Written by Martin Ågren - May 26, 2026
A "custom AI chatbot" (or "custom AI assistant") is:
A tailored (customized) version of an AI chatbot.
Example: Inside ChatGPT, you can create TWO different types of custom AI assistants: Projects (project-folder assistants) and Custom GPTs.
Save time and mental energy by letting AI behave in a consistent way, so you can avoid giving the same instructions to AI at the beginning of every new chat.
Get higher-quality responses from AI, because you have optimized the AI chatbot in advance.
Share your assistant with others (partners / colleagues / customers) to provide value to them, so they too can get valuable help from your customized AI assistant.
Here are general tips for how to create a customized AI assistant. In other words: more specific tips for ChatGPT itself can be found on the page that describes projects and GPTs respectively. But here are general tips that are not connected to any specific AI service:
Role
Assign a role to the assistant when you create it. Example of role assignment:
"Your role: Act as a senior, professional advisor for our management team".
Assigning a role to AI usually leads to more relevant, higher-quality output.
Behavior
Make sure you provide the assistant with instructions about the desired behavior. Example: Should it generally provide short answers? Should it avoid including emojis in its responses to the user?
Uploading files versus writing text
You can usually choose whether you want to write the instructions/information when creating the assistant, or if you want to upload file(s) containing the instructions/information. Writing the text directly can provide more accurate output, but sometimes it can be convenient to upload file(s) with text content, for example if you already have a document containing information that is relevant to the AI assistant you are creating.
Information the assistant needs
When creating your customized AI assistant, you should provide the necessary information. For example, you can explain what type of organization the assistant will work for when creating it, so you do not have to explain this in every new chat you start with the assistant in the future. Think of it as a new colleague who has never set foot in your office before. Are there documents you need to upload to the assistant for it to do a good job, or is it enough to provide the necessary information directly when creating it?
Guardrails (boundaries)
Consider limiting the assistant’s behavior using so-called "guardrails" (in other words, boundaries for its behavior). You may therefore need to instruct your new AI assistant regarding what it is NOT allowed to do. Although AI companies usually already have basic guardrails in place for their AI models and chatbots, those preset safeguards are broad and not specifically adapted to your organization’s unique needs. Therefore, you may need to define boundaries yourself regarding output. Clear guardrails increase predictability and make the assistant safer to use within the organization.
Examples of guardrails:
“Your assistance should be limited to professional matters only. Avoid topics of a private nature.”
English
When creating your customized AI assistant: write your instructions in English. LLMs are usually trained on significantly more English material than Swedish material. Therefore, instructions written in English usually lead to higher-quality output. This applies even if the AI assistant is mainly expected to communicate in Swedish with the user. However, there are exceptions: in some cases, LLMs may understand instructions best in Swedish if most of the available material about the topic exists in Swedish. Example 1: Evert Taube. Example 2: The Right of Public Access.
Iterate
Optimize output quality by iterating: test, improve the assistant, test again, improve it further (and so on) until the AI assistant generates output of optimal quality based on your needs.
GDPR
Make sure you comply with GDPR before, for example, uploading files containing personal data. Has your organization signed DPA agreements with the AI provider? Should you anonymize the personal data ("Employee no. 14" and so on)? Does your organization have policy documents providing guidance regarding what type of information you should NOT enter into an AI tool? How can you ensure compliance with GDPR when creating and using the assistant?
Training data
Make sure training data is turned off before you start entering instructions or uploading files to the assistant.
Trade secrets
Regarding text and/or documents you plan to provide to the assistant: does the text or files contain confidential/sensitive information? If so: how should you handle it? Does your organization have policy documents that provide guidance regarding what type of information you should NOT enter into an AI tool? Even if training data is turned off and GDPR is followed, there is still a general risk of cyberattacks in cloud-based solutions.
Different options:
Project folder assistant
Custom GPT
Custom instructions
A customized AI assistant can be very useful for you. You can avoid repetitive tasks (such as frequently writing the same type of instructions to the AI). You can choose to have an assistant for yourself and/or share your assistant with others, for example within the management team.