AI Glossary

AI set up to act with some independence, taking multiple steps and using tools to finish a job instead of replying once. The more freedom an agent has, the more important it is to check its work and limit what it can access.

A hypothetical AI that could do any intellectual task a person can. It does not exist today, and experts disagree on whether or when it will. You may see it in headlines, but it has no bearing on the tools you use now.

An AI that does more than answer a question: it takes a goal, plans steps, uses tools like search, email, or a calendar, and works through the task. Not to be confused with a real estate agent, which is why this site says 'AI agent' when it means the software.

Telling clients when AI is used in ways that matter, such as altered listing photos or a chatbot answering their questions. Some disclosures are required by law or MLS rules and others are good practice.

The questions about fairness, transparency, accountability, and harm that come with using AI. For agents, it mostly comes down to honesty with clients, protecting their data, and staying responsible for what goes out under your name.

The rules a company sets for how AI may be used: approved tools, what data can go in, who reviews the output, and how problems are reported. NAR has published an optional policy template for brokerages.

A search tool that gives you a written answer, often with source links, instead of a list of web pages. ChatGPT search, Perplexity, and Google's AI answers work this way, and clients are starting to use them to find agents.

A set of step-by-step instructions a computer follows to do a task. In everyday conversation about AI, 'the algorithm' usually means the logic a platform uses to rank, recommend, or score something.

Making sure an AI's behavior matches what people intend and value. It is also how developers describe their work on keeping AI systems safe and avoiding harmful output.

A way for one piece of software to talk to another. When an AI tool connects to your CRM or email, it is usually using an API.

Computer software that does tasks which normally take human judgment, like writing, summarizing, recognizing images, or answering questions. Most of what agents call AI today is generative AI chatbots and the tools built on them.

Software that estimates a property's value from data like sales, size, and location. Online home value estimates are AVMs. They can be a useful starting point and they can be wrong, especially for unusual homes.

Using software to do repetitive tasks with little human effort, like sending a form to every new lead. AI adds the ability to handle tasks that need a little judgment, like summarizing a reply.

Unfair or skewed results from an AI system, often because of the data it learned from or how it was designed. In housing it can show up as steering language in listings or ads, or in tools that score people or properties.

An AI that operates a web browser or computer screen like a person would, clicking, typing, and filling in forms. Useful for repetitive tasks, but it can make mistakes and should not be given access to bank, escrow, or wire accounts.

A scam where a criminal takes over or imitates a real email account, often to send fake wire instructions during a transaction. AI makes the messages more polished and the impersonation more believable.

Software you talk to in a chat window. Some are general assistants like ChatGPT, and others are built into a website or texting tool to answer questions from leads. A growing number of states require chatbots to say they are not a person.

AI that can interpret photos and video, such as identifying objects, reading text, or detecting what is in a room. It is behind photo tagging, virtual staging, and tools that review listing photos.

A standard for attaching tamper-evident information to an image or video about where it came from and how it was edited. Support is growing, but not every tool adds it and it can be stripped, so it is not proof on its own.

Deciding what information to put in front of the model for a task: the instructions, the right documents, and examples. As tools get better, it often matters more than clever wording.

How much text a model can consider at one time, including your instructions, any documents you upload, the conversation so far, and its reply. If a long contract does not fit, the model may miss parts of it. You will also see this called the token limit.

An AI assistant built into an app you already use, like Microsoft 365 Copilot in Word and Outlook, or the AI features inside your CRM. It can also describe any AI that works alongside you instead of replacing you.

A pre-configured assistant with its own instructions and reference files, such as a custom GPT in ChatGPT, a Project in Claude, or a Gem in Gemini. Useful for tasks you repeat, like writing in your voice.

A form of machine learning that uses neural networks with many layers. It is the technology behind today's chatbots, image generators, and voice tools.

Realistic fake images, video, or audio made with AI. A cloned voice can sound like a client, an agent, or a title officer, which makes wire fraud more convincing. Verify any change to payment instructions by calling a number you already have.

A way of turning words, documents, or images into numbers that capture their meaning, so software can find things that are similar in meaning and not just matching keywords. They power document search tools and many 'chat with your files' features.

How well you can understand why an AI produced a given result. It matters when a tool influences decisions about people or property, because you may need to show how a conclusion was reached.

Including a few examples in your prompt so the AI can copy the pattern. For instance, pasting two listing descriptions you like and asking for a third in the same style.

Further training a model on a smaller set of examples so it performs better at a specific job. Most agents will not need it. Custom instructions, Projects, and a good knowledge base usually get you what you want.

A large model trained on a broad range of data that can be adapted to many different jobs. The models behind ChatGPT, Claude, and Gemini are foundation models.

One of the most capable models from the leading AI companies at a given time. The label changes as new releases come out.

AI that creates new content, such as text, images, audio, or video, instead of only analyzing existing content. ChatGPT, Claude, Gemini, and virtual staging tools are all generative AI.

Making your website and profiles easier for AI search tools to understand and cite. It overlaps with traditional SEO: clear, specific, consistent information about who you are and where you work.

Tying an AI's answer to specific source material, ideally with citations you can click. Grounded answers are easier to verify than answers from the model's memory.

When an AI states something false with total confidence, like a disclosure item that is not in the document, a made-up statute, or a wrong number. It happens because models predict plausible text and do not look up facts. Always check important details against the source.

A process where a person reviews or approves the AI's work before it goes out or takes effect. It is the simplest safeguard for client communications, marketing, and anything involving money or legal terms.

What happens when you use a trained model: it takes your input and produces an output. Training is how the model learns; inference is every time you ask it something.

Using AI to remove or add something in part of a photo, like taking out a trash can or adding furniture. It is fine for temporary clutter, but not for hiding defects or permanent features.

A collection of your own documents, notes, and examples that an AI can draw on, like your buyer process, FAQs, or market notes. A good one is what makes an assistant answer like you and not like everyone else.

A model trained on huge amounts of text that can read, write, summarize, and answer questions in plain language. ChatGPT, Claude, and Gemini are built on LLMs.

A plain text file at the root of a website that gives AI tools a short, organized map of the site's content. It is a proposed convention, and not every AI tool uses it.

A kind of AI where software learns patterns from large amounts of data instead of following rules a programmer wrote by hand. Online home value estimates and lead-scoring tools are common examples.

An open standard for connecting AI assistants to outside tools and data, such as a calendar, a document folder, or a CRM. It is a common way AI tools get access to your information, so check what you are connecting.

A feature that lets an assistant remember details between conversations, like your preferred tone or your market. It is convenient, but check what is being saved and avoid letting it store client information.

The trained system behind an AI tool. When people say 'the model,' they mean the engine (like GPT, Claude, or Gemini) that produces the answers, as opposed to the app or website you type into.

When an AI company retires an older model. A prompt or workflow that works well today may behave differently after an update, so recheck anything important when a tool changes.

A model that can work with more than one kind of input, such as text, photos, audio, and documents. It lets you upload a photo of a room or a PDF and ask questions about it.

The field of getting computers to work with human language, in text or speech. Transcription, translation, and chatbots all rely on it.

A system loosely inspired by the brain, made of layers of connected nodes that pass signals to each other. Neural networks learn by adjusting those connections as they see more examples.

Technology that reads text out of scanned pages and images. AI tools often need it to work with scanned contracts or disclosures, because an image of a page is not the same as searchable text.

An open-weight model's files are published so anyone can download and run them on their own hardware. A closed model can only be used through the company's own app or service. Open-weight gives you more control over your data, and closed models are usually easier to use.

Personally identifiable information (PII) is anything that can identify a person. Nonpublic personal information (NPI) is financial information about a consumer, like account numbers or income. Federal privacy rules for financial institutions often cover NPI, and that can include title and settlement companies. Keep both out of tools your brokerage has not approved.

A prompt is the instruction you give an AI. Prompt engineering is the habit of writing prompts that are specific enough to get a useful answer: who you are, what you want, what to avoid, and how it should look.

Hidden instructions placed in a document, email, or web page that try to trick an AI reading it into doing something else, like revealing information or sending a message. It is a risk whenever an AI agent reads content you did not write.

When a factor that looks neutral, like a ZIP code, ends up standing in for a protected characteristic and produces a discriminatory result. It is one way an automated tool can cause a fair housing problem without using protected information directly.

A setup where the tool first looks up relevant passages from your documents and then gives them to the model along with your question. It is how many 'ask questions about your files' tools work, and it helps keep answers tied to your sources.

A limit on how much you can use an AI tool in a given period, based on your plan. If a tool suddenly stops responding or slows down, you may have hit one.

A model that spends extra time working through a problem step by step before it answers. It tends to do better on complicated questions, like comparing several documents, and it is usually slower.

Fraud where someone poses as the owner of a property, often vacant land or a home with no mortgage, to sell it or collect the proceeds. AI-generated documents and voices make it harder to spot.

When people on a team use AI tools the company has not approved, often with good intentions. It is how client data ends up in places no one meant it to, and a written AI policy is the usual fix.

A smaller, lighter language model that needs less computing power. Some run on a laptop or phone, which can matter if you want to keep sensitive documents off outside servers.

Guiding people toward or away from neighborhoods or homes based on protected characteristics like race, religion, familial status, or disability. It is a fair housing violation, and AI-written copy can include steering language without anyone intending it.

Code on a web page that labels its content for machines, such as the business name, address, reviews, or article author. It helps search engines and AI tools read your pages correctly.

Images, video, or audio created or changed by AI, including AI-generated people, voices, and altered photos. Rules on labeling it are developing state by state.

Standing instructions the AI follows in every conversation, like your market, your tone, and what never to include. Setting them up once saves you from repeating yourself in every prompt.

A model that creates images from a written description. Most work by starting with random noise and gradually refining it into a picture. Useful for graphics, but never present an AI-generated image as a real photo of a property.

The small chunk of text (a word or part of a word) that a language model reads and writes. Tokens matter because they determine how much a model can handle at once and often how usage is priced.

A model's ability to call outside tools, like a search engine, a calculator, a calendar, or your CRM, instead of answering from memory alone. Coordinating several tools in one task is often called orchestration.

The information a model learned from. It also matters for privacy: depending on the tool and your settings, what you type in may or may not be used to train future models, so check before you enter client information.

The design behind most modern language and image models. It lets a model weigh how every part of the input relates to every other part, which is why chatbots can follow long instructions and documents.

Adding digital furniture and decor to photos of an empty room. AI has made it fast and realistic, which is why many MLSs and some states require staged photos to be labeled.

Connecting your apps so one event triggers the next steps, such as a new lead in your CRM that sends an intake form and creates a task. Tools like Zapier and Make do this, and AI can be one of the steps.

Settings or plans where a provider does not keep your conversations or use them to train its models. What is available depends on the vendor and the plan, so read the terms before putting client information in.

Asking an AI to do a task with no examples, just the instruction. It works for simple requests, and adding examples usually improves harder ones.