Imagine having a personal assistant who knows exactly how you work, which language you prefer, and which tasks you complete regularly. Not a generic chatbot, but a tailored helper who understands your needs. That is exactly what AI assistants enable you to do. In this article, I will show you step by step how to set up your own AI assistants and revolutionize your daily workflow.
In the previous article about AI Automations, you learned how to automate repetitive processes. Now we are taking it a step further: you will build specialized AI assistants that function like virtual team members and help you with specific tasks.
What Are AI Assistants and Why Do You Need Them?
AI assistants are specially configured versions of language models that have been optimized for specific tasks. Instead of starting from scratch every time and explaining the full context to the AI model, you store all important information once. The assistant then "knows" what it is about with every interaction.
The difference between a normal chat and an AI assistant is comparable to the difference between a stranger on the street and a long-time colleague. The stranger needs the full explanation every time. Your colleague immediately understands what you mean, knows your processes, and delivers fitting results.
Specifically, AI assistants offer you the following advantages:
- Time savings: You do not have to rebuild context every time
- Consistency: The assistant delivers consistent quality in your desired style
- Specialization: Each assistant is an expert in its field
- Efficiency: Complex tasks are completed in seconds
- Scalability: You can create any number of assistants for different areas
Typical use cases include: email assistants that reply in your style, content assistants that create texts according to your specifications, research assistants that prepare information based on your criteria, coding assistants that know your programming style, and learning assistants that explain topics at your level.
Custom GPTs: Your Own Assistant at OpenAI
Custom GPTs are one of the most accessible ways to create your own AI assistant. You need a ChatGPT Plus subscription and can then configure your own assistants in the GPT Builder.
How to create your first Custom GPT:
Open ChatGPT and click on "Explore GPTs" and then on "Create." You will land in the GPT Builder, which guides you through the process. You have two options: the guided mode (Create) or the manual mode (Configure). I recommend the manual mode because you have full control there.
In the Configure tab, you will find the following fields:
- Name: Give your assistant a clear, descriptive name
- Description: Briefly describe what the assistant is for
- Instructions: This is where the system prompt goes, the heart of your assistant
- Conversation Starters: Suggested opening questions for users
- Knowledge: Files the assistant uses as a knowledge base
- Capabilities: Web search, image generation, and code execution
A practical example: You create a "Blog Editor" Custom GPT. In the Instructions, you store your writing style, your target audience, SEO guidelines, and the structure of your blog articles. In Knowledge, you upload example articles. From now on, you just need to tell the assistant the topic, and it delivers articles in exactly your format.
Tips for better Custom GPTs:
The more precise your Instructions are, the better the assistant works. Describe not only what it should do but also what it should avoid. Provide examples of good and bad results. Define the tone, structure, and length of outputs. Upload relevant documents to the Knowledge section so the assistant can access real knowledge.
Claude Projects: Assistants with Anthropic
Claude Projects from Anthropic offer an alternative approach to AI assistants. They work similarly to Custom GPTs but have some unique features that make them particularly attractive for certain use cases.
What makes Claude Projects special:
Claude Projects allow you to upload extensive documents as project context. Claude has a particularly large context window and can therefore process long documents completely. This makes Projects ideal for tasks where a lot of background knowledge is needed.
How to set up a Claude Project:
- Open Claude and create a new Project
- Give the Project a meaningful name
- Write Custom Instructions (your system prompt)
- Upload relevant files as Project Knowledge
- Start a new chat within the Project
The big advantage: Every chat within the Project automatically has access to all uploaded documents and the Custom Instructions. You can have different conversations, and Claude always knows the full context.
Practical example "Contract Assistant":
You upload your terms and conditions, contract templates, and legal guidelines as Project Knowledge. In the Custom Instructions, you define that Claude should act as a legal pre-reviewer who checks contracts for critical clauses and makes improvement suggestions. From now on, you can load any new contract into the Project and receive a well-founded analysis immediately.
Keep in mind: Claude does not replace a lawyer. But it can provide a valuable initial assessment and help you ask the right questions before consulting a legal professional.
System Prompts: The Heart of Every AI Assistant
The system prompt is the DNA of your AI assistant. It defines personality, behavior, knowledge, and boundaries. A good system prompt makes the difference between a mediocre and an excellent assistant.
The anatomy of a perfect system prompt:
An effective system prompt consists of several building blocks that together create a clear picture:
1. Role definition: Who is the assistant? Define a clear identity. Example: "You are an experienced social media strategist specializing in B2B marketing in the German-speaking market."
2. Task description: What should the assistant do? Be specific. Example: "Your task is to create social media posts for LinkedIn that convey expertise and generate engagement."
3. Context and knowledge: What background knowledge does the assistant need? Example: "The target audience is decision-makers in mid-sized companies. The tone is professional but approachable. We address our community informally."
4. Format specifications: What should the output look like? Example: "Each post has a hook (first line), a body with 3 to 5 paragraphs, and a call-to-action. Maximum length: 1,300 characters."
5. Restrictions: What should the assistant not do? Example: "Do not use emojis at the beginning of sentences. Avoid superlatives. Do not make unsubstantiated claims."
6. Examples: Show the assistant good results. Example: Include 2 to 3 sample posts that match your desired style.
Here is a complete system prompt template:
"You are my personal email assistant. Your task is to write professional emails on my behalf. My style is friendly, direct, and solution-oriented. I address colleagues informally and external contacts formally. Every email starts with a personal greeting and ends with my standard sign-off: Best regards, [Name]. Keep emails short (maximum 150 words). Do not use phrases like 'I hope this email finds you well.' When I tell you who the email is for and what it is about, create a complete draft."
Setting Up Assistants for Different Task Areas
The true power of AI assistants unfolds when you create specialized helpers for different areas. Here I show you proven configurations for the most common use cases.
The Research Assistant:
This assistant helps you collect, evaluate, and summarize information. Configure it so that it always cites sources, distinguishes between facts and opinions, and presents information in a structured format. Enable web search so it can find current information.
System prompt core: "You are a critical research assistant. Your task is to collect information on a topic and prepare it objectively. Always distinguish between established facts and interpretations. Cite sources. Structure your results with headings and bullet points. Point out contradictory information."
The Writing Assistant:
An assistant that writes texts in your style. Upload example texts as a knowledge base and describe your style precisely. Define target audience, tone, and format specifications. This assistant gets better over time when you give it feedback.
The Analysis Assistant:
Configure an assistant that analyzes data and information. It should proceed systematically: first understand the data, then identify patterns, and finally derive action recommendations. Enable the code interpreter function so it can also visually prepare data.
The Learning Assistant:
A personal tutor that explains topics at your level. Define your prior knowledge, your preferred learning style, and the subject areas. The assistant should start with simple explanations and go deeper when needed. It should regularly ask comprehension questions and use practical examples.
The Project Management Assistant:
This assistant helps you plan projects, prioritize tasks, and maintain an overview. Configure it with your project methodology (for example Scrum, Kanban, or GTD) and your typical project structures. It can create schedules, identify risks, and write status reports.
Best Practices for AI Assistants
After setup, the real work begins: optimization. Here are the most important best practices I have gathered in practice.
1. Improve iteratively:
Your first system prompt will not be perfect. And that is completely fine. Start with a basic version and refine it based on the results. When the assistant does something wrong, adjust the system prompt. Document what changes you made and why.
2. Use versioning:
Save different versions of your system prompt. This way, you can always return to an earlier version if a change did not have the desired effect. A simple text document with date and version number is sufficient.
3. Test with different scenarios:
Do not test your assistant only with the ideal case. Try edge cases: What happens with vague requests? How does it react to topics outside its area? How does it handle contradictory instructions? A good assistant should respond sensibly even in unexpected situations.
4. Keep the knowledge base up to date:
When your processes, guidelines, or information change, update your assistant's knowledge base as well. An assistant with outdated information can do more harm than good. Schedule regular reviews, for example once a month.
5. Set clear boundaries:
Define in the system prompt what the assistant should not do. Should it ask for clarification instead of guessing when uncertain? Should it refer to a human on certain topics? Clear boundaries prevent unwanted results and increase trust in the assistant.
6. Build feedback loops:
Give your assistant regular feedback. Tell it when a result was particularly good or bad. Many platforms allow this feedback to flow into optimization. The more you work with your assistant, the better it becomes.
7. Consider data privacy:
Do not upload sensitive personal data to the knowledge base unless the platform guarantees appropriate data protection. Check the provider's privacy policy and consider which information you want to entrust to the assistant.
Advanced Techniques for AI Assistants
Once you master the basics, you can take your assistants to the next level with advanced techniques.
Multi-step workflows:
Configure assistants that break complex tasks into multiple steps. Example: A content assistant that first creates an outline, then develops the chapters, and finally performs an SEO review. Describe the workflow in the system prompt and the assistant executes it automatically.
Assistant chains:
Use multiple specialized assistants in sequence. The output of one becomes the input of the next. For example: Research assistant delivers facts, writing assistant creates an article from them, SEO assistant optimizes the metadata. This way you leverage each assistant's strengths.
Dynamic system prompts:
Create system prompts with placeholders that you can adapt to different situations. Example: "You write for the target audience [AUDIENCE] in the tone [TONE] about the topic [TOPIC]." This way, you can use one assistant flexibly for different contexts.
API integration:
If you are technically inclined, you can integrate AI assistants into your own tools via APIs. OpenAI offers the Assistants API, Anthropic the Messages API with system prompts. This way, you can embed assistants directly into your workflows, for example in Slack, in your CRM, or in custom web applications.
Automatic quality control:
Set up a second assistant that reviews the results of the first. The "reviewer" evaluates the output based on predefined criteria and provides improvement suggestions. This four-eyes principle significantly increases quality and catches errors before they reach you.
Exercise: Configure Your First AI Assistant
Now it is your turn! In this exercise, you will create an AI assistant for a task you complete regularly.
Step 1: Choose a task
Choose a recurring task that you complete at least once a week. This can be email writing, content creation, research, data analysis, or something entirely different. What matters is that you know the task well and have experience with it.
Step 2: Write the system prompt
Use the Prompt Generator at optiprompt.io to create a professional system prompt. Define role, task, context, format, and restrictions. Follow the structure you learned in this article. Start with a clear role definition and then become increasingly specific.
Step 3: Create the assistant
Create your assistant on the platform of your choice (ChatGPT, Claude, or another). Add the system prompt and upload relevant documents as a knowledge base. Choose appropriate conversation starters and enable the necessary capabilities.
Step 4: Test and optimize
Ask your assistant at least five different requests that are typical for your chosen task. Evaluate the results and adjust the system prompt accordingly. Note what works well and what needs improvement.
Step 5: Document
Create a brief documentation of your assistant: name, purpose, system prompt (current version), knowledge base files, and experiences from testing. This documentation helps you with future updates and when creating additional assistants.
Bonus task: Create a second assistant that reviews the results of the first. How does the quality change? Which errors are detected that you might have missed?
Summary and Outlook
AI assistants are powerful tools that can massively boost your productivity. You have learned how Custom GPTs and Claude Projects work, how to write effective system prompts, and how to set up assistants for different task areas. With the best practices and advanced techniques, you can continuously improve your assistants.
The most important point: just start. Your first assistant does not have to be perfect. With each iteration, it gets better, and you develop a sense for what works and what does not. After just a few adjustments, you will notice a significant difference in your work efficiency.
In the next article, we will dive into Data Analysis for Beginners. You will learn how to use AI to understand data, identify trends, and make data-driven decisions. Even if you are not a data expert, you will see how AI makes this important area accessible to you. Until then: have fun setting up your personal AI assistants!