AI Glossary for Business Owners: What Your Vendor Actually Means

Contents
- One question that sorts out every term
- Bot, chatbot and AI agent: three words, three things
- What is under the hood
- Words about connections
- Words about people and responsibility
- Words about money and measurement
- Five questions that translate any proposal
- Where WhaleBiz fits in this glossary
- Frequently Asked Questions
One vendor offers you a "WhatsApp bot", the second an "AI agent", the third a "chatbot connected to GPT and trained on your data". Sometimes all three sell exactly the same thing, and sometimes three things with almost nothing in common. The difference usually shows up a month later, when a customer writes in the middle of the night "Do you have an appointment tomorrow?" and gets a menu of three buttons. This glossary is for the moment before you sign: an AI glossary for business owners, in plain words, with one question for the vendor next to each term. At WhaleBiz we wrote and run our own system, AI agents and a CRM, so we hear these words every day, including from customers who already have three proposals.
One question that sorts out every term
Almost every technical word in a vendor's proposal answers, in the end, one of two questions: what the system actually does, and who deals with it when it gets something wrong. That is why every term here comes with a question you can ask in the meeting, without feeling awkward. One more thing before we start: when a vendor uses a word you do not know, ask "show me that in a conversation". A five-minute demo on WhatsApp, with a question your customers really ask, will tell you more than any slide deck.
Bot, chatbot and AI agent: three words, three things
Chatbot, or just bot
Software that answers from a fixed script: a menu, buttons, "press 1 for opening hours". It works as long as the customer asks exactly what was planned in advance. The moment they write "Can I come with my kid after kindergarten?", the bot gets stuck or sends them back to the menu. Many business owners tried a bot like this and gave up on it, and rightly so.
AI bot, or "connected to GPT"
Here the software understands free text and answers in its own words. That is already much better, but "understands and answers" still does not mean "does". If its best answer is "you can call the clinic during opening hours", the customer still has to call, and at that hour nobody picks up.
AI agent
An agent understands and also acts: it checks the calendar and books the appointment, sends a payment link or a photo of the product, asks the qualifying questions, records the lead in the CRM and hands to a person whatever needs a person. Someone who searches Google today for "WhatsApp bot" or "chatbot for business" is, in most cases, looking for exactly this, just under the old name. A full explanation with examples is in the article what an AI agent is and what it does in a business.
Question for the vendor: "What does it do by itself, not just answer? And in which systems?"
What is under the hood
Language model
The engine that understands and writes text (vendors sometimes say LLM). OpenAI's GPT, Anthropic's Claude and Google's Gemini are names of models, not business products. When a vendor says "we work with GPT", they told you which engine is inside, not what the conversation with your customer will look like. The models do differ, and we laid it out in our comparison of AI models, but for a business owner what was built on top of the engine matters more.
Question for the vendor: "When a new, better model comes out, who switches, and what do I have to do?"
Instructions, or prompt
The text that tells the agent who it is and how your business talks: what tone, what is allowed and what is not. For example: no medical diagnosis, no promised discounts, every question about medication goes to the doctor. Most of the difference between a good agent and an embarrassing one sits here, not in the model.
Question for the vendor: "Who writes the instructions, and who updates them when something changes on my side?"
"Trained on your data"
The most confusing phrase on the market. In most cases nobody trains any model. The agent gets access to your documents, such as a price list, FAQ, procedures and treatment descriptions, and searches them before it answers. That is good, and usually exactly what you need. What matters is finding out which documents it answers from, and what happens when the price list changes. If you have to send the vendor a new file and wait, better to know that in advance.
Question for the vendor: "If I change a price today, when does the agent know?"
Hallucination
When a model answers something wrong with full confidence: it invents an opening hour, promises parking that does not exist. It happens when it has no answer in the data and the instructions, and fills the gap by itself. Nobody can promise it will never happen, but clear instructions and up-to-date data reduce it a lot.
Question for the vendor: "What does the agent say when it does not know?" The answer you want to hear: it says it will check, and hands over to a person.
Words about connections
API
The official door through which one system talks to another. On WhatsApp this is the most important difference in a proposal: the WhatsApp Business app on your phone is one thing, and the WhatsApp Business Platform, what vendors call "the WhatsApp API", is another, with its own rules and billing. Since 1.10.2026 Meta bills service messages on the platform, including replies written by an AI assistant, and the app on your phone was not affected. All the details are in our article on Meta's new WhatsApp pricing.
Question for the vendor: "Which of the two am I on, and who pays Meta?"
Integration and two-way sync
An integration is a connection to a tool you already have: a calendar, clinic management software, a CRM, payments. Two-way sync means an appointment the agent booked shows up in the secretary's calendar, and an appointment the secretary cancelled frees up for the agent too. Syncing in one direction only is a reliable way to book two customers for the same hour.
Question for the vendor: "Show me an appointment booked in the chat appearing in my calendar."
CRM
The place where every lead is kept: name, phone, what they asked, where they came from and what happened next. Without a CRM, a great WhatsApp conversation stays on the phone of whoever answered, and a week later nobody remembers it.
Question for the vendor: "Where do I see all conversations and leads in one place, and can I export them?"
No-code
Tools where you connect building blocks by dragging, without writing code. Quick to start, and sometimes enough. But when something breaks in a chain of ten connections, you need someone who knows where to look.
Question for the vendor: "Who fixes it when it breaks on a Friday evening?"
Words about people and responsibility
Handoff to a person
The moment the agent passes the conversation to a person: an angry customer, a medical question, a special request, or a customer who simply wrote "I want to talk to someone". A good agent hands over with a short summary, so the person does not ask everything again from the start.
Question for the vendor: "Where does the handoff land, and what happens when nobody is available?" A lead handed to nobody is a lead that is gone.
24/7
The agent answers at any hour, on Shabbat and on holidays too. That is true, and it is the easy part. The real question is what happens to what it passed on in the middle of the night: who sees it in the morning, and in what order.
"Autonomous"
A word you will hear a lot, from us too. For us, the "autonomous sales system" is the direction: one system that brings customers, answers, sells and keeps everything in the CRM. What WhaleBiz does today, honestly: the agent handles most conversations by itself and hands to a person whatever needs a person. If someone promises you a business that runs without a single person, ask to see it working at a real customer.
Want to consult with us?
We can help you choose, build and deploy the perfect AI solution for your business. Leave your details and we'll get back to you.
Words about money and measurement
"Message"
Vendors count messages in different ways. For us, a message is one reply from the agent, and a reply to a voice message or a photo is still one message. With another vendor it can be every message, in both directions. That difference changes the bill quite a bit.
Question for the vendor: "What exactly counts as a message?" Our plans and what each one includes are on the plans and pricing page.
KPI
One number that tells you whether it works: how many enquiries got an answer within a minute, how many appointments were booked in the chat, how many hot leads reached a person. If the vendor cannot say what you count after a month, you will not know either.
Pilot
A trial period on one channel or on part of the enquiries, with a goal set in advance. A pilot without a target number is just another month.
Five questions that translate any proposal
If there is no time to read the whole glossary, these are the five questions worth a meeting:
- What does the agent do by itself, beyond answering?
- What information does it answer from, and how do I update it?
- What does it say when it does not know, and who does it hand over to?
- Which channels does it work in, and where do I see all the conversations?
- What do we count after a month, and what counts as a message?
Where WhaleBiz fits in this glossary
Our term is "AI agent". WhaleBiz agents answer on WhatsApp, the website, Instagram, Telegram and Facebook, in any language. They qualify leads, book appointments in the calendar with two-way sync, send photos, documents and payment links, understand voice messages and answer from the business's own documents. Every conversation is saved in the built-in CRM, and whatever needs a person goes to a person. How this looks in a sales conversation, from the first question to the booked appointment, you can see on the WhaleBiz sales agent page. And if anyone, another vendor or us, uses a word that is not here, ask for a demo. That is the question no term replaces.
Frequently Asked Questions
What is the difference between a chatbot and an AI agent?
A chatbot answers from a fixed script, with a menu and buttons, and gets stuck when the customer asks something nobody planned. An AI agent understands free text and also acts: it books an appointment in the calendar, sends a payment link, records the lead in the CRM and hands to a person whatever needs a person. When people search for a "WhatsApp bot" today, in most cases they are really looking for an agent.
Does "trained on our data" mean my data is used to train a model?
In most cases, no. Usually the agent searches your documents, such as a price list, FAQ and procedures, before it answers, and no model is retrained. It is worth asking the vendor in writing which documents the agent answers from, how they are updated, and what is done with your data.
Can an AI agent run without any staff?
A good agent handles most conversations by itself, at any hour, but there will always be conversations that need a person: an angry customer, a medical question, a special request. At WhaleBiz the agent hands such conversations to a person with a short summary. Decide in advance who receives the handoffs and what happens when nobody is available.
What is an AI hallucination, and how do you reduce it?
A hallucination is a confident, wrong answer, for example an opening hour the model invented. It happens when the agent has no answer in the data and the instructions. You reduce it with up-to-date data, clear instructions and one simple rule: when the agent does not know, it says it will check and hands over to a person.
How much does an AI agent for a business cost?
It depends on what the agent needs to do and in which channels. WhaleBiz plans and what each one includes are on the pricing page of our site. When you compare proposals, also check what each vendor counts as a "message": for us a message is one reply from the agent, even when it is a reply to a voice message or a photo.

Michael Romm
Michael is the founder and CEO of WhaleBiz, leading business and marketing strategy. An expert in data (SQL, Python) and developing automation and AI solutions for businesses.