Four Limitations of the AI Content Writing Workflow

1. Inconsistent performance: There is no specified rule book. A part of the process is to figure out the process. It involves figuring out the prompts, determining which tasks should be done by humans, and the degree of machine dependency.

2. No fine editing: If there is a section of an article or a sentence in a paragraph that needs to be rephrased, LLMs suck at that. Either they will mess it up even more or they will make additional unsolicited changes that will give you gamer’s rage.

3. Bloated tech stack: You will have to use multiple LLMs as ChatGPT can’t do it all. In addition, you will need an intermediate scratch-pad type tool (I use Notion) where you dump the AI-related research material.

4. Learning curve: Each AI-powered chatbot responds differently to the same prompt. It kills me internally when I see “AI influencers” comparing two models with a common creative prompt.

In the rest of the article, I’ve dived into how you can minimize the impact of these limitations and make sure you get the best bang for your buck while writing content with AI.

Acknowledge strengths and weaknesses

Based on my last two years of wrestling with various AI models, here’s my feedback:

• ChatGPT is the best overall tool. If you could get one, this is it. It allows easier customizability and is the best general-purpose LLM on the market. The GPT ecosystem just makes it even better. That’s why my content writing workflow is built on it.

• Perplexity.ai is absolutely necessary for conducting research, verifying facts, and finding authoritative sources. The best part is, you won’t have to pay anything. The free version is pretty awesome.

• Claude 3.5 Sonnet offers the best model period. It is great at understanding your commands and producing a desirable output that adheres to your instructions. However, its limited world knowledge limits its utility.

• Gemini 1.5 has the highest context window. If you have a bunch of heavy PDFs, audio and video recordings, or long YouTube videos to go through, this can help you immensely. It’s free and Google gives you more control over its creativity and political correctness.

Yes, there are some pretty phenomenal open-source models as well. I’ve tried the Mistral 8x7b and the Llama 3.1 8B model on my computer. The uncensored Dolphin models are amazing too.

These understand your requests well and are pretty solid at carrying out editorial tasks like rephrasing sentences and generating meta details.

However, they have massive downsides. They are limited to their small knowledge bases and lack internet access. Plus, they have the smallest context window, regardless of where you use them, and aren’t multimodal. 

(Although Meta just released the Llama 3.2 models which are open-source and multimodal.)

Do your freaking job

You have heard this a million times by now — AI is not here to replace content writers.

The only people who believe AI can create content are noobs who have just discovered the magic of LLMs and bad writers.

Anyway, chatbots are here to fetch the specific information you need for the context of your article. This is massively useful for writers as jumping through multiple tabs mid-sentence is a productivity killer.

LLMs can collate all the information you need in one place. It is presented in the context of your article and in a desired level of comprehension. 

Authors should remember that they are the ones communicating with the reader, not the machine. 

While relaying a message, you need to back it up with stats, facts, and logical reasoning. On top of that, you need to package it in an engaging format so the reader actually attentively consumes your content.

AI can help a lot in finding the right information, as mentioned earlier, but when it comes to persuasive communication, it just sucks.

I am perfectly aware that it is enticing to push the GPT-generated output and call it a day, but in the long term, it will ruin everything. In the previous chapter (linked below) I explained the rare conditions when it may work. And in the next chapter (also linked below) I talk about the potential consequences so you know what’s at stake here.

Get a larger monitor and use Notion

One of the primary reasons I dislike laptops for deep work is their screen size and the position of the keyboard. If the keyboard is too close to the screen, you will have to hunch over, which is just terrible for your neck and back.

If you upgrade to a mere 22-inch display, which isn’t that big for most workspaces, you can easily boost your productivity. Usually, I split my screen 70-30. The larger part has the word editor, where I write content, and the smaller portion has the research material that I scroll through as needed.

Of course, you will need an extra keyboard and mouse, but I’ve seen a lot of professionals use an external monitor with their laptop’s keyboard and trackpad.

Now, let’s talk a bit about Notion’s role in content creation. 

It is the best tool where you can just create a page and pool in all the information you need to write your articles. Web links, ChatGPT outputs, content outlines, and more. This can stay in the 30% zone in the split I mentioned earlier. Check it out below:

You can rest the cursor on the Notion side and scroll up and down to get the information you need.

Learn and tough it out

Honestly, you will make plenty of mistakes while incorporating AI chatbots into your content production workflow, even with this book. It’s not because you aren’t hard-working or my workflow is useless. It’s just that there are too many use case-specific variables.

For instance, the definition of a “good article” depends on your audience’s preferences and brand guidelines. Similarly, if your articles are time-sensitive (related to sports, for example) you will find most of the LLMs useless.

There will be also days when you will work hard to write prompts and build workflows that will lead to utter trash and you will have to go back to square one.

And frankly, this is the best part of learning something new, particularly when it involves Gen AI and content creation.

Wrapping up: The advantages are worth it

Adopting LLMs into your content production workflow comes with challenges. There is no definitive playbook, focused tasks can only be done manually, you will have to adopt multiple tools, and there is a learning curve.

At the same time, it will reduce cognitive fatigue, save you time, and make you more money. You will spend less time finding the right information in a digestible format and more time conceptualizing your message.

It is essential to acknowledge what tools like ChatGPT can do and fight complacency with all your might. I highly recommend writers invest in a larger monitor to save their eyes and boost their productivity. Finally, there is no substitute for hands-on learning.

—

Next Chapter: Should You Publish AI-Generated Content

Previous Chapter: How AI-Generated Content Performs Organically

Index (with Prologue): Content Writing With AI: Principles, Strategies, and Prompts