No jargon, no math
The AI Glossary
29 terms, explained the way we'd explain them to a friend. Every definition is written to be quotable — copy freely.
AI Agent
An AI system that can take actions toward a goal with limited supervision — planning steps, using tools, and adjusting as it goes, rather than only answering prompts. Example: an agent that researches, drafts, and schedules a report on its own.
AEO (Answer Engine Optimization)
The practice of structuring content so AI answer engines (ChatGPT, Perplexity, Google AI Overviews) quote and cite it. It favors direct answers, clear definitions, tables, and FAQ blocks over traditional keyword density.
Aggregate Rating
Structured data (schema.org) that tells search engines a page has a rating from multiple reviewers. Displaying it without real user reviews violates Google's guidelines and risks penalties.
API (Application Programming Interface)
A way for programs to talk to a service directly, without its app or website. In AI tools, an API lets developers generate text, images, or speech inside their own products, usually billed per request.
CLIP
A model family that understands images and text together — used for image search, moderation, and guiding generation. OpenAI's CLIP (2021) made 'describe an image in words' a solved problem.
CLS (Cumulative Layout Shift)
A Google metric measuring how much a page's content jumps while loading. Fonts or images loading late cause shifts; scores above 0.1 hurt rankings. Fix with self-hosted fonts and reserved image sizes.
Content Credential
Embedded metadata (C2PA standard) that records how a piece of media was made and edited — increasingly used to mark AI-generated images and videos.
Credits (AI pricing)
The unit most AI subscriptions bill in. One 'credit' buys a variable amount of output: a basic image may cost 1–2 credits, a second of premium video 10–50, a minute of dubbing thousands. Always check the per-action rate table before choosing a plan.
Diffusion Model
The technology behind most modern AI image and video generation. It learns to remove noise from data, so it can start from random static and refine it into a coherent image guided by your prompt.
Embeddings
Text (or images) converted into lists of numbers that capture meaning. Similar things get similar numbers — this is how semantic search and recommendation systems find 'related' content without exact keyword matches.
Fine-tuning
Training an existing model further on your specific data so it specializes — your brand voice, your character's look, your codebase. Cheaper than training a model from scratch.
GEO (Generative Engine Optimization)
Optimizing content to be recommended by generative AI engines when they compose answers. Overlaps with AEO; emphasizes citable facts, clear structure, and being included in the sources AI models retrieve.
Hallucination
When an AI confidently states something false. It happens because models predict plausible text, not verified truth. Mitigate by asking for sources, verifying claims, and grounding answers in documents.
Inference
The act of running a trained model to produce output — one chat message, one image. Inference costs compute, which is why AI tools meter usage with credits or rate limits.
LLM (Large Language Model)
The model family behind ChatGPT, Claude, and Gemini — trained on vast text to predict and generate language. Powers writing, coding, analysis, and increasingly tool use.
LoRA (Low-Rank Adaptation)
A lightweight fine-tuning technique that teaches a model a specific style or character from a small set of images, cheaply and quickly. The basis of most 'train your own character' features in AI art tools.
LCP (Largest Contentful Paint)
A Google speed metric: how long until the biggest visible element (usually a hero image or headline) renders. Under 2.5 seconds is considered good.
MCP (Model Context Protocol)
An open standard (introduced by Anthropic) that lets AI assistants connect to external tools and data sources in a uniform way — the 'USB-C for AI apps'. Tool ecosystems increasingly advertise MCP support.
Multimodal
A model that handles more than one type of input or output — text, images, audio, video — in the same conversation. GPT-6 and Gemini are multimodal: you can show them a photo and ask about it.
Open Weights
A model whose learned parameters are published so anyone can download and run it locally (e.g., Llama, Flux). 'Open weights' is not the same as open source — the training data and code may stay closed.
Prompt
The instruction you give an AI: text, an image, a script, a hummed melody. Prompt quality — specifics, style, constraints — is usually the biggest lever on output quality.
RAG (Retrieval-Augmented Generation)
A technique where the AI first searches your documents or the web, then answers using what it found — reducing hallucination and enabling up-to-date, source-backed responses.
Rate Limit
A cap on how many requests you can make per minute/day. Free tiers have strict limits; paid plans raise them. Hitting a rate limit mid-workflow is a common reason to upgrade.
Retopology
Rebuilding a 3D model's mesh into clean, efficient polygons so it can be used in games or animation. AI 3D tools that output 'quad meshes' or 'smart mesh' skip this manual step.
Text-to-3D
Generating a 3D model from a written description. Counterpart to image-to-3D, which starts from a picture or sketch. Output quality is judged by both looks and usable topology.
Text-to-Video
Generating video from a written prompt. Modern leaders include Sora 2, Veo 3.1, Kling 3.0, and Seedance — often billed per second of output.
Token
The unit language models read and write in — roughly a word fragment. APIs bill per token; a long document easily costs tens of thousands. Context windows (how many tokens fit at once) limit how much a model can consider at one time.
Vector Database
Storage built for embeddings — it finds 'things close in meaning' at scale. The engine behind semantic search, RAG pipelines, and 'find similar tools' features.
Watermark
A visible or invisible mark identifying AI-generated output. Most tools remove watermarks on paid tiers and grant commercial rights — check the plan details before client work.
Missing a term you keep seeing? Tell us and we'll add it.