The words, in plain English

Every AI term you keep hearing, explained the way a friend would — with everyday examples instead of jargon.

AI Agent

An AI agent is a piece of software that can take real actions on your behalf — such as searching the web, filling in forms, sending emails, or making bookings — working through a task step by step until it's done. Unlike a chatbot that simply answers questions, an agent actively does things in the world. That also means mistakes have real consequences: an agent could send the wrong email or submit a form incorrectly, so it's worth reviewing what it did when the task matters.

AI Companion

An AI companion is a chatbot or dedicated app built for ongoing friendly conversation and emotional support — it is always available and never gets impatient. These tools are most often used by older adults living alone and by people who feel isolated or lonely, giving them a low-pressure space for everyday conversation. They can genuinely help at the margins — easing feelings of loneliness or providing a listening ear at odd hours — but they are designed to supplement human relationships, not replace them. Experts caution that leaning on AI companionship too heavily can make it harder to build or maintain real-world connections.

AI Image Generator

An AI image generator is a tool that turns a text description — like 'a golden retriever wearing a party hat in a sunny garden' — into an actual image in seconds, no drawing skills required. Popular free options include Adobe Firefly, Microsoft Designer, and Canva AI. Results can be impressive, but watch for common quirks: hands and fingers often look strange, and any text written inside an image is usually garbled. Before using a generated image commercially, check the platform's terms — copyright ownership of AI-generated images is still legally unsettled in many countries.

AI Mode (Google Search)

AI Mode is Google's newer search experience that generates a conversational, written answer to your question — like asking a knowledgeable assistant — instead of (or alongside) showing a list of links to websites. Unlike AI Overviews, which adds a short summary snippet above the usual search results, AI Mode opens a back-and-forth conversation where you can ask follow-up questions. Like all AI systems, it can still state incorrect information with confidence, especially on recent or niche topics, so it is worth checking important answers against reliable sources.

AI Overviews

AI Overviews is Google's name for the AI-written summaries that sometimes appear at the very top of Google search results, before the regular links. Instead of showing you a list of websites to click, Google's AI reads those websites and writes a short answer for you. It's handy for quick questions, but the summary can sometimes be wrong, so it's worth scrolling down to check the source links.

AI Slop

AI slop is the flood of low-effort, AI-generated articles, videos, and social media posts published at scale to attract clicks rather than to genuinely help readers. Because this content is extremely cheap to produce, it overwhelms search results with vague, generic text that all sounds alike. The quickest signs you are reading it: there is no named author or credible expert source, the writing sounds polished but never gives concrete facts or actionable details, and the whole piece could apply to almost any topic without saying anything specific.

AI Subscription Tier

An AI subscription tier is a paid plan — typically around $20 a month — that gives you access to more powerful AI models, higher daily usage limits, and extra features like image generation or voice mode that are restricted or absent on the free version. The free tier is genuinely capable for light, everyday use: many people draft emails, get quick answers, or brainstorm ideas without ever paying. If you use AI heavily throughout the day, run into usage caps, or want to generate images and use advanced tools, the paid tier is likely worth the cost.

AI Training Data

AI training data is the enormous collection of text, images, audio, and other content that an AI model learned from — including public websites, books, and potentially your conversations with AI chatbots. During training, the AI didn't memorise individual examples; instead it spotted patterns across billions of pieces of content to learn how language and information work. The catch for users is that your private chats may be used by default to help improve future versions of the model. Most major AI services let you opt out of this in their privacy settings.

AI Watermark

An AI watermark is a tag — sometimes visible, often invisible and embedded in the file itself — that marks an image, audio clip, or piece of text as AI-generated, making it harder to spread AI content as if it were real. Invisible watermarks hide data inside the pixels or audio samples rather than stamping anything on top, so the file looks exactly the same to you. EU law now requires platforms and AI tools to label AI-generated content, and watermarking is one of the main ways to meet that requirement. Watermarks can sometimes be removed or stripped out, so they are a useful signal rather than an absolute guarantee.

AI detector

An AI detector is a tool that tries to figure out whether a piece of writing, an image, or a video was made by AI or by a human. These tools look for patterns that AI tends to produce. They're not very reliable — they make mistakes in both directions, flagging real human writing as AI or missing real AI content — so you shouldn't treat their verdict as definitive proof.

ATS (Applicant Tracking System)

An ATS, or Applicant Tracking System, is software that large employers use to automatically receive, sort, and filter job applications before any human recruiter reviews them. When you upload a resume, the ATS scans it for keywords, job titles, and formatting — and applications that don't match the job posting's language may be rejected by the computer before a person ever reads them. To improve your chances, mirror the exact wording from the job description, avoid multi-column layouts and tables, and be aware that many employers now add an AI scoring layer on top of the basic keyword filter.

Artificial intelligence (AI)

Artificial intelligence is when a computer is built to do things that normally require human thinking — like understanding language, recognizing faces, or making decisions. It's not a robot brain; it's software trained on huge amounts of data so it can spot patterns and respond in useful ways. AI is already in things you use every day, from spam filters in your email to the voice assistant on your phone.

ChatGPT

ChatGPT is a popular AI chatbot made by a company called OpenAI. You type a question or request, and it writes back a human-like response — it can answer questions, summarize articles, write emails, or help brainstorm ideas. It's free to try at chat.openai.com, with a paid version that adds extra features.

Chatbot

A chatbot is a computer program you can have a text conversation with, like chatting with a person online. Older chatbots could only answer a set list of questions, like a phone menu. Newer AI chatbots — powered by large language models — can understand almost any question you type and reply in plain, natural sentences.

Chatbot Memory

Chatbot Memory is a feature that lets an AI assistant remember facts about you — such as your job, dietary restrictions, or writing preferences — so you do not have to re-explain yourself at the start of every conversation. Unlike your chat history, which is a full transcript of past messages, a memory is a short saved note about a single fact. You can view what the AI has stored and delete any individual entry at any time. You can also turn the feature off entirely if you would rather start every session from scratch.

Chatfishing

Chatfishing is using an AI chatbot — often paired with stolen or AI-generated photos — to fake a romantic relationship on a dating app or social media, building emotional trust over weeks before asking for money or personal information. Today's bots are convincingly human: they remember details from earlier chats, mirror your communication style, and message at any hour without getting tired. Classic warning signs include always having an excuse to avoid a video call and suddenly facing a financial emergency that only you can solve. Protect yourself by asking for a live video call early, reverse-image-searching profile photos, and never sending money to someone you have only met online.

Context Window

A context window is the maximum amount of text an AI can read and consider at once — everything you have typed plus its own replies within a single conversation. Once a chat grows beyond that limit, the AI effectively 'forgets' the earliest messages, as if they were never there. Uploading a short email or a news article works fine for most AI tools, but processing a full book or a very long document requires a model with a large context window. Context window size is one of the main differences between free and paid AI plans.

Custom GPT / Gem

A Custom GPT (in ChatGPT) or Gem (in Google Gemini) is a personalized AI assistant you set up once — giving it a name, instructions like your dietary rules or preferred writing style, and optionally some files — so every new conversation starts with that context automatically, without you having to re-explain anything. You can create as many as you like for different purposes, such as a cooking helper, a work email writer, or a study partner. They are free on most plans and private to your account by default, so nobody else can see or use them.

Deepfake

A deepfake is a photo, video, or audio clip where AI has replaced or altered someone's face, voice, or both to make it look like they said or did something they never did. The word blends 'deep learning' (the AI technique used) and 'fake.' Deepfakes range from harmless fun — like swapping faces in a movie — to seriously harmful, like fake videos designed to spread lies about real people.

EU AI Act

The EU AI Act is the world's first comprehensive law regulating artificial intelligence, and it applies to any AI service used or offered in Europe. As of August 2026, chatbots must clearly disclose that they are AI, and images, audio, or video generated by AI must carry visible labels. You may already notice these disclosures in chat windows, on company websites, and in social media posts. Other countries are watching closely, and similar laws are now being developed around the world.

Generative AI

Generative AI is a type of AI that creates new content — text, images, audio, video, or code — instead of just answering yes-or-no questions or sorting data. When you ask ChatGPT to write a birthday message, or ask an AI image tool to draw a sunset, that's generative AI at work. It generates something new based on patterns it learned from massive amounts of existing content.

Griefbot (Deadbot)

A griefbot is an AI system trained on a deceased person's text messages, emails, voice recordings, or social media posts to simulate conversation with them after death. Some people find comfort in talking to one, but researchers warn that it can interfere with healthy grieving by making it harder to accept the loss. It is a pattern-matching system, not the person — it can produce responses the deceased would never have said. Laws around creating a griefbot without the person's prior consent are still catching up in most countries.

Hallucination

In AI, a hallucination is when an AI chatbot confidently states something that is completely made up. It might invent a fake news article, cite a book that doesn't exist, or give you a wrong phone number — all while sounding totally certain. It happens because the AI is very good at generating plausible-sounding text, but it doesn't actually 'know' facts the way a person does. Always double-check important information from an AI.

Large language model (LLM)

A large language model is the AI engine behind most modern chatbots, including ChatGPT. It learned from an enormous amount of text — books, websites, articles — and got very good at predicting what word should come next, which lets it write and talk in a natural, human-like way. 'Large' refers to the sheer amount of data it was trained on and the computing power it took to build.

Machine learning

Machine learning is the way AI gets smart — by practicing. Instead of a programmer writing out every rule, the computer looks at thousands or millions of examples and figures out the patterns on its own. For example, a machine-learning system learns to spot spam email by reading millions of spam and non-spam messages until it can tell the difference. Most AI you use today is powered by machine learning.

On-Device AI

On-Device AI means the artificial intelligence processing happens directly on your phone or computer rather than being sent to a company's remote servers. This makes it faster for everyday tasks and more private, because your requests, photos, and messages never leave your device. Products like Apple Intelligence, Google Pixel's AI features, and Samsung Galaxy AI use on-device processing for many basic tasks. The trade-off is that on-device AI is less powerful than cloud-based AI, so it may struggle with more complex requests.

Opt-Out (AI Training)

Opting out of AI training is a privacy setting, available in most major AI apps, that tells the company not to use your future conversations to improve its models. You will usually find the toggle in Settings under Privacy or Data Controls. It is separate from deleting your chat history — those are two different steps. Opting out also does not undo any training that already happened using your past conversations, and it does not make the AI less helpful for you personally.

Prompt

A prompt is the message or question you type into an AI tool to tell it what you want. Think of it like giving instructions to a very capable assistant — the clearer and more specific your instructions, the better the result you get back. For example, instead of typing just 'recipe,' you might type 'Give me a simple pasta recipe for two people with no dairy.'

Quishing (QR Code Phishing)

Quishing blends the words QR code and phishing: scammers place fake QR codes on parking meters, restaurant tables, email invoices, or posters that redirect you to convincing fake websites designed to steal your login credentials or payment details. QR codes are especially dangerous because you cannot see the destination URL just by looking at the code, and most email security filters do not scan inside images. Before tapping any QR code, wait for the URL preview to appear on your screen and check that it looks legitimate — this one habit stops most attacks.

See also: Chatfishing, Deepfake, Voice cloning

Smart Speaker

A smart speaker is a small home device — like the Amazon Echo, Google Nest Mini, or Apple HomePod — that listens for your voice and uses a built-in AI assistant to answer questions, play music, or control smart lights, all without picking up your phone. It is always listening for its wake word, which makes it convenient but also means there is a live microphone in your room around the clock. Smart speakers are best for quick, hands-free tasks; they cannot browse the web freely or handle the long, back-and-forth conversations that AI assistants on your phone or laptop can.

Voice Assistant

A voice assistant is an AI you control by speaking out loud — Siri, Google Assistant, or Amazon Alexa — built into your phone, tablet, or smart speaker to answer questions, set reminders, play music, and control smart home devices. Unlike a text chatbot, it works hands-free and is faster for simple, quick tasks, but less suited for long or complex conversations. One privacy consideration: the device listens passively for a wake word such as 'Hey Siri' or 'Alexa', so the microphone is always on and may occasionally pick up conversations that were not intended as commands.

Voice cloning

Voice cloning is when AI listens to a recording of someone's voice — even a short clip — and learns to copy it, so it can say anything in that same voice. It sounds real enough to fool people on the phone. Scammers use it to impersonate family members in distress. Legitimate uses include helping people who have lost their voice due to illness.