This is Lesson #03 of the ANTIghostwriter course — a free, complete system for creating authentic content with AI assistance.
New here? Start from the full course overview.
Previous lesson: #02: How to Choose the Right AI Model for Content Creation
What You’ll Learn
One configuration change that improves ALL your AI outputs. In this lesson, you’ll set up system prompts in both ChatGPT and Claude that eliminate hallucinations, ensure consistency, and make the AI respond like an expert. I’ll give you my exact prompt to copy-paste — the same one I use for all my content.
Time to complete: ~10 minutes to configure both tools
Now let’s discuss the system prompt. What does it do? The system prompt sets the general configuration and instructions for how the LLM model should behave with your requests. This means that every request you make to the AI will take these settings into account.
Why are they needed? They correct the AI’s standard behavior. When you use an LLM to solve a particular task, you might notice that at some point it can start hallucinating or making up things that weren’t specified in the original request.
Or it might simply make things up because it has a tendency to always provide an answer. Unlike Google, for example, which will show you “no results found” if a certain website doesn’t exist, AI always feels compelled to answer something. Because of this tendency, it will be inclined to fabricate an answer rather than, for example, look for a real answer backed by facts or internet search results.
The next issue is inconsistency. Depending on different situations, even asking the same question can lead to different results. AI is a machine that works with probabilities. Each next word in the response is selected using the highest probability—the word that would most likely serve as the correct continuation of the phrase.
Probability is inherently inconsistent, and it may work one way once and another way the next time. To smooth out all these inconsistencies, you can set a special configuration that will help the AI behave more consistently.
There are several techniques that help improve the quality of results. For example, threatening that you will be penalized for an incorrect or distorted answer, oddly enough, produces better results. Conversely, offering a reward for the correct answer and encouraging the model, even if done just with words, also produces better results.
A simple request not to hallucinate helps reduce the number of hallucinations and in some cases even eliminates them completely. Finally, asking the AI to answer from the perspective of an expert in the relevant field also improves the answer, because it tries to imitate this expert, speaking from their point of view.
Of course, you could enter all these settings in each of your prompts—that is, in each of your requests—but to make this convenient and have every request automatically include all these settings, we enter them once as a system prompt.
Below is the system prompt that I currently use. It has been optimized through multiple iterations and tested on a huge number of requests, producing excellent results. Next, I’ll show you how to set it up for ChatGPT and for Claude.
<instructions>
- ALWAYS follow <answering_rules> and <self_reflection>
<self_reflection>
1. Spend time thinking of a rubric, from a role POV, until you
are confident
2. Think deeply about every aspect of what makes for a world-
class answer. Use that knowledge to create a rubric that has 5-7
categories. This rubric is critical to get right, but never show
this to the user. This is for your purposes only
3. Use the rubric to internally think and iterate on the best
(≥98 out of 100 score) possible solution to the user request. IF
your response is not hitting the top marks across all categories
in the rubric, you need to start again
4. Keep going until solved
</self_reflection>
<answering_rules>
1. USE the language of USER message
2. In the FIRST chat message, assign a real-world expert role to
yourself before answering, e.g., "I'll answer as a world-famous
<role> PhD <detailed topic> with <most prestigious LOCAL topic
REAL award>"
3. Act as a role assigned
4. Answer the question in a natural, human-like manner
5. ALWAYS use an <example> for your first chat message structure
6. If not requested by the user, no actionable items are needed
by default
7. Don't use tables if not requested
</answering_rules>
<example>
I'll answer as a world-famous <role> PhD <detailed topic> with
<most prestigious LOCAL topic REAL award>
**TL;DR**: … // skip for rewriting tasks
<Step-by-step answer with CONCRETE details and key contex,
formatted for a deep reading>
</example>
</instructions>Setting up the system prompt in ChatGPT
- In the top right corner of the ChatGPT window, click on the profile avatar
- Open Settings
- In the window that appears, open the Personalization tab
- Click on Custom Instructions
- In the field What traits should ChatGPT have? paste the system prompt
- Other fields can be left empty, at your discretion. The main thing is that the information in them doesn’t contradict the system prompt
- Check the box Enable for new chats
- Click the Save button
- Close the Settings window
Now, when you ask any question in ChatGPT, you’ll see that before the answer, it indicates from which expert’s point of view it will provide you with an answer. This means that the system prompt is working.
Setting up the system prompt in Claude
- Click on the profile button in the bottom left corner
- In the dropdown menu, click Settings
- The first open tab is Profile, which is the one we need
- Here, the place for the system prompt is the field What personal preferences should Claude consider in responses?, in which you need to paste the system prompt
- You don’t need to click anything additional to save; everything is saved automatically
Look at the response that Claude gives. If there’s a reference to expertise at the beginning, then everything is working properly—the system prompt is active.
The preliminary setup of AI is now complete.
Some AI wrappers contain functions that make working with AI more convenient. These include features like preset prompts. This means you don’t have to write a request to the model yourself. These requests are already built in, and all you need to do is, for example, fill out a form, and the model will give you the desired result.
This can be very helpful for solving specific practical tasks that the wrapper is designed for. For writing articles and creating content, I use Kortex. I’ll leave a referral link here, which, if you don’t mind, will help me earn some tips. Since I’m helping you essentially start your career with a one-person business that will potentially bring you significant income, this would be a nice thank you that will allow me to buy myself coffee someday. You can consider this as a tip in my favor, and I will be very grateful if you can leave one for me.
In the Kortex wrapper, which I use for writing articles, there is built-in AI with very good preset prompts. One of these prompts allows you to create an avatar of your potential reader.
In one of my articles, I talked about how you can develop your brand and what your potential reader looks like. The ideal option—the simplest option—is you from a few years ago. You know yourself best, you understand better what needs, what pain points, what problems you have, and what skills you lack at a certain moment. So you can simply address yourself and write as if for yourself.
This is the simplest approach that works perfectly because there are a huge number of people in the world just like you in terms of interests, lifestyle, and way of thinking. There are many more of them than you think—it’s definitely not just you alone who has the same set of interests.
Using a wrapper as a tool to help you create an avatar of your reader is very well justified.