Quick answer

An AI glossary is a plain-English reference to the vocabulary of artificial intelligence, from foundational ideas like machine learning and neural networks to newer terms like agentic AI, RAG and answer engine optimization. This guide defines 162 AI terms in 10 categories. Browse them A to Z or by category.

You can’t evaluate an AI vendor, brief an agency or write an AI policy if your team doesn’t share the same vocabulary.

88%
of organizations use AI in at least one business function
McKinsey, 2025
62%
are at least experimenting with AI agents
McKinsey, 2025
51%
have experienced at least one negative consequence from AI, most often inaccuracy
McKinsey, 2025
280x+
drop in the cost of GPT-3.5-level AI performance, late 2022 to late 2024
Stanford AI Index, 2025

AI adoption is now the norm, but shared understanding has not kept pace. McKinsey’s 2025 State of AI survey found that while most organizations use AI somewhere, only about a third have begun scaling it across the enterprise. Much of that gap is practical: teams struggle to compare tools, set guardrails and measure results when “AI” means something different to everyone in the room.

This glossary closes that gap. We used public reference glossaries from Stanford HAI and Hyland as a starting point, rewrote every definition in plain English, and added the marketing and search terms businesses actually run into, such as answer engine optimization, AI Overviews and intent data. Where a term affects how you plan, target or measure campaigns, you’ll find an In marketing note. Looking for SEO terms like canonical URL or crawl budget? See our SEO glossary of terms.

Table of contents

  1. Browse the AI glossary: A to Z or by category
  2. Why AI terminology matters for marketers
  3. Core AI concepts
  4. How machines learn: training methods and techniques
  5. AI models and architecture
  6. Generative AI and LLM terms
  7. AI agents and automation
  8. Data terms behind every AI system
  9. Measuring and running AI models
  10. AI ethics, safety and governance
  11. AI terms for marketing and search
  12. AI infrastructure and roles
  13. How UnReal Web Marketing puts these AI terms to work
  14. Frequently asked questions

Browse the AI glossary: A to Z or by category

Find any of the 162 terms two ways: scan the alphabetical index, or explore terms by category to see how related concepts fit together. Every term links to its full definition.

AI terms A to Z

A

B

C

D

E

F

G

H

I

K

L

M

O

P

R

S

T

U

V

W

Z

AI terms by category

No terms match your search. Try a shorter word, such as “agent” or “data.”

Why AI terminology matters for marketers

Vocabulary is a buying skill. When a platform claims to be “AI-powered,” the difference between a rules engine, a predictive model, a generative model and an autonomous agent determines what it can do, what data it needs and what can go wrong.

Vocabulary is also a governance skill. Terms like hallucination, human-in-the-loop and prompt injection describe real risks. Teams that can name them can design controls around them.

And vocabulary is now a visibility skill. Buyers increasingly research through AI assistants and AI-generated search answers. Understanding how LLMs retrieve, chunk and cite content is the foundation of answer engine optimization.

The 6 AI terms to know first

TermWhat it meansWhy marketers care
Generative AIAI that creates new text, images, audio, video or codeDrafts emails, ads and landing pages at scale
Large language model (LLM)The text model behind chat assistants and AI searchDecides how AI search answers describe your brand
Machine learningSystems that learn patterns from dataPowers lead scoring, lookalikes and send-time optimization
AI agentSoftware that plans and takes steps toward a goalCan run multi-step campaign tasks with human approval
RAGA model looks up trusted sources before answeringKeeps AI answers grounded in your real data
AEOOptimizing content to be cited in AI answersVisibility in Google AI features, ChatGPT and Perplexity

Core AI concepts

These are the foundational ideas every other term in this glossary builds on. If your team agrees on these definitions, vendor demos and agency proposals get much easier to evaluate.

Artificial intelligence (AI)

The branch of computer science focused on building systems that can do work that normally requires human judgment, such as understanding language, recognizing patterns, making predictions or choosing an action. The term traces back to a 1955 research proposal that led to the 1956 Dartmouth summer workshop.

In marketing: AI is an umbrella label. When a vendor says a product “uses AI,” ask which kind: a predictive model, a generative model, a rules engine or an agent.

Narrow AI

AI designed to do one well-defined job, such as filtering spam, transcribing calls or scoring leads. Nearly every AI product in commercial use today is narrow AI, even when it feels general-purpose.

Artificial general intelligence (AGI)

A hypothetical form of AI that could learn and reason across almost any subject at or beyond human level, rather than excelling at a single task. Researchers disagree on how to define it, how to measure it and when, or whether, it will arrive.

Traditional AI

Earlier, more rule-driven approaches to AI that analyze or classify information inside clearly defined boundaries. Examples include fraud-detection rules, recommendation engines and diagnostic support tools. Traditional AI predicts or decides; it does not create new content.

In marketing: Many proven marketing workflows, like segmentation rules and suppression logic, are traditional AI and still outperform generative tools for compliance-sensitive decisions.

Generative AI (GenAI)

AI that produces new material, including text, images, audio, video or code, by learning statistical patterns from very large datasets and then generating plausible new output in response to a prompt.

In marketing: Useful for first drafts of emails, ad variations, social posts and landing-page copy. Every output still needs a human check for accuracy, compliance and brand voice.

Machine learning (ML)

A subset of AI in which a system learns patterns from examples instead of following hand-written instructions. As it sees more relevant data, its predictions typically improve.

In marketing: Machine learning powers lead scoring, churn prediction, send-time optimization and lookalike audiences. Google also uses machine learning in its ranking systems; see RankBrain in our SEO glossary.

Deep learning (DL)

A branch of machine learning that uses neural networks with many stacked layers. Deep learning handles messy, high-volume inputs such as images, speech and free-form text, but it requires large amounts of data and computing power.

Algorithm

A defined, step-by-step set of instructions for solving a problem, much like a recipe. In machine learning, the algorithm is the learning procedure; the model is what that procedure produces.

Model

The trained result of a machine learning process: a mathematical representation that takes an input (a lead record, a sentence, an image) and returns an output (a score, a reply, a label).

Expert system

An early style of AI that captures a specialist’s knowledge as a large set of if-then rules so a computer can make decisions in a narrow field, such as equipment troubleshooting or credit approval.

Natural language processing (NLP)

The field of AI that lets computers read, interpret and produce human language. Search engines, chat assistants, translation tools and sentiment analysis all rely on NLP.

Natural language understanding (NLU)

The part of NLP concerned with meaning: working out intent, context and relationships in text so a system knows what a person is actually asking for.

Natural language generation (NLG)

The part of NLP that turns data or instructions into readable sentences, such as an automated performance summary or a drafted product description.

Automatic speech recognition (ASR)

Technology that converts spoken audio into written text. It powers meeting transcription, voice assistants and call analytics.

Computer vision

The field of AI that helps machines interpret images and video, including recognizing objects, reading text in photos and detecting faces or logos.

Robotics

The design and building of machines that carry out physical tasks. Modern robots often combine computer vision, sensors and machine learning to navigate and adapt.

Turing test

A thought experiment proposed by Alan Turing in 1950: if a person in conversation cannot reliably tell a machine from a human, the machine could be said to show intelligent behavior. Today it is viewed as a historical milestone rather than a rigorous benchmark.

Human-centered AI

An approach to designing AI that puts people’s needs, rights and well-being first, aiming to augment human work rather than simply replace it.

↑ Back to the glossary index

How machines learn: training methods and techniques

These terms describe how a model gets from raw data to useful predictions. You do not need to build models to benefit from knowing them, but they help you ask vendors the right questions about where results come from.

Training data

The examples a model studies to learn its task. The quality, size, freshness and representativeness of training data shape almost everything about how a model behaves.

In marketing: If a scoring model was trained on last year’s ideal customer profile, it will keep favoring last year’s buyers. Ask how often training data is refreshed.

Supervised learning

Training a model on labeled examples where the correct answer is already known, such as emails marked spam or not spam, so it can predict labels for new examples.

In marketing: Feeding a model historical leads tagged “closed-won” and “closed-lost” so it can score new leads is supervised learning.

Unsupervised learning

Training on data that has no labels, so the model has to discover structure on its own, for example grouping similar records into clusters.

In marketing: Commonly used to uncover customer segments you did not know existed.

Self-supervised learning

A method where the model creates its own learning tasks from raw data, such as hiding a word in a sentence and predicting it. This is how large language models learn from enormous amounts of unlabeled text.

Reinforcement learning (RL)

Learning through trial and error. The model takes actions, receives rewards or penalties, and gradually favors the actions that lead to better outcomes.

In marketing: Bid optimization and multi-armed bandit testing in ad platforms borrow from reinforcement learning.

Reinforcement learning from human feedback (RLHF)

A training step in which people rate or rank a model’s responses and the model is adjusted to prefer the answers humans judged more helpful, honest or safe. It is widely used to refine chat assistants.

Transfer learning

Reusing what a model learned on one task as the starting point for a related task, which saves time and data compared with training from scratch.

Fine-tuning

Continuing to train an already-trained model on a smaller, specialized dataset so it performs better in a specific domain, tone or task.

In marketing: A company might fine-tune a model on approved product copy so drafts match its terminology and style.

Few-shot learning

A model’s ability to pick up a new task after seeing only a handful of examples, often supplied directly in the prompt.

Zero-shot learning

A model’s ability to handle a task or category it was never explicitly shown during training, relying on general knowledge instead.

Contrastive learning

A training approach that teaches a model by comparing pairs of examples, pulling similar items closer together and pushing dissimilar items apart in its internal representation.

Federated learning

Training a shared model across many devices or organizations without moving their raw data to one place. Only model updates are shared, which can help with privacy.

Knowledge distillation

Training a smaller “student” model to imitate a larger “teacher” model, producing a faster, cheaper model that keeps much of the original’s ability.

Ensemble methods

Combining the predictions of several models so the group outperforms any single member. Random forests and gradient-boosted trees are well-known examples.

Decision tree

A model that reaches a prediction by asking a sequence of yes-or-no questions about the data, branching at each answer. Decision trees are easy to explain, which makes them popular in regulated settings.

K-nearest neighbors (k-NN)

A simple algorithm that classifies a new item by looking at the most similar items it already knows and going with the majority.

Bayesian network

A probabilistic model drawn as a graph that shows how variables influence one another, useful for reasoning about cause and effect under uncertainty.

Markov chain

A model of a sequence of events in which the probability of the next step depends only on the current state, not on the full history.

Dimensionality reduction

Techniques that simplify a dataset by reducing the number of variables while keeping the information that matters most, making patterns easier to see and models faster to train.

Loss function

A formula that measures how far a model’s predictions are from the correct answers. Training is the process of driving that number down.

Backpropagation

The core technique neural networks use to learn: errors at the output are traced backward through each layer so the network knows how to adjust its internal settings.

Optimization algorithm

The method used to adjust a model’s settings step by step to reduce error. Gradient descent is the most common example.

Hyperparameter

A setting chosen by people before training starts, such as learning rate or the number of layers, as opposed to parameters the model learns on its own.

↑ Back to the glossary index

AI models and architecture

This is the vocabulary behind the AI tools your team already uses. It explains why some models are faster, cheaper or better at certain jobs than others.

Neural network

A model made of layers of connected computing units, loosely inspired by neurons in the brain. Each connection carries a weight that is adjusted during training.

Parameter

An internal value a model learns during training. Model size is often described by parameter count, which can run into the hundreds of billions for large language models.

Weights

The numerical values that set the strength of the connections inside a neural network. Weights are a type of parameter, and together they hold what the model has learned.

Tensor

A container for numbers arranged in one or more dimensions, like a spreadsheet that can extend into extra dimensions. Tensors are the basic data structure AI hardware processes.

Convolutional neural network (CNN)

A type of neural network designed for grid-shaped data such as images, using small filters that scan for edges, shapes and textures.

Generative adversarial network (GAN)

A pair of neural networks trained against each other: a generator creates synthetic examples and a discriminator tries to spot the fakes, pushing both to improve until the output looks realistic.

Diffusion model

A generative model that learns to create images, audio or video by starting from random noise and removing it step by step until a coherent result appears. Many popular image generators use diffusion.

Transformer

The neural network architecture behind most modern language models. Transformers use attention to look at every part of an input at once, which lets them capture context across long passages.

Attention mechanism

The technique that lets a model weigh which parts of the input matter most for the output it is producing, for example linking a pronoun to the right noun several sentences earlier.

Large language model (LLM)

A transformer-based model trained on massive amounts of text so it can understand and generate language: answering questions, summarizing, drafting and translating.

In marketing: LLMs power chat assistants and AI search answers. How your content is structured affects whether LLMs cite it, which is the focus of answer engine optimization.

Small language model (SLM)

A compact language model with far fewer parameters than frontier LLMs. SLMs are cheaper and faster to run, and they can operate on a laptop or phone for focused tasks.

GPT (generative pre-trained transformer)

A family of transformer-based language models that are first pre-trained on large text collections and then adapted to generate text. The term is often used loosely for chat assistants in general.

Foundation model

A large model trained on broad data that can be adapted to many different tasks, from chat to classification to code, rather than being built for one purpose.

Frontier model

A highly capable, general-purpose model at the leading edge of what AI can currently do. Frontier models typically get the most scrutiny from researchers and regulators.

Multimodal AI

AI that can take in or produce more than one kind of data, such as text, images, audio and video, within a single system.

In marketing: A multimodal model can review a display ad image and its copy together and suggest edits to both.

Reasoning model

A model designed to work through complex problems in intermediate steps before answering, trading some speed for better accuracy on math, logic, planning and analysis.

Mixture of experts (MoE)

An architecture that splits a model into specialized sub-networks and activates only the most relevant ones for each input, improving efficiency without shrinking total capability.

Vision transformer (ViT)

A transformer adapted for images by cutting a picture into small patches and treating them like words in a sentence.

Latent space

The compressed internal representation a model builds of its data, where similar items sit close together. Generative models create new output by sampling from this space.

Open-source model

A model released with its code, weights and, ideally, training details available for anyone to inspect, use and modify under an open license.

Open-weight model

A model whose trained weights can be downloaded and run independently, even if its training data or full code is not published. Open-weight is not always the same as fully open source.

Closed model (proprietary model)

A model whose code and weights are kept private by its developer and accessed only through an app or paid API.

In marketing: Closed models are easy to start with; open-weight models give more control over data residency and cost. The right choice depends on what data you plan to send to the model.

↑ Back to the glossary index

Generative AI and LLM terms

These are the day-to-day working terms for anyone writing prompts, building content workflows or evaluating AI writing tools.

Prompt

The instruction, question or input you give a generative AI model. A prompt can include examples, background documents and formatting rules.

System prompt

Standing instructions set by the developer or administrator that shape how a model behaves across every conversation, such as tone, role and rules it must follow.

Prompt engineering

The practice of writing and refining prompts to get more accurate, relevant and consistent output from a model, for example by assigning a role, giving examples and specifying the format.

Context engineering

Designing everything a model sees when it works, not just the prompt: relevant documents, data, memory of past steps, tool access and business rules. It has become essential for reliable multi-step AI workflows.

In marketing: A content assistant that can pull your approved messaging, persona notes and product specs before drafting is the result of good context engineering.

Context window

The maximum amount of text (measured in tokens) a model can consider at one time, including the prompt, any attached material and its own response.

Token

The small unit of text a language model reads and writes. A token is often part of a word; roughly, 100 English tokens equal about 75 words. Usage and pricing for AI APIs are usually counted in tokens.

Tokenization

The process of breaking text into tokens so a model can process it numerically.

Inference

The moment a trained model is put to work on new input to produce a prediction or response. Training happens once; inference happens every time someone uses the model.

Temperature

A setting that controls how predictable or varied a model’s output is. Lower values produce more consistent answers; higher values produce more creative, less predictable ones.

Hallucination

When an AI model states something false or invented, such as a fake statistic or citation, with the same confidence as a fact.

In marketing: Hallucinated stats in published content damage credibility and search trust. Require a source for every number before anything goes live.

Grounding

Tying a model’s answers to specific, verifiable sources, such as your documents or a live database, so output reflects real information rather than guesswork.

Retrieval-augmented generation (RAG)

A technique that lets a model look up relevant information from an external source, such as a knowledge base or website, and use it when writing an answer. RAG reduces hallucinations and keeps answers current without retraining the model.

In marketing: AI search tools use retrieval to decide which web pages to read and cite. Clear, well-structured pages are easier for them to retrieve.

Embeddings

Lists of numbers that represent the meaning of a word, sentence, image or record, so items with similar meanings end up mathematically close together, even if they are worded differently.

Vector database

A database built to store embeddings and quickly find the items most similar in meaning to a query. Vector databases are a core component of most RAG systems.

Chunking

Splitting long documents into smaller sections before they are embedded and stored, so a retrieval system can find and pass along just the relevant passage.

In marketing: Pages organized with descriptive headings and short, self-contained answers tend to chunk cleanly, which helps AI search engines quote them.

Semantics

The meaning behind words and phrases. Semantic understanding lets AI treat synonyms and related ideas as connected.

Ontology

A formal map of the concepts in a domain and how they relate, for example how a product, its features and the industries that buy it connect to each other.

Knowledge graph

A network of entities (people, companies, products, topics) and the relationships between them, often built on an ontology. Search engines use knowledge graphs to understand how things connect.

Chatbot

Software that holds a conversation with people through text or voice. Older chatbots followed scripts; modern ones are usually powered by LLMs.

Conversational AI

The broader set of technologies, including NLP, speech recognition and LLMs, that allow software to hold natural, multi-turn conversations.

↑ Back to the glossary index

AI agents and automation

Agents are the fastest-moving part of the AI vocabulary. These terms help separate genuine autonomous systems from rebranded automation.

Automation

Technology that carries out tasks by following explicit instructions written by people. Automation is predictable and repeatable, but it does not decide what to do on its own.

In marketing: Classic marketing automation triggers a follow-up email when someone downloads a guide. Adding AI lets the system choose the content or timing.

AI agent

Software that pursues a goal by planning steps, using tools such as search, a CRM or a spreadsheet, and adjusting its approach based on results, with limited human direction.

Agentic AI

AI systems with a high degree of autonomy that can set sub-goals, plan, act across multiple applications and adapt over time to complete entire workflows rather than single tasks.

In marketing: An agentic workflow might research target accounts, draft personalized outreach and log activity in the CRM, pausing for human approval before anything is sent.

Multi-agent system

A setup in which several AI agents with different roles work together, such as one researching, one writing and one checking, to solve problems too complex for a single agent.

Tool use (function calling)

A model’s ability to call external software, such as a search engine, calculator, database or API, to get information or take action instead of relying only on what it learned in training.

Model Context Protocol (MCP)

An open standard that gives AI applications a consistent way to connect to outside data sources and tools, so agents can work with business systems without a custom integration for each one.

In marketing: MCP and similar standards make it easier to connect AI tools to the systems marketers already use, such as a CRM, analytics or a content management system.

Copilot

An AI assistant built into an existing application that helps users as they work, for example by drafting text, summarizing records or suggesting next steps, while the person stays in control.

Human-in-the-loop (HITL)

Designing an AI system so people review, correct or approve its work at key points. HITL is a core safeguard for anything customer-facing or high-stakes.

In marketing: Any AI that sends email, changes bids or edits customer records should have a human approval step until it has a proven track record.

AI workflow

The end-to-end process of building, deploying and maintaining an AI system, or a business process in which AI handles one or more steps.

Deep research

An AI capability that carries out extended, multi-step investigation, searching many sources, comparing findings and producing a structured report, rather than answering from a single lookup.

↑ Back to the glossary index

Is your business showing up in Google and AI search?

UnReal Web Marketing has been helping businesses win in search since 1997. Find out where you rank today, in traditional results and in AI-generated answers.

Get your FREE SEO auditFree consultation

Or call 845-201-5300

Data terms behind every AI system

AI is only as useful as the data it runs on. These terms describe the raw material, and they are where most marketing AI projects succeed or stall.

Structured data

Information stored in a fixed format with defined fields, such as a CRM table, a spreadsheet or a database of contacts with job title, company and industry columns.

Semi-structured data

Information with some organization but no rigid schema, such as emails, invoices, JSON files or web forms.

Unstructured data

Information with no predefined model: documents, call recordings, images, videos and free-text notes. Most enterprise data is unstructured, and generative AI has made it far more usable.

Big data

Datasets so large, fast-moving or varied that traditional tools struggle to process them, requiring specialized storage and analytics, often including AI.

Data mining

Using statistics and machine learning to uncover patterns, relationships and anomalies in large datasets.

Pattern recognition

The automatic detection of regularities in data, such as recurring purchase behavior, using statistical and machine learning methods.

Synthetic data

Artificially generated data that mimics the statistical properties of real data. It is used to train or test models when real data is scarce, sensitive or expensive.

Data augmentation

Expanding a training set by creating altered copies of existing examples, such as rotated images or reworded sentences, to help a model generalize.

Predictive analytics

Using historical data and models to forecast what is likely to happen next, such as which accounts will buy, churn or respond.

In marketing: Predictive analytics helps decide which keywords, audiences and campaigns deserve more budget.

Prescriptive analytics

Analytics that goes beyond forecasting to recommend the best action to reach a goal, given the options and constraints available.

↑ Back to the glossary index

Measuring and running AI models

These terms tell you whether a model actually works and keeps working. They are the questions to ask before you trust a vendor’s performance claims.

Accuracy

The share of all predictions a model gets right. Accuracy can mislead when one outcome is rare; a model that predicts “won’t buy” for every lead can look highly accurate while being useless.

Precision

Of everything the model flagged as positive, the share that truly was positive. High precision means few false alarms.

In marketing: In lead scoring, precision answers: of the leads marked “hot,” how many were actually sales-ready?

Recall

Of all the truly positive cases, the share the model managed to find. High recall means few missed opportunities.

F1 score

A single number that balances precision and recall, useful when you care about both avoiding false alarms and avoiding misses.

Overfitting

When a model learns its training data too closely, including random noise, so it performs well on familiar examples but poorly on new ones.

Underfitting

When a model is too simple to capture the real patterns in its data, so it performs poorly even on the examples it trained on.

Model drift

A gradual decline in a model’s performance as real-world conditions move away from the data it was trained on, for example after a market shift or a change in buyer behavior.

In marketing: Scoring models should be monitored and retrained on a schedule, not set and forgotten.

AI benchmarks

Standardized tests used to compare models on specific skills such as reasoning, coding or reading comprehension. Benchmarks are useful signals, but they rarely reflect your exact business task.

Evals (AI evaluations)

Custom tests an organization builds to check a model or AI feature against its own real tasks and quality standards before and after launch.

Scaling laws

Observed, fairly predictable relationships showing that model performance improves as model size, training data and computing power increase.

MLOps

Practices and tools for deploying, monitoring, retraining and governing machine learning models reliably in production, similar to DevOps for software.

LLMOps

MLOps adapted for large language model applications, covering prompt management, evaluation, cost tracking, safety filters and monitoring for drift or hallucinations.

↑ Back to the glossary index

AI ethics, safety and governance

Governance terms are no longer just for legal teams. Regulators, customers and search engines increasingly reward brands that use AI transparently and responsibly.

AI ethics

The principles that guide how AI should be built and used, covering fairness, transparency, privacy, accountability and the prevention of harm.

Responsible AI

The practical frameworks, policies and controls an organization puts in place to develop and use AI safely, legally and ethically.

AI governance

The policies, roles and review processes that decide which AI uses are allowed, who approves them and how they are monitored over time.

In marketing: Effective governance is attached to the decision being automated, such as who gets contacted or what a customer sees, not to the label on the tool.

Bias in AI

Systematic errors that cause a model to favor or disadvantage certain groups, often because the training data under- or over-represents them.

Explainable AI (XAI)

Methods that make a model’s reasoning understandable to people, for example showing which factors most influenced a score.

Interpretability

How easily a person can understand why a model produced a particular result. Simpler models are usually more interpretable than deep neural networks.

AI alignment

The work of making sure an AI system’s goals and behavior match human intentions and values, including in situations its designers did not anticipate.

AI safety

The field focused on keeping AI systems reliable, secure and free from harmful behavior, from everyday errors to long-term risks.

Prompt injection

An attack that hides malicious instructions in content a model reads, such as a web page or email, to trick it into ignoring its rules or leaking data.

In marketing: Any agent that reads inbound email or browses the web needs prompt-injection safeguards before it gets access to customer data.

Data poisoning

An attack that corrupts a model by inserting, altering or mislabeling examples in its training data.

Adversarial machine learning

The study of inputs deliberately designed to fool models, and of the defenses that make models more robust against them.

Deepfake

Realistic synthetic audio, video or images that depict a person saying or doing something they never did, typically created with generative AI.

Algorithmic monoculture

A situation where many organizations rely on the same or very similar models and data, so their decisions, and their mistakes, become uniform.

In marketing: If every competitor uses the same AI targeting defaults, everyone ends up bidding for the same prospects. Proprietary data is the differentiator.

AI sovereignty

A country’s or company’s ability to control its own AI capabilities, data and infrastructure in line with its strategic interests and laws.

Recursive self-improvement

A theoretical scenario in which an AI system improves its own design or code, and each improvement makes the next one easier.

↑ Back to the glossary index

Most AI glossaries stop at the technology. These are the terms marketers run into every week, and where AI turns into traffic, leads and revenue.

Answer engine optimization (AEO)

Structuring and writing content so AI-powered answer engines, such as Google’s AI features, ChatGPT and Perplexity, can understand, trust and cite it in their responses.

In marketing: AEO builds on classic SEO: clear question-based headings, concise direct answers, structured data and verifiable sources.

Generative engine optimization (GEO)

A closely related term for improving a brand’s visibility and accuracy inside answers produced by generative AI search tools. Many practitioners use AEO and GEO interchangeably.

AI Overviews

Google’s AI-generated summaries that can appear above traditional search results, drawing on and linking to web sources.

In marketing: Being cited in an AI Overview can put your brand above the first organic result. See how Google, ChatGPT and DeepSeek treat sensitive topics in our YMYL analysis.

AI citation (LLM visibility)

How often, and how accurately, AI assistants mention or link to your brand and content when answering questions in your category.

In marketing: AI visibility is now a budget line of its own. Our 2026 SEO budget guide covers where to invest.

Predictive lead scoring

Using machine learning to rank leads by their likelihood to convert, based on firmographics, behavior and historical outcomes, instead of manual point rules.

In marketing: Better scoring helps marketing hand sales the leads most likely to close, rather than raw volume.

Propensity modeling

Statistical or machine learning models that estimate how likely a person or account is to take a specific action, such as buying, upgrading, responding or churning.

Lookalike modeling

Finding new prospects who share the characteristics of your best existing customers, using AI to compare attributes at scale.

In marketing: Lookalike audiences are a core targeting option in social media advertising and PPC campaigns.

Intent data

Signals that suggest a person or company is researching a topic or solution, such as content consumption, search behavior or engagement with gated assets.

In marketing: Search queries are among the strongest intent signals there are, which is why SEO and PPC capture buyers who are already looking.

Next-best-action

A decisioning approach that uses data and models to recommend the most valuable next step for each customer, such as an offer, message or channel.

In marketing: Next-best-action works best when every recommendation is tested against a control group, so you can prove it lifts results.

Identity resolution

Matching records from different sources, such as email, CRM, web forms and events, to the same real person or company so AI works from one accurate profile.

In marketing: Without identity resolution, AI personalization targets duplicates and outdated records.

Data enrichment (data append)

Adding missing or updated attributes to existing records, such as verified email addresses, phone numbers, job titles or firmographics.

In marketing: Common examples include adding verified email addresses, phone numbers or company details to a contact list.

Data hygiene

The ongoing work of cleaning, validating, de-duplicating and updating records so models and campaigns run on accurate information.

In marketing: Clean data is the cheapest way to improve the output of any marketing AI tool.

AI personalization

Using AI to tailor content, offers, timing and channel to an individual or account based on their data and behavior.

In marketing: On a website, AI personalization can change headlines, offers and calls to action by visitor, a growing part of modern web design.

Send-time optimization

Using machine learning to predict when each recipient is most likely to open or engage, and scheduling delivery accordingly.

Programmatic advertising

Buying and placing digital ads automatically through software and real-time auctions, with AI increasingly used for targeting, bidding and creative selection.

In marketing: AI now drives most bidding and targeting decisions in paid media. Learn how we manage it in our PPC management services.

Click-bot detection (invalid traffic filtering)

Identifying and excluding non-human activity, such as security scanners and bots that click links, so reports and lead counts reflect real people.

In marketing: Email security tools often “click” every link automatically. Without bot filtering, engagement and lead data can be badly inflated.

AI content syndication

Distributing gated content, such as white papers, reports and webinars, to targeted audiences, using data and AI to match assets with the right decision-makers and to qualify the resulting leads.

In marketing: Syndicated content works best when the asset answers a specific buyer question and leads are screened for bots before they reach sales.

↑ Back to the glossary index

AI infrastructure and roles

The hardware, platforms and job titles you will see in AI strategy conversations and vendor org charts.

GPU (graphics processing unit)

A chip built to perform many calculations in parallel. Originally designed for graphics, GPUs are now the workhorse for training and running AI models.

TPU (tensor processing unit)

A chip designed by Google specifically to accelerate machine learning workloads.

Hyperscaler

A very large cloud provider that operates massive networks of data centers and rents out computing, storage and AI services at global scale.

Edge AI

Running AI directly on local devices, such as phones, sensors or laptops, instead of in a remote cloud, for faster responses and better privacy.

AI tech stack

The combined layers of hardware, cloud infrastructure, data platforms, models and applications that support an organization’s AI work.

AI framework

A software library that gives developers prebuilt building blocks for creating, training and deploying models. PyTorch and TensorFlow are common examples.

AI fluency

The ability to use AI tools productively, critically and responsibly: knowing what they do well, where they fail and how to check their work.

Chief AI officer (CAIO)

An executive responsible for aligning AI investment with business strategy, prioritizing use cases and managing AI risk across the organization.

AI architect

The person who designs how AI systems connect with existing data, applications and business processes.

AI engineer

A builder who turns models into working products, typically by integrating pre-trained models with company data, tools and user interfaces.

Context engineer

A newer role focused on designing the information, memory, retrieval and rules that AI agents need to complete multi-step work reliably.

Data scientist

A specialist who uses statistics, programming and machine learning to extract insights and build predictive models from data.

Data architect

The person who designs an organization’s overall data structures, storage and flows so information is consistent and usable across systems.

AI governance lead (AI ethicist)

The person who sets standards for fair, transparent and compliant AI use, reviews higher-risk applications and monitors for bias.

↑ Back to the glossary index

How UnReal Web Marketing puts these AI terms to work

AI terminology only matters if it turns into more visibility, more traffic and more customers. Here is how the concepts in this glossary map to the services we provide.

GoalAI concepts involvedUnReal Web Marketing service
Rank in Google and get cited in AI answersAEO, GEO, AI Overviews, chunking, semantic search, knowledge graphSEO and AEO services
Get more from every ad dollarPredictive analytics, lookalike modeling, intent data, reinforcement learningPPC management
Reach the right audiences on socialLookalike modeling, AI personalization, multimodal AISocial media advertising
Turn visitors into leadsAI personalization, conversational AI, chatbotsWeb design
Know where you stand todayAI citation, AI Overviews, model drift in rankingsFree SEO audit

Not sure which of these matters most for your business? Compare our SEO packages or read a client case study to see the results in practice.

Frequently asked questions

What is an AI glossary?

An AI glossary is a reference list that defines artificial intelligence terminology in plain language. It helps teams use the same vocabulary when they evaluate AI tools, brief agencies and set AI policies. This glossary covers more than 160 terms across ten categories, with notes on how each applies to marketing.

What is the difference between AI, machine learning and deep learning?

Artificial intelligence is the broad field of building systems that perform tasks requiring human-like judgment. Machine learning is a subset of AI in which systems learn from data instead of fixed rules. Deep learning is a subset of machine learning that uses many-layered neural networks to handle complex data like images, speech and natural language.

What is the difference between generative AI and traditional AI?

Traditional AI analyzes or classifies existing information, for example scoring a lead or flagging fraud. Generative AI creates new content, such as text, images or code, based on patterns learned from training data. Many marketing programs use both: predictive models decide who to reach, and generative models help draft what to say.

What is the difference between a chatbot, an AI agent and agentic AI?

A chatbot holds a conversation and responds to questions. An AI agent goes further by planning steps and using tools, such as search or a CRM, to complete a goal. Agentic AI describes systems with a high degree of autonomy that can manage entire multi-step workflows and adapt as they go, ideally with human approval at key points.

What is a large language model (LLM)?

A large language model is an AI model trained on enormous amounts of text so it can understand and generate human language. LLMs power chat assistants, writing tools and AI-generated search answers, and they are built on the transformer architecture.

What does RAG mean in AI?

RAG stands for retrieval-augmented generation. It is a technique in which an AI model first retrieves relevant information from an external source, such as a company knowledge base or the web, and then uses that information to write its answer. RAG makes responses more accurate and current and reduces hallucinations.

What is an AI hallucination, and how do you reduce it?

An AI hallucination is a confident but false or fabricated output, such as an invented statistic or citation. You can reduce hallucinations by grounding the model in trusted sources with RAG, giving clear instructions, asking for citations, and keeping a human reviewer in the loop before anything is published or sent.

What is the difference between AEO and SEO?

SEO (search engine optimization) aims to rank pages in traditional search results. AEO (answer engine optimization) aims to have content understood, trusted and cited inside AI-generated answers from tools like Google’s AI features, ChatGPT and Perplexity. AEO builds on SEO fundamentals and adds concise direct answers, question-based headings, structured data and strong sourcing.

Which AI terms matter most for marketers?

Start with generative AI, large language models, machine learning, predictive lead scoring, intent data, identity resolution, RAG, hallucination, human-in-the-loop and answer engine optimization. Together they cover how AI creates content, decides who to target, relies on clean data and affects how customers find you in AI search.

DM

About the author: David Montalvo

David is an SEO and digital marketing strategist with 28+ years of experience and the founder of UnReal Web Marketing. He writes about SEO, AI search visibility and digital marketing. Connect on LinkedIn.