Not too long back, AI content was something that people casually talked about. A few AI writing tools would charge a bomb and marketers hated how insipid it sounded. Professors cared about it only if the content appeared red in the plagiarism tools because good AI content was unheard of.
But the status quo changed with the advent of ChatGPT that hit everyone like a Tsunami and no one knew how to react. Big organizations like Google were left fumbling for answers while freelance writers started losing clients. Search engines almost stopped caring about AI content because there was just so much of it! Six months down the line, AI content has become pervasive. And one can only learn how to use it effectively because there is no going back from it.
Today, what matters is how you use AI content and still convert leads from it because no matter how advanced AI gets, it will never bring in the human touch. But customers want to buy from humans, not from bots, so it’s essential that your content reads human and empathetic — even if the majority of it came from an AI tool.
Before we get to the meat of this article, remember that cleaning content happens in steps, it begins with your first prompts and takes many iterations. It also involves a lot of human input which means it will take time — not as much as writing a piece from scratch would — it’s a one-prompt-done game.
Here are the 12 steps to truly create undetectable AI content:
1. Understand how AI content writing tools work
The first step to conquering something is analyzing how it works. By learning how AI writes content, you can reverse engineer and remove elements that make content sound machinic.
ChatGPT, for example, consumed a lot of content and learned how people write. It is designed only to determine what the next word should be.
For example, if you ask it to write a poem on an elephant, it will draw its inspiration from millions of words about the elephant that were published online. It gauges your prompt and decides that the expected content should be poetic, so it will take content from scientific journals, children’s books, religious books, history books, etc. but uses poetic words it picked from literature books and novels.
But this explanation is opinionated and it’s one of the many theories going around about how this AI machine works. Other guesses come from the common AI detector tools like “contentatscale”. This tool, as shown in the below image, analyzes the pattern of content fed into it and the predictability of text.
No one knows the algorithm behind it but the general idea is to measure the perplexity of text and determine if AI can reach or match it.
2. Remember that no AI detection tool is perfect.
The best these tools do is some sort of calculated guessing. They claim to be 98% accurate but based on my experience, none of these tools are perfect or reliable. At times, content I write shows AI while AI content is detected as human-written. However, they are about 80% accurate, which is good enough for most purposes.
That brings the question, how accurate is Google’s AI detection tool. If it claims that it will penalize AI content, it must have a mechanism to detect it. It must be based on some pattern analysis for sure, but what if human writers match that style by mistake? They will be penalized and there would be no way to prove that the content is not AI. So there always will be an element of doubt and a margin of error.
There, however, is an indirect way for Google to detect AI content — by reading the signals sent by visitors. Now, even human feedback can be inaccurate but the ranking signals like low bounce rate, high average time on page, and high page count per session might indicate quality natural content.
3. Train yourself to detect AI content without tools
In God we trust, for everything else, bring data — It is a famous quote that marketers should abide by, but what if we have no data? What if there is no metric to measure something? What if the little data that is available is not reliable? We have already established that tools can’t detect AI content, so there is no concrete frame of reference — there’s no wrong or right. So what do we do in that case?
As marketers, we face such situations often, and it’s when intuition is the only tool that works. You need to have someone with solid editorial experience, someone who has edited a lot of content to determine if a phrase of text “reads human”. Even their judgment can’t be perfect but it’s the best you can do.
If you can’t afford an experienced editor, there is an alternate method.
You can give the content a read and ask yourself if you feel involved and are able to relate to what the writer is trying to convey. If you can, then the content is human, even if it’s written by an AI tool.
4. Be prepared to write at least 40% of content yourself
You don’t want to hear this, I know, but it is what it is. AI content, with its current capabilities, can’t write content that people enjoy reading. It makes for a brilliant brainstorming tool, as we will discuss later. It will generate some cool ideas and bring in a perspective. But it is limited when it comes to fleshing out an outline.
This means, you would want to write most content yourself. AI can provide a rough first draft after a bunch of iterations but you need to package content better, add more context, edit text that doesn’t “feel” right. And when you do all this, you will end up writing about 40-45% of content yourself.
And if you think you can save a lot of time — you would be partially right because you can save up to 25-30% of your writing time with AI. That’s because you would be investing in strategic writing and the part when you have to deliberate the next word is taken care of by AI. In short, you tell AI what to write and how to write but you step in to make it strategically accurate.
5. Start with a human-friendly outline
AI tools require customized human input and prompts to generate a high-quality, human-lik output. The process begins with creating a detailed human-sounding blog outline that serves as the foundation for the content. The more human and useful the outline is, the more human-like the resulting content will be.
You can start by writing a prompt that includes your intent for writing the piece, what you plan to achieve from it, and who your target persona is. Clearly state the purpose and objective of the content, whether it’s to inform, persuade, entertain, or educate your target audience.
Weave the details of your target audience into your prompt to help the AI understand the context and style of writing required. Specify their demographics, interests, and any other relevant characteristics to shape up the content accordingly.
After you submit the prompt, AI will create the initial outline draft in a minute or so. Review the outline, carefully examine each item. Remove irrelevant subtopics that do not contribute to the primary keyword that you target.
Need help from AI in generating topic ideas and keyword research?
Access the guide to using ChatGPT for keyword Research
Next, identify any additional items that are relevant to the topic and plug them into the outline. For example, AI does not know about your product capabilities, so it will have no context on how to match the content with customer pain-points — so you will have to add an H2 with features of your product manually.
Once you have made edits to the initial outline draft, ask the AI to suggest further improvements or refinements to the outline. New items in the outline will give more ideas to your AI writer, a few that you may not have thought about.
Iterate this process, incorporating the AI’s suggestions and making necessary adjustments until you are satisfied with the final outline.
Remember that you possess the subject matter expertise and knowledge about your product or topic, which the AI may lack. Therefore, actively participate in the refinement process to ensure that the generated content aligns with your expectations and effectively conveys your message to your target audience.
6. Give AI a reference of the content you wrote
AI writing tools need a sample of your content to understand your style, tone, and approach — Simply instructing it to “make it sound human” or “make it sound professional and use Oxford commas” won’t yield the desired outcome. You must provide a reference piece or a couple of paragraphs for AI to emulate.
The reference should ideally exhibit the specific tone, style, and grammar you wish to replicate in the AI-generated content. The more representative and detailed the reference, the better equipped the AI will be to generate human sounding content.
So a prompt like…
I want you to write content about [Topic]. Emulate the style, grammar, and tone used in the below text. [Two paragraphs of the content you wrote]
…would generate a far more human sounding copy than a generic prompt that leaves a lot to AI’s gumption.
7. Cover one sub-topic at a time
One of the biggest limitations of AI writing tools is their ability to retain context after generating a significant amount of text. To keep AI content contextually and topically relevant, try to write prompts that require up to 500 words in response. You can do that by focussing on one sub-topic at a time instead of prompting for the whole article in one go.
8. Use the right AI prompts to generate semi-unique content
The best way to generate natural sounding content from AI is by giving it a detailed prompt with sample text for reference, just the way we did for the outline. As we will see in an example later, a detailed prompt gives content that sounds less robotic and scores better on AI detection tools.
9. Copy the content as it is to your text editor
After you receive the first H2 from the AI, give it a cursory glance. If it meets your standard, copy-paste the content into your editor tool. The idea is to get all content together on a page before editing.
If the content is not as per your expectations, you may have to run a few more iterations by supplying more information to the AI writing tool. It’s best that you start a new chat with the updated prompt, otherwise, the AI tool might build up on the content it produced based on the previous prompt.
10. Read the content aloud to identify robotic sounding voice
AI often provides grammatically consistent content but you still need to clear out the obviously robotic content. By reading aloud, you will identify instances of excess usage of passive voice and over emphasis on the process of doing something, which AI often does. AI tools often turn everything into a stepwise process and bullet points.
Another way to make content less robotic is by using the text-to-speech tools. They provide an auditory feedback that highlights areas where the content sounds robotic or lacks a human touch.
Everytime you feel that the content sounds a bit off, pause the text-to-speech tool and edit the content till it sounds natural. Do this for the whole article before moving to the next step.
11. Mix your personality while rewriting
Now your content reads human but it may not be sound engaging yet. To make it sound human you need to add a bit of personality in it. Use metaphors to make content more relatable — that’s one thing that AI can’t do with eclat.
Also sprinkle in some instances of real pain-points of your audience. You can get those insights from your sales and customer support teams. Also consider adding examples to prove you point and sell the value proposition better.
12. Run your content by an AI detection tool
We spoke about how AI detection tools are not accurate, so this step is optional. At times, however, you may need to prove to your boss the content is indeed AI free. So it’s best that you run the content by at least 3 AI detection tools, take screenshots once they approve of your content, and send it to your boss as proof.
AI tools like Content Scale detect robotic sounding phrases in your content and highlight it. You can take a call on editing that text to satisfy the tool but it’s more of an intuitive process than anything else.
An example of AI content becoming undetectable
The best way to show how AI content can be undetectable is by sharing examples of content that are AI generated and human generated. And that’s what we are going to do now. And to keep it real, we will take content from this very article.
We will refer to point# 5 in the list – “Start with a human friendly outline”With help of screenshots we will show what raw AI content reads like, how the right prompts make it sound more human without editing, and how human intervention turns the same few paragraphs into more engaging copies.Let’s get started.
The Output without human guidance
Notice how the content is immediately turned into a numbered list, even when that’s not truly required. ChatGPT also uses words that are not commonly used during a conversation. The text is a bit too formal and bland. You may also find AI tools being too cautious while making any claims — this hurts the authenticity and authority of your content.Let’s see how it performs in an AI detection tool called ZeroGPT.
Performance of all AI content in AI detection tools
Not too bad now, is it? It’s already 28% AI as per the tool. Let’s see how it changes when you provide a more detailed prompt.
Output with human guidance
The above text was generated by ChatGPT when I provided with some bullet points to cover.
Performance of human-guided AI content in AI detection tools – Before human edits
This one step alone, reduced the AI content by 6 percentage points! I made no efforts in editing the text yet but the tool already thinks that the content is human written.Finally, let’s see what happens after I edit the content.
Performance of human-guided AI content in AI detection tools – After human edits
The AI content has come down to 12%. Now I put no effort in cleaning up the content that is flagged as AI generated in yellow highlight. That’s because I wrote that content! It’s a short-coming of the tool to call it AI-generated.I am happy with the output and have used the same copy in point #5.
In this blog post, we covered in detail how you can make your AI content totally undetectable. Remember, all the steps here ask you to use intuition first and then use tools to support your claims. Don’t be too concerned if the content is detected by tools as AI-generated, if the content “sounds” right to you. Finally, there is no benchmark to detect AI content, so don’t push it too hard but just make it readable and worthy of engagement.
