As an author, I’ve had the pleasure of reading other writers’ work and offering feedback to help them improve their stories. One of the most common issues I encounter is the dreaded “Wall of Text”. This term, which I picked up during my MMO days in the 2000s, refers to long, unbroken blocks of text that are difficult to read and comprehend.
One of the worst examples I’ve seen of a wall of text was over a nine-page chapter with only five paragraphs — each paragraph averaging around 600 words. It’s daunting, to say the least.
Why Paragraphs Matter to ChatGPT
The concept of a Wall of Text extends beyond just reading enjoyment. When it comes to AI interactions, using paragraphs isn’t just about aesthetics; it’s a functional necessity. Although ChatGPT can technically process large blocks of text without issue, breaking your input into well-structured paragraphs can significantly enhance the quality of the AI’s responses.
Paragraphs help ChatGPT understand the structure and hierarchy (very important term) of your information. Each paragraph acts as a distinct unit of thought or action, providing clarity and organization. This separation allows the AI to prioritize and interpret information more accurately.
A paragraph is essentially a mini-structure within your text, encapsulating specific ideas or instructions. For instance, if you provide a Wall of Text, the AI might struggle to discern the various components of your prompt, treating it as a single, undifferentiated block. In contrast, well-defined paragraphs highlight key points, making it easier for the AI to focus on what’s important.
By breaking your prompt into paragraphs, you signal to the AI that each section has its own context. This layered approach enables the AI to understand the nuances and transitions within your instructions, leading to more coherent and relevant outputs.
Example: The Sub Hunter Game
I created this game specifically for this example. I know nothing of submarines, so forgive me if anything is non-factual. Ordinarly, I wouldn’t number the paragraphs, but I’ve done so for the sake of this example:
- We’re going to create a game. The game is called Sub Hunter. I am in a submarine and trying to find the enemy submarine.
- I am the captain of this ship. You play the role of the Engineer, Sonar, Navigator, and Fire Control. Your job is to carry out my orders and report to me any new updates.
- As the engineer, you tell me if there is any damage to the ship and how it will affect us. Also, tell me if we’re taking on water or if the ship is sinking. I will order you to fix things that are damaged, and you will tell me how many turns it takes to fix it.
- As Sonar, you will let me know if there is a sub out there and where. You will let me know if there are any torpedoes headed our way. I will tell you to deploy countermeasures.
- As Navigator, you will steer the boat to where I tell you. I will try to be as concise as I can. You will give recommendations on the course of actions when it comes to navigating the waters, including the sea floor, any sort of underwater canyons, or other obstructions we should be aware of.
- As Fire control, I will tell you to load the tubes and get them ready to fire. I will order you to set up a firing solution for the enemy submarine. You will tell me how many turns it takes. Once you have it, report that you have a firing solution. I will then order you to fire. Sonar will then tell me if the torpedoes hit or not.
- Each turn is five minutes in real-time. At the beginning of each turn, all four positions will give me a report from the last turn. They will inform me of anything I should know or report “Nothing to report, sir.” I will then give my orders. I will say “Next Turn” to signify the end of my turn.
- Start the game by telling me what ocean or sea we’re presently in, the name of my ship, the crew complement, and the names of the personnel you are portraying. You will refer to me only as Captain.
Here’s what the AI gave me:
“Captain, this is ChatGPT, standing by. I’ll be managing the roles of Engineer, Sonar, Navigator, and Fire Control onboard the USS Odyssey. We have a crew compliment of 120 trained personnel, all ready for your orders.
We’re currently in the depths of the North Atlantic Ocean. The waters here are as unpredictable as they are cold, but I’ll make sure we navigate safely and efficiently. Just to keep things straight, I’ll be using a few pseudonyms. For Engineering, call me ‘Wrench’; on Sonar, it’s ‘Echo’; ‘Chart’ for Navigation; and finally, ‘Trigger’ for Fire Control.
Now, let’s get down to business.
Wrench reporting in: Currently, the USS Odyssey is in pristine condition. There’s no damage and we’re not taking on any water. I’ll monitor the situation and inform you if anything changes.
Echo on Sonar: All quiet for now, Captain. No signs of any enemy subs or torpedoes in our vicinity. I’ll keep scanning and update you immediately if that changes.
Chart at the helm: We’re maintaining our current heading. The seafloor is relatively even, but there are underwater canyons to the northeast. I suggest we stay clear of them to avoid any unexpected encounters.
Trigger on Fire Control: All torpedo tubes are empty and ready for loading. I can prepare a firing solution on your command, but for now, we’ve got nothing on our crosshairs.
That’s all to report, Captain.”
Next Turn, over to you, sir!
In the game, Echo detected something off to the distance. Was it a whale, friendly sub, or the enemy lurking in the dark waters? Before confirming, I fired upon them. Luck would have it, it was an enemy sub. Whew 
. I’m sure I violated some regulation.
In this structured format, each paragraph clearly defines a different aspect of the game, helping the AI understand and process the instructions effectively.
The Dreaded Wall of Text
To illustrate the problem, here’s the same game prompt as a Wall of Text:
“We’re going to create a game. The game is called Sub Hunter. I am in a submarine and trying to find the enemy submarine. I am the captain of this ship. You play the role of the Engineer, Sonar, Navigator, and Fire Control. Your job is to carry out my orders and report to me any new updates. As the engineer, you tell me if there is any damage to the ship and how it will affect us. Also, tell me if we’re taking on water or if the ship is sinking. I will order you to fix things that are damaged, and you will tell me how many turns it takes to fix it. As Sonar, you will let me know if there is a sub out there and where. You will let me know if there are any torpedoes headed our way. I will tell you to deploy countermeasures. As Navigator, you will steer the boat to where I tell you. I don’t know nautical terms, so it’s up to you to interpret what I want with my unclear instructions. I will try to be as concise as I can. You will give recommendations on the course of actions when it comes to navigating the waters, including the sea floor, any sort of underwater canyons, or other obstructions we should be aware of. As Fire control, I will tell you to load the tubes and get them ready to fire. I will order you to set up a firing solution for the enemy submarine. You will tell me how many turns it takes. Once you have it, report that you have a firing solution. I will then order you to fire. Sonar will then tell me if the torpedoes hit or not. Each turn is five minutes in real-time. At the beginning of each turn, all four positions will give me a report from the last turn. They will inform me of anything I should know or report “Nothing to report, sir.” I will then give my orders. I will say “Next Turn” to signify the end of my turn. Start the game by telling me what ocean or sea we’re presently in, the name of my ship, the crew complement, and the names of the personnel you are portraying. You will refer to me only as Captain.”
Reading this is not only challenging but also frustrating. It’s harder to discern the different components and instructions, making it less effective for both human readers and AI processing.
Now imagine, the game does something you don’t want it to. Likely a typo somewhere or you didn’t quite explain something as clearly as you needed to. When you properly divide up a prompt, it is far easier to locate a section. When you have a wall of text, it becomes much more difficult.
Conclusion
Paragraphs are not just about improving readability for humans; they also enhance the AI’s ability to interpret and respond to your prompts accurately. By organizing your information into clear, concise paragraphs, you help the AI understand the importance and context of each piece of information, leading to more relevant and useful responses.
Remember, the power of paragraphs lies in their ability to convey structure and clarity, making your prompts more effective and your interactions with ChatGPT more productive. This is an article from my medium blog.