allowing a model to have enormous total capacity while spending only a fraction of that compute on any single input.

Model Context Protocolvolume_upvideo_camera_back

An open protocol that gives AI models a standard way to connect to external tools and data sources; WebMCP brings the idea to websites.

Model Distillationvolume_upvideo_camera_back

Training a smaller student model to replicate the behaviour of a larger teacher model, producing compact models that punch above their parameter count.

Model Probingvolume_upvideo_camera_back

Systematically querying a language model with structured prompts to map its latent biases, brand associations, and internal knowledge graphs without relying on traditional rank trackers.

Multi-label Classificationvolume_upvideo_camera_back

A classification task where each input can be assigned more than one label simultaneously — a query can be both informational and local; a sentence can be both positive and sarcastic.

Multimodal Modelvolume_upvideo_camera_back

A model that processes and reasons across more than one type of input — text, images, audio, video — within a single unified architecture rather than routing each modality to a separate specialist.

Node IDvolume_upvideo_camera_back

A sequential identifier assigned to each content node during extraction via depth-first traversal. These IDs appear in the structured Markdown output as {#ID} references, enabling the AI to target specific elements on the page for reading, citing, or interaction.

Number of Citationsvolume_upvideo_camera_back

The absolute count of times your pages are cited in AI answers.

Number of Mentionsvolume_upvideo_camera_back

The absolute count of times your brand is named in AI answers.

On-Device Modelsvolume_upvideo_camera_back

Machine-learning models that run locally in software like Chrome — fast, private, and shaping how your content is read before it reaches a server.

Open Knowledge Formatvolume_upvideo_camera_back

An open, vendor-neutral format for packaging the knowledge AI systems need, as plain markdown files any model or agent can read.

Otsu's Thresholding (for Queries)volume_upvideo_camera_back

An unsupervised method borrowed from computer vision that sets optimal per-label confidence cut-offs in query classification by maximizing the variance between positive and negative score clusters.

Overfittingvolume_upvideo_camera_back

When a model memorises training examples rather than learning generalisable patterns — it scores well on training data but poorly on new inputs it hasn't seen.

Pre-trainingvolume_upvideo_camera_back

Training a model from scratch on a large general corpus to build broad language understanding before any task-specific work begins.

Precisionvolume_upvideo_camera_back

Of all the times a model predicted positive, the fraction it was actually right — a measure of how trustworthy positive predictions are.

Primary Biasvolume_upvideo_camera_back

An AI model's ungrounded confidence in an entity, formed during training and present before any retrieval; in AI search it is the largest single factor in whether a source is selected.

Query Classificationvolume_upvideo_camera_back

Assigning search queries to labels — like intent categories — so they can be routed, mapped to content, and reported at scale.

Query Deserves Groundingvolume_upvideo_camera_back

Google's internal decision about whether a query needs a live web search; below its threshold, the model answers from memory.

Query Fanoutvolume_upvideo_camera_back

The step where an AI system splits one prompt into several single-intent sub-queries, each retrieving its own sources; a page can be grounded for one angle of a question and absent for another.

Query Well-formednessvolume_upvideo_camera_back

The property of a search query being grammatically natural and unambiguous — a signal used to distinguish full-sentence questions from fragmentary keyword strings, with implications for query expansion and intent classification.

Rankvolume_upvideo_camera_back

How prominently your citations and mentions appear within an AI answer.

Recallvolume_upvideo_camera_back

Of all the actual positive cases in the data, the fraction the model correctly identified — a measure of how much the model misses.

Relevance Engineeringvolume_upvideo_camera_back

Engineering a page's semantic relevance to a query with embeddings and vector math, treating visibility as an engineering problem rather than keyword optimization.

Retrieval-Augmented Generationvolume_upvideo_camera_back

A framework that retrieves relevant passages first, then feeds query plus evidence to the model so it answers from real text.

Reverse Promptingvolume_upvideo_camera_back

Reconstructing the most likely prompt that generated a piece of AI output — used for detection, content auditing, and understanding how models respond to specific framings.

Rufusvolume_upvideo_camera_back

Amazon's AI shopping assistant — a multi-component RAG system that plans a query, retrieves across Amazon's own sources, and streams an answer.

Secondary Biasvolume_upvideo_camera_back

The post-retrieval layer of selection: how an AI model treats, weights, and is swayed by content once it has been retrieved. Unlike primary bias, it is addressable now.

Selection Ratevolume_upvideo_camera_back

The rate at which an AI system chooses a given source from the available grounding candidates when composing an answer; the AI-search equivalent of click-through rate.

Selection Rate Optimizationvolume_upvideo_camera_back

The AI-search counterpart to CTR optimization: improving how often AI systems select your content as the grounding source for their generated answers.

Semantic Compressionvolume_upvideo_camera_back

Writing dense, self-contained passages so that when an AI system extracts an isolated grounding chunk, the core context like the brand or product name travels with it.

Share of Citationsvolume_upvideo_camera_back

The proportion of cited sources in AI answers that are yours.

Share of Mentionsvolume_upvideo_camera_back

The proportion of brand mentions in AI answers that are yours.

Share of Voicevolume_upvideo_camera_back

Your overall slice of the AI answer space for a topic, relative to competitors.

Site Engagement Scorevolume_upvideo_camera_back

Chrome's internal 0–100 per-domain user engagement metric, built from navigation events, input interactions, and media consumption, decaying over time when a site is not visited.

Site Search in AI Modevolume_upvideo_camera_back

Vertex AI's website-search capability that lets an owner run an AI-Mode-style search over their own content, confirming the Prepare→Retrieve→Signal→Serve flow.

Synthetic Training Datavolume_upvideo_camera_back

AI-generated datasets used to train or fine-tune language models; in AI SEO, a potential early-stage strategy for influencing which brands and associations future models learn.

Temperaturevolume_upvideo_camera_back

A setting that rescales next-word probabilities to control randomness — low values make output focused, high values more diverse.

Tokenvolume_upvideo_camera_back

The atomic unit a language model reads and generates — a subword chunk, not a whole word — that determines how text is measured, priced, and processed.

Token Probabilityvolume_upvideo_camera_back

The likelihood a model assigns to each candidate next token; low probabilities flag where the model is uncertain about your brand or topic.

Tokenizationvolume_upvideo_camera_back

Breaking text into subword tokens a model can process; each token carries a frequency bias from how often it appeared in training.

Tokenizervolume_upvideo_camera_back

The component that converts raw text into the sequence of tokens a model actually processes — fixed at pre-training time, and different for every model family.

Top-p Samplingvolume_upvideo_camera_back

A decoding method that samples the next word from the smallest set of top candidates whose probabilities sum to a threshold p.

Training Datavolume_upvideo_camera_back

The labelled or unlabelled examples a model learns from — the single biggest determinant of what a model knows, what it believes, and which brands and entities it associates with which topics.

Transfer Learningvolume_upvideo_camera_back

Applying knowledge learned on one task or dataset to a different but related task — the principle that makes fine-tuning a pre-trained model far more efficient than training from scratch.

Tree Walkervolume_upvideo_camera_back

Our brand-analysis tool that reveals how Gemini talks about a brand by walking the probabilistic paths of the model's own language.

URL Context Toolvolume_upvideo_camera_back

Google's Gemini capability that fetches and reads the content of a specific supplied URL, enabling document grounding.

Vector Embeddingvolume_upvideo_camera_back

A dense numeric representation of text that captures meaning and intent, so machines can compare content by similarity rather than keywords.

Vector Embedding Optimizationvolume_upvideo_camera_back

Choosing embedding dimension and quantisation to balance speed, storage and accuracy — often gaining efficiency with little loss.

Vector Searchvolume_upvideo_camera_back

Finding the most relevant content by comparing meaning-carrying vectors, ranking by nearest neighbour rather than keyword overlap.

Web Search Groundingvolume_upvideo_camera_back

The process where an AI model runs a live web search, reads the results, and weaves some of those pages into its answer.

WebMCPvolume_upvideo_camera_back

A proposed web standard that lets sites expose structured tools to AI agents, so agents invoke defined actions instead of screen-scraping the page.