A
Agent (AI agent)
An AI tool that can take steps on its own to finish a task, such as searching, filling in forms, sending messages, or changing files, instead of only replying with text. Because it can act for you, give it only the access it needs and review what it plans to do before it sends, buys, or deletes anything.
AI (artificial intelligence)
A broad label for computer systems that do tasks we usually associate with human thinking, such as recognizing speech, sorting photos, or writing text. It covers many kinds of tools, from spam filters to chat assistants.
AI use policy
A short written agreement about how your church or business will and will not use AI tools: what information stays out, who reviews AI-assisted work, and when you tell people AI was used. It works best when leadership discusses it together and revisits it from time to time.
Alt text
A short written description of an image that screen readers read aloud for people who cannot see it. AI tools can draft alt text, but a person who knows what the picture shows should check it before it is published.
API (application programming interface)
A way for one piece of software to talk to another directly, without a person clicking through screens. Many AI services offer one so other apps and websites can send them requests. The settings and terms for this kind of use can differ from the regular app, so check them separately.
B
Bias
When an AI tool's output leans unfairly toward or against certain people, groups, or viewpoints, often because of patterns in the material it learned from. Read AI drafts that describe or affect people, such as job posts or summaries, with fairness in mind.
C
Chatbot (chat assistant)
A program you talk with by typing or speaking, which replies in conversation. Most of today's chatbots are built on large language models; some are general-purpose and some are set up for one website or task.
Context window
The amount of text an AI model can take into account at one time, including your instructions, anything you paste or upload, and the conversation so far. In a very long chat or document, earlier details may be dropped or given less weight, so restate what matters or start a fresh conversation.
D
Data retention
How long a service keeps what you type, upload, or record, and which parts it keeps, such as conversations, files, or audio. Before you share anything sensitive, check your tool's settings and terms for chat history, auto-delete, and how to delete your data.
Deepfake
Audio, video, or images made or altered with AI so that a real person appears to say or do something they never did. Be cautious with surprising clips or urgent messages, and confirm through a contact method you already trust before acting on them.
Disclosure (AI disclosure)
Telling people when and how AI was used to help create something they read, see, or hear, or to handle their information. Being open about it, and asking permission where personal information is involved, helps keep people's trust.
E
Embedding
A list of numbers that represents the meaning of a piece of text or an image, so a computer can find items with similar meaning even when the exact words differ. Embeddings power many "smart search" features and are a common building block of retrieval (RAG).
F
Fine-tuning
Training an existing model further on a smaller, focused set of examples so it handles a particular style or task better. Most church offices and small businesses do not need it: clear prompts, good examples, and reusable templates are the place to start.
G
Generative AI
AI that produces new content, such as text, images, audio, or video, rather than only sorting or labeling existing content. Treat what it makes as a draft for a person to check.
H
Hallucination
When an AI tool states something false or made up as if it were true, such as an invented fact, quote, source, or date. Check names, numbers, quotes, and references against a trusted source before you use them.
Human review (human in the loop)
Having a person read, check, and approve AI output before it is used or sent. It is a basic safeguard for anything that goes to members, customers, or the public.
I
Image generation
Creating new pictures from a written description. Generated images can contain odd details or garbled lettering and may look like real people or existing artwork, so review them closely and say they were made with AI where someone could mistake them for real photos.
K
Knowledge cutoff
The point in time after which a model has no training information. Unless the tool can search the web, it may not know about recent events or changes, so check anything time-sensitive yourself.
L
Large language model (LLM)
The kind of generative AI behind most chat tools, trained on a very large amount of text. It writes by predicting what text is likely to come next, which makes it fluent with language but not a reliable source of facts on its own.
M
Machine learning
A way of building AI in which the system learns patterns from many examples, instead of following rules a programmer wrote by hand.
Model (AI model)
The trained system at the heart of an AI tool: the part that actually produces the answers. One app may offer several models, and the same model can sit behind several different apps.
Multimodal
Able to work with more than one kind of input or output, such as text, images, audio, and documents in the same conversation. A photo, screenshot, or scan you upload can contain private details just as typed text can.
O
Opt-out of training
A setting or request that asks a service not to use your conversations or files to improve its models. Whether it exists, what it covers, and whether it is on by default varies by tool and account type, so check your tool's settings and terms instead of assuming.
P
Personal information (PII)
Details that identify a person or can be linked to them, such as names with addresses, phone numbers, birth dates, giving records, health or prayer details, or account numbers. PII stands for "personally identifiable information". Leave it out of prompts, or remove or replace it, unless your church or business has approved the tool and its settings for that use.
Plug-in / integration (connector)
An add-on that connects an AI tool to other services, such as email, calendars, file storage, or a website. It can save time, but it also lets the AI tool see or change information in those services, so connect only what you need.
Prompt
What you type or say to an AI tool: your question, your instructions, and any background it needs. A clear prompt says what you want, who it is for, and what form the answer should take.
Prompt injection
Hidden or sneaky instructions placed in a web page, email, or document that try to make an AI tool ignore your instructions and do something else. It matters most for agents and connected tools, so keep their access limited and review what they propose to do.
Prompt template
A reusable prompt with blanks to fill in, such as [event name] or [date], so routine tasks get consistent results. Keeping a small set of tested templates saves time and makes it easier to leave private details out.
R
Retrieval-augmented generation (RAG)
A technique in which a tool first looks up relevant passages in a chosen set of documents, such as your handbook or FAQ, and then has the model answer using them. It can keep answers closer to your own material, but the documents need to be current and the answers still need checking.
S
Summarization
Using AI to shorten a long text, such as meeting notes or a report, into its main points. A summary can leave out or subtly change important details, so compare it with the original before you share it.
System prompt
Background instructions given to a model before your conversation starts, usually by whoever built or set up the tool, that shape its role, tone, and rules. Some tools let you add your own standing instructions; keep private information out of those too.
T
Temperature
A setting, mostly found in developer tools, that controls how predictable or varied a model's wording is. Lower settings give more consistent, focused answers; higher settings give more varied, creative ones.
Token
A small chunk of text, often a whole word or part of a word, that AI models read and write in. Length limits, and on some services usage charges, are measured in tokens.
Training data
The large collection of text, images, or other examples a model learned from. Its gaps, errors, and assumptions can show up in the model's answers.
Transcription (speech-to-text)
Turning recorded or live speech into written text, for example for meeting notes, captions, or a transcript of a recording. Names, accents, and background noise can cause mistakes, so review it before sharing, and let people know when a conversation is being recorded.
Translation (machine translation)
Using AI to turn text from one language into another. It can give a useful first draft, but a fluent speaker should review anything important before it is shared.
V
Voice cloning
Using AI to imitate a specific person's voice from recordings of them. Use someone's voice only with their clear permission, and be wary of unexpected calls or messages asking for money or private information, even when the voice sounds familiar.
W
Watermark (AI watermark)
A visible label or hidden signal added to AI-generated images, audio, or text to show it was made with AI. Watermarks are not always present and can sometimes be removed, so a missing watermark does not prove that something is real.