AI UX Glossary
Plain definitions of the patterns, risks and principles behind products with AI features, alongside the classic UX laws designers still search for. Each entry says what the term means, why it matters and how to apply it, with links to the primary sources: Google's People + AI Guidebook, Microsoft's Guidelines for Human-AI Interaction, Apple's Human Interface Guidelines, Nielsen Norman Group and the original research papers.
AI UX patterns
Interface patterns for products with AI and LLM features: prompting, streaming, review steps, feedback and recovery.
- Agentic UX
- Design for AI agents that take multi-step actions for a user, focused on showing the plan, asking approval before consequential steps, and keeping control.
- AI empty state
- What an AI input shows before the first prompt: what the tool can do, its limits, and an easy way to start, instead of a blank text box.
- AI error recovery
- Designing for the moments an AI feature fails, so people notice the problem, understand it and can correct, retry or take over without losing work.
- AI feedback loops
- Ways users tell an AI product what worked or failed, such as thumbs up/down, edits and corrections, and how the product shows that input had an effect.
- AI loading states
- What an AI product shows while a model is working, from spinners to step-by-step status, so waits of seconds or minutes feel clear rather than broken.
- AI onboarding
- How an AI product introduces itself: what it can and can't do, how to get good results, and how it will change, taught in short steps when needed.
- AI reasoning disclosure
- Showing how an AI reached its answer (sources, steps, inputs used) in layers: a short summary first, with more detail available on request.
- AI-to-human handoff
- Passing a conversation or task from an AI system to a person, or back to the user, with enough context that nobody has to start over.
- Confidence indicators
- UI cues that show how certain an AI system is about an output, such as labels, ranked alternatives or ranges, so people know how much to rely on it.
- Conversational UI
- An interface where people use natural language, typed or spoken, and the system replies in kind, as in chatbots, voice assistants and AI chat.
- Explainability
- How well people can understand why an AI system produced an output. Often paired with interpretability: what that output means for their task.
- HAX guidelines
- Microsoft's 18 Guidelines for Human-AI Interaction: design rules for AI features, grouped by when they apply, from first use to long-term use.
- Human-in-the-loop
- A design where a person reviews, approves, corrects or can stop an AI system's work at key points, instead of the system acting fully on its own.
- Personalization controls
- Settings that let people see, adjust, pause or reset what an AI has learned about them and how it uses that to tailor results.
- Prompt suggestions
- Ready-made example prompts or follow-up questions an AI product offers so people can start, or continue, without writing a prompt from scratch.
- Streaming responses
- Showing an AI model's output as it is generated, token by token, instead of waiting for the full answer before displaying anything.
- Undo for AI actions
- Letting people reverse what an AI changed or did, quickly and in one step, so automation mistakes are cheap to fix instead of costly to clean up.
AI trust & safety
How people come to trust (or over-trust) AI output, and the design choices that keep that trust earned.
- AI data privacy and consent
- Asking clear permission before an AI feature uses personal data, explaining how it is used and stored, and letting people opt out later.
- AI disclosure
- Telling people clearly when they are interacting with an AI system or seeing AI-generated content, at the point where it matters.
- AI guardrails
- Rules, filters and permission limits around an AI system that keep its inputs, outputs and actions within safe, intended bounds.
- AI mental models
- People's beliefs about how an AI system works, what it can do and how their actions affect it. Wrong beliefs lead to misplaced trust and misuse.
- Automation bias
- The tendency to over-rely on automated advice, accepting a system's suggestions or missing problems it fails to flag, even when other evidence disagrees.
- Calibrated trust
- When how much people trust an AI system matches how reliable it actually is, avoiding both over-reliance and under-reliance.
- Citations and grounding
- Tying AI answers to specific, checkable sources and showing those sources so people can verify claims and see where information came from.
- Hallucination
- When a generative AI system produces output that sounds plausible and confident but is false, made up or unsupported by its sources.
Classic UX laws
The named laws and effects from psychology and HCI research, with what they do and do not say.
- Aesthetic-usability effect
- People tend to perceive attractive designs as easier to use, and are more tolerant of minor usability problems in products that look good.
- Doherty threshold
- The idea that productivity rises sharply when a system responds fast enough (usually cited as under 400 ms) that neither user nor computer waits.
- Fitts's law
- The time to hit a target depends on how far away it is and how big it is: distant, small targets take longer and cause more misses.
- Goal-gradient effect
- People put in more effort the closer they get to a goal. Visible progress, even progress given for free, makes them speed up and finish.
- Hick's law
- Choice reaction time grows with the number of equally likely options, but logarithmically: each doubling of options adds a roughly fixed amount of time.
- Jakob's law
- Users spend most of their time on other sites and apps, so they expect yours to work the same way as the ones they already know.
- Miller's law
- People can hold only a small number of items in short-term memory at once, so chunk information. It is often misused to cap menus at seven items.
- Peak-end rule
- People judge a past experience mostly by its most intense moment and its ending, not by the average of every moment or by how long it lasted.
- Postel's law
- Be liberal in what you accept and conservative in what you send: tolerate varied user input, but produce output that is consistent and precise.
- Serial position effect
- People best remember the first and last items in a list (primacy and recency) and tend to forget the ones in the middle.
- Tesler's law
- Also called the law of conservation of complexity: every application has some complexity that can't be removed, only shifted between users and builders.
- Von Restorff effect
- When several similar items appear together, the one that differs from the rest is the one people are most likely to notice and remember.
Interaction & usability
Core usability principles that apply to every interface, AI or not.
- Affordance
- An affordance is an action an object or interface makes possible for a person, such as a button that can be pressed or a field you can type in.
- Cognitive load
- Cognitive load is the mental effort needed to use an interface. Good design cuts the effort that does not help people reach their goal.
- Deceptive patterns
- Deceptive patterns, also called dark patterns, are design tricks that push people into choices they would not otherwise make, often to their harm.
- Mental model
- A mental model is what a person believes about how a system works. People use it to predict what will happen when they act.
- Progressive disclosure
- Progressive disclosure shows only the most important options first and reveals advanced or rarely used ones when people ask for them.
- Recognition over recall
- Recognition over recall means keeping options, actions and information visible so people can spot what they need instead of remembering it.
- Signifier
- A signifier is any perceivable cue that tells people what action is possible and where to do it, like a push plate on a door or an underlined link.
- Usability heuristics
- Usability heuristics are broad rules of thumb for interface design. The best known are Jakob Nielsen's 10, used to review designs for problems.