November 12,2025

How to Write a Prompt for a Telegram Chatbot with Domino CRM

Domino CRM makes it easy to get the most out of AI in chatbots. The platform includes a scenario block called AI Assistant. To integrate it:

• Go to the AI Assistant section in our chatbot builder
• Give your AI Assistant a name
• Write the prompt for the chatbot
• Upload one or more knowledge base files
• Set the creativity level for the AI-powered Telegram chatbot
• Finalize settings, then place the AI Assistant at the right point in your scenario

To speed things up, you are free to even let the built-in AI draft the Telegram chatbot prompt for you. But what if you want to control each and every applicable detail personally?

This guide will explain how to work with chatbot prompts to achieve results that are both reliable and repeatable.

Practical Techniques and Tips to Try 

You can maximize the effectiveness of your chatbot prompt by using:

• Roles
• Examples in the request
• Reverse dialogue
• Reflection
• Self-check and practice

Let’s look at each in turn.

Roles in a Telegram Chatbot Prompt

A common way to guide model outputs is to specify the role from which the language model should act. This adds context and expertise and sets the right point of view.

Example: ‘You are a business consultant with 15 years of experience in HoReCa. Create a step-by-step business plan for a coffee shop.’

Specifying a role tells the model what stance to take and how much detail to include. If you say the response should come from a marketer, teacher, financial analyst, or student, the model adapts style and content accordingly. You can also set the audience role, e.g., ‘Answer as if you’re speaking to a five-year-old.’

This approach helps when you want to:

• Emphasize a professional tone or level of expertise
• Adapt the output to a specific audience
• Get multiple perspectives on the same task

Tip: Try the same prompt from different roles, for example, from a CFO and a marketer. Comparing answers reveals different angles and helps you choose the best option. 

Examples in the Request 

How many examples you provide determines how accurately the language model understands your task and delivers the desired result. There are three main approaches:

1. Zero-Shot Prompting (without examples). In a zero-shot setting, the model gets only the task description and responds based on its existing knowledge. This is the fastest approach, but the output may be too generic or inconsistent.

Prompt: ‘Determine the sentiment of the review as ‘positive,’ ‘neutral,’ or ‘negative.’

2. One-Shot Prompting (with one example). In a one-shot setup, you show the model what the answer should look like. For example, through a template, an outline, or a style sample. This gives it a reference for format and tone.

Prompt: ‘Determine the sentiment of the review as ‘positive,’ ‘neutral,’ or ‘negative.’ 
Example aka shot: The movie was very long, but overall not bad.
Sentiment: positive

3. Few-shot / Multi-shot Prompting (with several examples). In a few-shot or multi-shot approach, you provide multiple (usually 3–5) examples so the model can recognize patterns, context, and expected style. This method is ideal for chatbot training, client communication, or educational content generation, where a single example may not capture all variations.

Prompt: ‘Determine the sentiment of the review as ‘positive,’ ‘neutral,’ or ‘negative.’

Example aka shot:
Review: Great movie, we loved it!
Sentiment: positive

Review: An okay movie, worth one watch.
Sentiment: neutral

Review: The movie was boring, and the nachos were cold.
Sentiment: negative

The more relevant examples you include, the more precisely the model grasps the logic, tone, and structure of your desired answer.

Reverse Dialogue for Prompt Creation

Reverse dialogue is a technique where the model takes the initiative and asks clarifying questions to gather missing inputs. This lets you shape the task gradually instead of trying to describe everything in a single prompt.

For example, if AI needs to prepare a business plan, you can write: ‘Your task is to create a business plan for a specified type of business. Ask one question at a time to collect the necessary information and, once you have the required inputs, produce the business plan.’

Be sure to specify that questions should come one at a time; otherwise the model may ask a dozen at once, making the dialogue unwieldy. With a step-by-step flow, the interaction is smooth:

• The model asks
• You answer
• The context gets refined

At any point, you can stop and ask for a final answer based on the data collected so far. Reverse dialogue is especially helpful for structuring ideas, gathering requirements, or surfacing missing details.

Reflection for Chat Prompts

Reflection is a technique in which a model evaluates and improves its own response.

For example: “Assess the strengths and weaknesses of your answer on a scale from 1 to 10, and suggest improvements for any points rated below 7.”

This approach can be useful when interacting with AI, but it cannot be built directly into the initial prompt. It can only be applied during the interaction stage with the assistant. At this stage, the model analyzes its text, identifies what can be refined, and offers an updated version. This helps improve responses quickly without additional requests.

Self-Check and Practice

To cement these techniques, formulate a task for your own Telegram chatbot and walk through it using the methods above. Start with a simple prompt, then add:

• Examples to define the structure
• A role to set the context
• Reverse dialogue to collect missing data

And of course, if you are creating an assistant for yourself—as a helper for your own projects—reflection will help improve the quality of the model’s responses over time. 

This is the real design of the language model’s thinking, step by step, like a live dialogue.

As a result, with a well-written prompt for your chatbot, the AI Assistant in Domino CRM will deliver more accurate, logical, and useful answers. And you will be able to use AI not just as a text generation tool, but as a full-fledged partner in day-to-day work!