What this landing page consists of

A working version has four parts.

1. The landing page itself. It shows the offer, prices, examples, stages and terms. AI should not replace a clear page.
2. AI chat. It finds answers in company materials, asks clarifying questions and helps determine the right next step.
3. Lead module. It saves the contact, source and conversation history, then sends the enquiry into the workflow.
4. Web panel. It is used to manage knowledge bases and AI and manager schedules, and stores leads and conversations.

Remove any one of these parts and you get either a conventional page, an uncontrolled chat, or a lead that a manager has to investigate again from the beginning.

The customer journey

This is what it looks like without technical language.

1. A person opens the landing page from search, an advertisement, a business listing or a direct link.
2. They see what you do, who it is for and on what terms.
3. They open the chat and ask a question in their own words.
4. AI searches the knowledge base and answers from the price list, services, terms, FAQ or documents.
5. If the task is a fit, the person leaves a name, phone number or another contact.
6. The system saves the lead in the web panel and can send it to Telegram or Bitrix24 together with the conversation history.
7. The manager sees the context and continues with the specific task.

Not every “hello” becomes a lead. An enquiry is handed over after clear interest or when the customer leaves contact details. A repeated enquiry can be marked as a duplicate so that the same contact is not sent to the manager many times.

Where AI gets its answers

A general language model does not know a particular company's current prices, deadlines and rules. The chat is therefore connected to a knowledge base: a price list, service descriptions, frequent questions, terms, procedures and other verified materials.

For each request, the system finds relevant passages and gives them to AI for the answer. This is called RAG. I explained the approach in detail in “A RAG agent for AI customer support”.

RAG reduces the risk of invented answers, but does not make AI infallible. I therefore configure rules to:

• answer from retrieved materials instead of the model's general knowledge;
• never invent prices, deadlines or terms;
• ask for clarification when context is missing;
• say directly when the knowledge base has no answer;
• hand a complex question to a person.

After a price list or document is replaced, the chat must be retested on key questions. A knowledge base alone, without testing, does not guarantee correct answers.

Web panel: what can be changed without code

The initial connection and system logic are configured by a developer. But you should not need a developer for every price list or schedule update.

My web panel lets you:

• create several separate knowledge bases;
• upload, replace and delete documents;
• change AI instructions and tone of voice;
• select an AI model and provider;
• set schedules by weekday and by individual date;
• define separate AI and manager hours for every connected channel;
• inspect leads, contacts, sources, conversation history and handover errors;
• add statuses and notes to leads;
• export conversations and leads for a selected period.

This is not simply “upload a PDF and forget it.” The panel shows what happens to enquiries, and the knowledge base can be updated as the business changes.

How AI hands a conversation to a person

The most dangerous setup is one in which AI and a manager answer the customer at the same time. Every channel therefore needs a rule: either AI or a person controls the conversation.

On a website, the basic version works like this: when the customer is ready to discuss the task or leaves contact details, the enquiry is saved and sent to a manager. If the manager must continue in the same channel, a CRM open line is connected.

This already works in my Bitrix24 open-lines module: when a manager joins the conversation, AI becomes silent. Automation can return after the manual stage ends, according to the configured scenario.

The same principle can be used in MAX, Telegram and other channels. I covered MAX business features in a separate guide.

What data passes through the chat

If the chat collects a name, phone number, email or other data, it cannot be treated as an ordinary website button. Before launch, decide:

• where conversations and leads are stored;
• which part of the request is sent to the AI model;
• who can access the history;
• how long the data is retained;
• how consent to process contact data is obtained.

Unnecessary personal data should not be sent to an external AI model if it is not needed for the answer. The page also needs a clear privacy policy and consent mechanism. The exact design depends on the hosting, model, CRM and the business's requirements.

Who benefits from this setup

A landing page with an AI consultant makes sense when:

• a service has many terms, options and frequent questions;
• customers arrive in the evening or at weekends;
• managers repeat the same answers;
• the task needs clarification before asking for a phone number;
• the lead must immediately reach the responsible person instead of sitting inside a separate chat.

AI is not necessary for everyone. If you have three simple services, one price and a short form, a good landing page may work perfectly without chat. AI should also not be connected if the company is not ready to keep its price list and knowledge base current.

How to tell whether the chat is useful

Instead of borrowing someone else's conversion percentages, I would track the business's own figures:

• how many visitors opened the chat;
• how many conversations reached a clear request;
• how many people left contact details;
• how many leads reached a manager;
• which questions remained unanswered;
• the cost of one handed-over lead.

This shows whether the chat solves a real business problem. If it merely holds long conversations but neither answers questions nor hands over enquiries, it is a polished widget rather than a working tool.

Honest costs

This setup has one-off and ongoing costs.

The one-off part covers the landing-page structure and development, AI consultant configuration, knowledge-base preparation and integrations. Ongoing costs include hosting, data storage, AI requests and embeddings for document search. Support, new scenarios and integrations are added as needed.

There is no single honest “chat price per month.” Cost depends on the model, conversation length, number of documents, traffic and connected channels. These costs should be shown before launch, not after the first invoice.

See it live

This setup is running on my RM Systems website. Open the chat and ask, for example, how much a landing page with an AI consultant costs, what materials are needed to start, or whether Bitrix24 can be connected. The answer will be built from materials that I manage through the web panel.

If you need a similar setup, see the scope of work in the catalog. A precise price cannot honestly be derived from one description: first we need to understand what already exists on the website, which materials are ready, where leads must go and who will maintain the knowledge base.