This guide fits a business that wants better answers on the website and a practical decision about the kind of solution. It explains the difference between a WhatsApp button, a menu bot, and an AI chatbot, and the questions to check before connecting information or systems.Connecting documents or site pages to a bot is not necessarily model training. It usually means using that information as a source or context for an existing AI service, together with quality checks, permissions, and clear rules for cases where there is not enough information for a satisfactory answer.
1. A WhatsApp button, a menu bot, and an AI chatbot
A WhatsApp button only moves the user into a conversation with the business. It is simple, clear, and fits when fast human handling is preferred. A menu bot shows fixed options such as opening hours, choosing a service, or leaving details.An AI chatbot is different because it receives a free-form question and tries to answer from sources defined in advance. It fits when the wording of questions changes and there is enough written information to rely on, but boundaries and checks still need to be planned. Sometimes a short menu or a form is enough, and not every site needs AI.
2. Possible information sources
Information sources can be site pages, service documents, public procedures, catalogs, frequently asked questions, or training material. The clearer, more current, and better written the information is, the easier it is for the bot to give a useful answer.If the information is scattered, contradictory, or missing, the bot may give partial answers. Part of the work is therefore not only a technical connection, but also arranging sources, defining topics, and deciding what does not enter the answer set.
- Information that has not been updated for a long time and needs a check before it is connected.
- Topics you do not want the bot to answer.
- Sources that should be shown or cited to the user when that is appropriate.
3. How to keep the information current
A good chatbot depends on the information that feeds it. If a service page changed, a document was updated, or a policy changed, there needs to be a way to update the sources and check that the bot is not still relying on old information.Options include a periodic scan of pages, uploading updated documents, a simple admin interface, or a defined manual process. The choice depends on how much information there is, how often it changes, and who in the business is responsible.
- Who in the business is responsible for updating the information?
- How often does the content change?
- How is a source removed once it is no longer relevant?
- How are answers checked after a significant update?
- Is a version history or approval before publishing needed?
- Is there seasonal or temporary information that should expire automatically?
4. Public information versus information that requires authentication and permissions
It is relatively straightforward to plan a bot that answers from public information on the site: services, opening hours, a work process, or general questions. Personal or internal business information needs a completely different plan: sign-in, permissions, a record, and access limits.If the bot is supposed to answer a customer about an order, an invoice, or a personal status, it needs to know who the user is and what they are allowed to see. An employee, a customer, and a manager do not necessarily see the same information. Without a suitable permission mechanism, the bot should not expose information that is not public.
5. What to do when there is not enough information for an answer
The bot should not be made to invent an answer. If there is no suitable source, the information is missing, or the question is unclear, the reply should say so plainly and offer a next step: a more focused question, a link to a relevant page, or a handoff to a person. Questions that keep coming back without a suitable source should be marked so the content can be filled in.Define in advance the topics the bot does not answer, such as legal advice, a binding price, a medical decision, or an action that requires a human check. Those boundaries matter as much as the ability to answer.
6. Handoff to a person, and what goes with it
An AI chatbot also needs a way to hand a conversation to a person, for example when there is no answer, a customer is upset, the request is personal, or the action is sensitive. The handoff can be a form, WhatsApp, a support system, or a CRM, according to the business process. The person should receive context, not an empty message.The information that moves can include the details the user left, a summary of the conversation, the last question, the source page, and the reason for the handoff. If sensitive information is involved, decide what may be passed and where.
7. An answer is not an action in a system
There is a difference between a bot that explains what to do and a bot that performs an action: opening a ticket, updating customer details, changing a status, or sending a message. Answering from information is one layer. An action in a system is automation that needs an API, permissions, tests, and failure handling.You can start with a bot that answers and hands off to a person, and later add limited actions when there is a reason. Even then, define which actions need human approval before they run, record an action that was performed, and prevent the same action from running twice. There is no promise that an action can be performed in every system.
8. A ready-made solution or custom development
A ready-made solution fits when you need simple chat on the site, public information sources, few adjustments, and a basic handoff to a person. Custom development is justified when there are permissions, several information sources, connections to systems, business logic, or a need for deeper control.There is no single right choice for everyone. An off-the-shelf solution can reduce complexity at the start, and custom development can be right when the bot becomes part of an important business process or is connected to protected information.
| Consideration | Ready-made solution | Custom development |
|---|---|---|
| Information sources | Site pages, an FAQ, and simple documents. | Several sources, a special information structure, or a need to control the source of an answer. |
| Permissions | Mainly public information. | Information by user, customer, employee, or manager. |
| Handoff to a person | A form or basic WhatsApp. | A conversation summary, tagging, CRM assignment, and handoff rules. |
| Actions in a system | Usually limited or unavailable. | Possible only with an API, permissions, tests, and a record. |
| Quality control | Manual checks and basic rules. | A test set, answer boundaries, logs, and improvement based on use. |
| Maintenance | Simpler while needs stay basic. | Requires responsibility for connections, information sources, and usage costs. |
9. Quality checks and the limits of the solution
Before turning it on for the site, test real questions: simple questions, different wordings, questions with no answer, questions outside the scope, and questions that need a person. The goal is to find where the bot helps and where it should stop.After launch, keep watching questions that were not answered well, missing sources, and usage costs. An AI chatbot is a product that needs maintenance, not a component you switch on and forget.
10. What affects the cost
The cost of an AI chatbot is not only the setup. It can include arranging information sources, building an interface, a handoff to a person or a CRM, integrations with systems, quality checks, usage costs for AI services, and maintaining the information sources over time.A cost estimate depends on the actual need: how much information is connected, how many users are expected, whether there are permissions, whether actions are performed in a system, and what level of control is required.
Hypothetical example only
A conversation in which one question is answered and one is handed to a person
- A visitor on the site asks: "What does MVP development include?" The bot finds an answer on the service page and briefly explains that it is a focused first version with one core action.
- The bot offers a link to the MVP page and asks whether the user wants to leave details for a conversation.
- The user then asks: "How much would a project like mine cost?" The bot does not have enough information to give an estimate, and it does not invent a price.
- The bot asks for a short description of the product and offers a handoff to a person. If the user agrees, the name, phone number, conversation summary, and the page the inquiry started from are passed along.
- The person receives context: the user asked about an MVP and wants a fit check, but no price or automatic timeline was promised.
A checklist before choosing a solution
You can arrive with partial answers. The list gathers what is worth preparing before choosing a solution.- A list of information sources: site pages, documents, and frequently asked questions.
- A decision about who is responsible for updating the information.
- A split between public information and information that requires authentication and permissions.
- Topics the bot does not answer and instead hands off to a person.
- The handoff channel and the information that goes with it.
- Whether the bot only answers or is also supposed to perform actions in a system.
- Test scenarios before it goes live on the site.
- An understanding of setup, integration, usage, and maintenance costs.
Common questions
Can an AI chatbot answer without mistakes?
We do not promise full accuracy or the complete prevention of mistakes. Information sources, answer boundaries, quality checks, and a handoff to a person are planned for when there is not enough confidence.
Does connecting documents mean a new model is being trained?
Not necessarily. The information is usually used as a source or context for an existing AI service. Training a model is a different process and is not the default. Connecting documents is not necessarily model training.
Can the bot be connected to customers' personal information?
Only if there is a suitable authentication and permissions mechanism, and it is defined exactly what each user is allowed to see. Without a clear permission, the bot should not expose personal information.
Want to check whether an AI chatbot fits your website?
Send which questions the bot should answer, which information sources it should use, and what happens today when a visitor needs a person. We will check whether a ready-made solution or custom development is the better fit.Want to check whether an AI chatbot fits your website?
Send which questions the bot should answer, which information sources it should use, and what happens today when a visitor needs a person. We will check whether a ready-made solution or custom development is the better fit.