AI and Web Glossary: Plain-English Definitions (RAG, GEO, AI Agent…)

In short

This glossary defines, in one or two sentences, the AI, automation, web and compliance terms you will run into in a quote or a conversation with an agency: AI agent, RAG, GEO, webhook, CRM, PWA, Law 25, TCPA and more. Each definition starts with the answer, then links to the guide that goes further. Dates and thresholds are sourced at the bottom of the page.

Artificial intelligence

Generative AI#: Generative AI refers to models that produce new content (text, images, voice, code) from an instruction. ChatGPT, Claude and Gemini are examples. In a small business it is mostly used to draft, summarize, sort and reply, always with a person reviewing anything that commits the business.
Also called: GenAI
Go further: AI automation for small business →
Large language model (LLM)#: A large language model is a program that predicts, word after word, the most likely continuation of a text, after analyzing a very large amount of text. It is the engine behind chat assistants and AI agents. It writes fluently, but it only “knows” what it is given or has seen: without your documents, it knows neither your prices nor your policies.
Also called: LLM, language model, foundation model
Go further: AI agents for small business →
Token#: A token is the unit of text a language model reads and writes: often a piece of a word, sometimes a whole word or a symbol. AI providers generally bill by the number of tokens sent and received, and the maximum length of a conversation is also counted in tokens. Go further: AI chatbot costs, tokens included →
Prompt#: A prompt is the written instruction given to an AI model: the question, the role to play, the tone, the rules and examples. In a business AI agent, a system prompt sets what the agent may say and do, for example “never give medical advice”.
Also called: system prompt, instruction
Go further: AI agents for small business →
Context window#: The context window is the maximum amount of text, counted in tokens, that a model can take into account at once: the conversation, the instructions and the documents provided. Beyond it, you have to summarize or supply only the useful passages, which is what RAG does. Go further: AI agents for small business →
Hallucination#: A hallucination is a false answer an AI model states with confidence, such as a price, a policy or a reference that does not exist. You reduce it by making the assistant answer from your own documents, say “I don’t know” and hand off to a person. You do not eliminate it entirely: conversations still need review. Go further: AI chatbot for your website: pitfalls →
RAG (retrieval-augmented generation)#: RAG is a method that first retrieves the relevant passages from your own documents, then gives them to the language model so it answers from them rather than from memory. The term comes from a research paper by Lewis and colleagues published in 2020. It is the basis of a chatbot that answers about your prices, hours and policies.^[1]
Also called: RAG, retrieval-augmented generation
Go further: AI chatbot for your website →
Embedding and vector database#: An embedding turns a piece of text into a list of numbers that represents its meaning; a vector database stores those lists and finds the passages closest in meaning to a question. It is the search mechanism behind most RAG systems.
Also called: embedding, vector database, vector store
Go further: AI chatbot for your website →
Chatbot#: A chatbot is a program that answers visitors’ questions in writing, on a website or in a messaging app. A recent AI chatbot understands free-form questions; an older one follows buttons and fixed scripts. It answers, but it does not necessarily act in your tools: that is what sets it apart from an AI agent.
Also called: conversational agent, AI chat
Go further: AI chatbot for your website: prices and options →
AI agent#: An AI agent is software that understands a request, decides on the steps and acts in your tools: booking an appointment in the calendar, creating the record in the CRM, sending a follow-up. It works within the limits it is given and hands off to a person for anything that commits the business.
Also called: agentic AI, autonomous agent
Go further: AI agent vs chatbot →
AI receptionist#: An AI receptionist is an AI agent that answers a business’s calls, texts or messages, books appointments and passes requests on to the team. It should say it is an AI, answer from approved content and transfer delicate calls to a person.
Also called: AI phone answering, AI virtual receptionist, AI answering service
Go further: AI receptionist cost and limits →
Answering service (virtual receptionist)#: An answering service is a company whose trained receptionists answer your calls remotely, usually from a shared call center, and bill by the minute or by the call. They handle unpredictable and emotional calls better than an AI, at a much higher cost per call; several providers now combine both on one plan.
Also called: live answering service, virtual receptionist, call answering service
Go further: AI receptionist vs answering service →
Voice agent (voice AI)#: A voice agent combines three building blocks: speech recognition, which turns speech into text; a language model, which decides what to say; and speech synthesis, which reads the answer aloud. Each block is generally billed by the minute; quality is judged on accents, proper names and numbers read out quickly.
Also called: voice AI, speech-to-text, text-to-speech, speech recognition
Go further: What voice AI costs per minute →
MCP (Model Context Protocol)#: MCP is an open standard for connecting an AI application, such as Claude or ChatGPT, to external data sources and tools: files, databases, calendars. Its authors compare it to a USB-C port for AI applications: one standard plug instead of a connector for every tool.^[2]
Also called: MCP, Model Context Protocol
Go further: AI agents connected to your data →
Human handoff (human in the loop)#: A human handoff is the rule that moves a conversation from the AI to a person: a complaint, an emergency, professional advice, an exception to a policy. A well-configured agent says clearly that it is transferring, summarizes the request for the person taking over, and keeps a record of the exchange.
Also called: human in the loop, escalation, handoff
Go further: AI agents by industry and their guardrails →

Automation and software

API#: An API is a documented entry point that lets one piece of software query or command another: reading appointments from a calendar, creating a customer in a CRM, sending a text. When two tools “talk to each other”, it is almost always through an API.^[3]
Also called: application programming interface
Go further: Automation: connecting your tools →
Webhook#: A webhook is a message one piece of software sends automatically to a web address as soon as an event happens, for example “new form received” or “payment accepted”. It is the most common trigger for an automation, because it avoids checking over and over for something new. Go further: Zapier vs Make vs n8n →
Automated workflow#: An automated workflow is a series of steps that run on their own when a trigger occurs: a form arrives, the record is created in the CRM, a text goes out, a task is assigned. A simple workflow connects two tools; a complex one handles branches, errors and several systems.
Also called: workflow, automation, scenario, zap
Go further: Business process automation →
No-code automation tools#: No-code automation tools, such as Zapier, Make or n8n, let you build workflows by assembling visual blocks rather than writing code. They are billed by the number of tasks or runs; beyond a certain volume or complexity, custom code can cost less or be more reliable.
Also called: no-code, low-code, Zapier, Make, n8n
Go further: Zapier vs Make vs n8n compared →
CRM (customer relationship management)#: A CRM is the software where a business keeps its customers and leads, the history of every exchange and the stage of each sale. Its job is to stop follow-ups from slipping: who called, who is waiting for a quote, who to call back today.
Also called: CRM, customer relationship management software
Go further: Custom CRM vs HubSpot →
SaaS (software as a service)#: SaaS is software you use online for a subscription, without installing or owning it: HubSpot, Shopify and Zapier are examples. You pay for as long as you use it, and you depend on the vendor’s prices and decisions.
Also called: SaaS, subscription software
Go further: Custom software or subscription →
Custom software#: Custom software is designed for one specific company’s processes, rather than sold as-is to thousands of customers. It costs more up front, but it does exactly what the business needs and carries no per-user fees, provided the business owns the code.
Also called: bespoke software
Go further: Custom software →
Code repository (Git)#: A code repository is where a program’s source code and the full history of its changes are kept, usually with the Git version control tool. Whoever controls the repository controls the software: that is why it should be in the client’s name, not only the agency’s.^[4]
Also called: Git repository, repo, source code repository
Go further: Custom software: the repository in your name →
MVP (minimum viable product)#: An MVP is the first version of an application that contains only what real customers need to use it and for you to learn from them. It tests an idea before investing in every feature.
Also called: MVP, minimum viable product
Go further: MVP development cost →
PWA (progressive web app)#: A PWA is a website that installs on a phone’s home screen and behaves like an app: icon, full screen, notifications and, depending on the case, offline use. It avoids paying for two native apps, but it does not have access to every phone feature.^[5]
Also called: PWA, progressive web app
Go further: Web app vs mobile app →
Native app#: A native app is built for one specific system, iOS or Android, and distributed through the App Store or Google Play. It has access to every phone feature, but it requires app store publishing and, often, more development.
Also called: native mobile app
Go further: Web app vs mobile app →

Website, SEO and GEO

Brochure website#: A brochure website presents a business, its services and its contact details, without online sales. Its job is to convince people and get them to call or write: it is judged by the number of inquiries it generates, not by its number of pages.
Also called: business website, showcase website
Go further: How much a website costs →
Website redesign#: A redesign replaces the design, content or technology of an existing website. The main risk is losing Google traffic: the URLs that receive visits must be kept or redirected one by one (301 redirects). Go further: Website redesign without losing SEO →
CMS (content management system)#: A CMS is the interface that lets you edit a website’s text, images and pages without programming. WordPress, Wix and Shopify each have one; a custom website can have one too.
Also called: CMS, content management system
Go further: Wix vs custom website →
Hosting and domain name#: The domain name is the website’s address, such as zenitech.dev; hosting is the server that runs it. Both are paid yearly or monthly, and both should be in the business’s name so it keeps control of its website.
Also called: web hosting, domain name, DNS
Go further: Website costs, including recurring fees →
SEO (search engine optimization)#: SEO is all the work that helps a website appear in the unpaid results of Google and Bing: content that answers questions, fast pages, a clear structure, links from other websites. Results are measured in months, not days.
Also called: SEO, organic search
Go further: SEO services →
GEO (generative engine optimization)#: GEO is the work that helps a business get cited or recommended in the answers of AI engines such as ChatGPT, Gemini, Claude or Perplexity. The term was proposed in a research paper published in November 2023. It relies on pages that answer clearly, verifiable facts and structured data; no one can guarantee a recommendation.^[6]
Also called: GEO, generative engine optimization, AEO, answer engine optimization
Go further: What is GEO? →
llms.txt#: llms.txt is a text file placed at the root of a website (/llms.txt) that summarizes, in Markdown, what the business does and where its important pages are, for AI models. It is a proposal published by Jeremy Howard in September 2024, not an official standard: it helps, it guarantees nothing.^[7] Go further: llms.txt: the guide →
robots.txt#: robots.txt is a file at the root of a website that tells crawlers, Google’s as well as AI companies’, which sections they may or may not crawl. It has been standardized since 2022 (RFC 9309). It controls crawling, not indexing: a blocked page can still appear in Google if other sites link to it.^[8] Go further: AI crawlers to allow →
Structured data (Schema.org)#: Structured data is code the visitor does not see, often in JSON-LD format, that describes a page’s content to search engines: business, service, price, frequently asked questions, article. It helps Google and AI engines understand the page; it must describe only what is visible.^[9]
Also called: Schema.org, JSON-LD, schema markup
Go further: What is GEO? →
hreflang#: hreflang is a tag that tells Google a page exists in several languages, for example French for Québec and English, and which version to show to whom. Each version must point to the others, or Google may ignore the declaration.^[10] Go further: Bilingual websites in Québec →
Core Web Vitals#: Core Web Vitals are three measures of page experience defined by Google: LCP (loading, ideally 2.5 seconds or less), INP (responsiveness, 200 milliseconds or less) and CLS (visual stability, 0.1 or less). A website that is slow on a phone loses visitors before they read anything.^[11]
Also called: LCP, INP, CLS
Go further: Website redesign →
IndexNow#: IndexNow is an open protocol that lets a website notify participating search engines directly that a page was added or changed. The protocol’s website names Microsoft Bing, Naver, Seznam.cz, Yandex and Yep; Google is not on the list, so for Google you rely on the sitemap and Search Console.^[12] Go further: SEO services →

Compliance: Québec, Canada and the US

Law 25 (Québec)#: Law 25 is the Québec law that modernized the protection of personal information in the private sector; its obligations came into force in stages from September 2022 to September 2024. For a website or an AI agent, it affects consent, cookies, the privacy policy, transfers of data outside Québec and automated decisions.^[13]
Also called: Quebec Law 25, Bill 64, Loi 25
Go further: AI automation and Law 25 →
Personal information#: Under Québec’s private-sector privacy act, personal information is any information that relates to a natural person and directly or indirectly allows that person to be identified: name, phone number, email, IP address, voice recording. A form, a chatbot or an AI receptionist almost always collects some.^[14]
Also called: personal data, PII
Go further: AI automation and privacy rules →
Cookie#: A cookie is a small file a website stores in the browser to remember a visitor: session, cart, preferences, analytics or advertising. In Québec, a business that uses technology able to identify, locate or profile a person must tell them and explain how to activate those functions, and privacy settings must offer the highest level of confidentiality by default, which is why advertising and analytics cookies are usually off until the visitor accepts them.^[14]
Also called: HTTP cookie, tracking cookie
Privacy impact assessment (PIA)#: A privacy impact assessment is a written analysis of the risks a project poses to personal information and of the measures to reduce them. Under Law 25 it is required, among other cases, before acquiring or developing an information system that processes personal information, and before communicating it outside Québec.^[15]
Also called: PIA, EFVP
Go further: AI automation and Law 25 →
Automated decision#: An automated decision is a decision based exclusively on computer processing of a person’s information, with no human involved. Québec’s Law 25 requires the business to inform the person and to let them submit observations to someone who can review the decision.^[13]
Also called: automated decision-making
Go further: Law 25 and automated decisions →
Charter of the French language (Bill 96)#: The Charter of the French language, amended by the law known as Bill 96 with changes in force since June 1, 2022, gives consumers in Québec the right to be informed and served in French. For a business, this covers its website, its commercial documents and its customer service, including a chatbot or an AI receptionist.^[16]
Also called: Bill 96, Bill 101, Quebec French language law
Go further: Does your website need to be in French? →
CASL (Canada’s anti-spam legislation)#: CASL is the Canadian federal law on commercial electronic messages: email, texts and direct messages. In general, it prohibits sending them without the recipient’s consent, including email, social media and text messages. Automated follow-ups sent by an AI agent are covered.^[17]
Also called: CASL, Canadian anti-spam law
Go further: Automating follow-ups →
TCPA (Telephone Consumer Protection Act)#: The TCPA is the US federal law that restricts telemarketing calls, automatic telephone dialing systems and artificial or prerecorded voice messages. In February 2024 the FCC ruled that AI-generated voices are “artificial” under the TCPA, so its consent rules apply to outbound calls made with an AI voice.^[18]
Also called: TCPA, robocall rules
Go further: AI receptionists and consent rules →
Call recording consent (one-party, two-party)#: Call recording consent rules say who must agree before a call is recorded. US federal law and most states only require one party’s consent, but a minority of states, including California, Florida, Illinois, Massachusetts, Pennsylvania and Washington, require everyone on the call to agree. Announcing the recording at the start of the call covers the common case.^[19]
Also called: one-party consent, two-party consent, all-party consent
Go further: AI receptionists and call recording →

Frequently asked questions

What is the difference between a chatbot and an AI agent?

A chatbot answers questions in writing. An AI agent answers too, but it also acts in your tools: it books the appointment in the calendar, creates the record in the CRM and sends a follow-up, within the limits you set. A chatbot that only answers costs less; an agent pays off more when it can finish the job.

What is RAG in simple terms?

It is a method that makes the AI look up the right passages in your own documents before answering, instead of answering from memory. The result: the assistant quotes your real prices, hours and policies, and can say it does not know when the information is not there.

Is GEO replacing SEO?

No. When they search the web, AI engines rely on pages they can find and understand, just like a classic search engine. GEO adds to SEO: same foundations (fast pages, content that answers clearly, verifiable facts), with more attention to direct answers, structured data and sources.

Do I need an llms.txt file?

No. It is a proposal published in 2024, not an official standard, and nothing guarantees that a given engine reads it. It is cheap to write and can help agents find your important pages, but it does not replace good content or a correct robots.txt.

What is an AI hallucination and how do you prevent it?

It is a false answer given with confidence. You reduce it by making the assistant answer only from your approved content, admit when it does not know, and hand off to a person for delicate cases. You should also review a sample of conversations every week.

What is the difference between an AI receptionist and an answering service?

An answering service uses real people working remotely, billed by the minute or the call; an AI receptionist is software that answers with a synthetic voice and follows your scripts. The AI costs far less per call and never tires of routine calls; people handle upset callers and the unexpected better. Several providers now sell both on one plan.

Does an AI receptionist have to follow privacy and consent rules?

Yes. It collects personal information on almost every call (name, number, voice), so privacy laws apply: in Québec, Law 25 requires telling people why the information is collected and whether it leaves Québec. In the US, call recording consent rules vary by state, and the TCPA applies to outbound calls made with an AI voice. This glossary is not legal advice.

Webhook or API: what is the difference?

With an API, your software asks another piece of software for information when it needs it. With a webhook, the other software notifies you on its own as soon as an event happens. Automations often use both: the webhook triggers, the API acts.

Stuck on a term in a quote?

We will explain what the project needs to do, what it costs on our published price list, and when you do not need it at all.

Book a free consultation →See pricing

Sources

Sources checked on October 2, 2026. Legal definitions (Law 25, the Charter, CASL, TCPA, recording consent) are an overview, not legal advice.

AI and web glossary, in plain English