How To Make AI Content Undetectable (Results From 9 Tests)

You want to create unique content for your website or blog, but you don’t have the time or budget to hire a professional writer.

So you turn to AI writing software tools that can help you write faster and cheaper.

But now you’re worried about being flagged by AI content detectors.

Or worse, what if Google can detect AI content?

That can ruin your search rankings and reputation. Then you’re pretty much screwed.

So what can you do to make it undetectable and high-quality?

In this article, I conducted 9 tests to see how to make AI Content undetectable and bypass detection.

Let’s get to it.

What Does AI-Detection Mean?

AI-Detection is a term that refers to the process of identifying whether a piece of text was written by a human or an artificial intelligence (AI) system. It can help prevent misinformation, academic dishonesty, or chatbot deception.

It’s performed using classifiers trained on large datasets of human-written and AI-written texts on different topics.

These classifiers use machine learning algorithms and natural language processing techniques to analyze the text and assign a confidence score that indicates how likely it is that an AI wrote the text.

Want to know more?

Check out my article on how AI detection works.

Generating a Baseline

Before we begin, it’s only logical to set up a baseline for comparison.

I generated a few paragraphs on 3 topics using ChatGPT 3.5 and then ran it through various AI detectors like Originality AI, Content at Scale AI Detector, and Open AI Text Classifier.

This establishes a benchmark for the performance of AI content detection tools and gives me a starting point to evaluate the effectiveness of the AI bypassing strategies in subsequent steps.

Creating Raw AI-Generated Content on 3 Topics

I got ChatGPT to generate AI content on the following topics:

• How the Future of AI Will Revolutionize Content Marketing
• Why Anime Is So Popular
• Existence of Aliens

To save you time, here are the outputs and AI test results for topic 1:

Here is a table of the baseline test summaries for all topics:

Topic	Originality AI	Content at Scale*	Open AI Text Classifier
how the future of AI will revolutionize content marketing	100%	60%	Possibly
why anime is so popular	97%	46%	Possibly
existence of aliens	97%	87%	Possibly

What Are The Methods To Make AI Content Undetectable

I got the inspiration for this experiment from Matt Diggity, founder of Diggity Marketing. You can watch his video here.

My tests included a few of his methods in addition to mine, but with different topics and tools.

Matt also used the OpenAI playground, but I chose ChatGPT because not everyone would have a subscription.

Here are the results.

Test 1: Using Quillbot Paraphraser

The Quillbot Paraphraser is quite a popular tool that many people use to get good results apparently.

How true? Let’s find out.

I ran it through the standard mode of the AI tool.

Here are the results:

Here is a summary of all the topics:

Verdict

For the most part, Quillbot’s paraphraser didn’t do much except in 2 cases:

• By-passing Originality AI in topic 2. (97% to 2%)
• Slight improvement in Content at Scale for topic 3. (87% to 42%)

It didn’t produce statistically significant results to bypass AI content detection in 7/9 cases.

So, I’d pass on this.

Test 2: Using TheLatestN0W ChatGPT Pre-Prompt

Next up was to try a secret GPT pre-prompt by TheLatestN0W.

The idea is to give ChatGPT some context before giving it a prompt which may yield better results.

This was done by giving ChatGPT some information about the following 2 concepts:

• Perplexity: The complexity of the text
• Burstiness: The variation of sentences

Human-written content tends to have the highest of both these metrics.

Here is the prompt to enter before the initial command.

Here are the results:

Here is a summary of all the topics:

Verdict

So in my case, the prompt didn’t really do much to outsmart the AI content detectors.

It did show an improvement from 60% to 33% for topic 1 with Content at Scale’s AI detector tool.

But that’s just 1/9 of cases.

I’d pass on this method too.

Test 3: Using Bing Chat

Bing Chat is one tool I don’t see many people using or discussing, maybe because of its accessibility.

You do need to download Microsoft Edge to use it.

This method was pretty straightforward. I simply gave the prompt: Write me 2 paragraphs [Topic]. Then ran it through the 3 AI content detection tools.

Here are the results:

Here is a summary of all the topics:

Verdict

Unfortunately, the output of topics 1 and 2 wasn’t enough to run through Open AI’s Text Classifier.

Nevertheless, this was actually quite decent at making AI-generated content human.

The results showed improvement in the following cases:

• Managed significant improvement with Originality AI and Content at Scale in topic 1.
• And a huge difference in topic 3 for both Content at Scale and Open AI’s Text Classifier.

That’s 4/7 cases.

Not bad if you asked me.

Test 4: Asking ChatGPT to Sound Human

In this test, all I did was ask ChatGPT to sound more human-like with extra instructions like:

• Use a conversational tone
• Add a few unorthodox sentence structures
• Make some jokes

Here are the results:

Here is a summary of all the topics:

Verdict

This method was so-so.

Topic 1 Content at Scale and Topic 3 Originality AI outputs showed about a 20% improvement from the baseline.

The rest either got worse or showed marginal improvements.

That leaves us with 2/8 successful cases. That’s a pass.

Test 5: Using WordHero Blog Paragraph Template

In my WordHero review, I mentioned that it could beat most AI content detection tools.

But let’s see how it does for different other topics.

I used their Blog Paragraphs template to describe the topics briefly. It doesn’t accept prompts, so all I could do was enter the keyword.

Here are the results:

Here is a summary of all the topics:

Verdict

I lost 3 data points due to insufficient characters for all topics to use the OpenAI Text Classifier.

But for the most part, the result was split mid-way.

3/6 showed improvement, 2/6 got worse, and 1/6 were neutral.

It’s decent if you already have WordHero or plan to get an AI writer with unlimited words.

Test 6: Remove Commas and Use Short Sentences

Next up was trying to get ChatGPT to remove all commas and use shorter sentences.

I used the same prompt with the additional instructions.

However, AI-generated content still included some commas.

Here are the results:

Here is a summary of all the topics:

Verdict

The output was too short again for the OpenAI’s Text Classifier, so I lost 3 data points.

It showed promising results for 2/6 cases and marginal for one.

Let’s just call it even again.

Test 7: Using ChatGPT To Rewrite the Raw AI-Generated Text

All I did here was ask ChatGPT to fix the already flagged content by simply rewriting it.

Here are the results:

Here is a summary of all the topics:

Verdict

This was actually one of the better-performing methods.

7/9 cases showed improvement from the baseline results.

Granted, not all of it was significant, but it beat Originality AI flat out for topic 1. Going from 100% to 1% AI-generated content.

Quite impressive for a simple tweak.

Test 8: Using Content at Scale

Test 8 uses the paid content generation feature by Content at Scale.

Here are some disclaimers first.

The content was created in existing article projects I already had, which were already optimized for a specific target keyword, unrelated to topics 1, 2 and 3.

I had to do this because you can’t create short-form, only long-form content in this tool. But you can create section headers by giving the AI prompts and further context.

Here are the results:

Here is a summary of all the topics:

Verdict

Content at Scale showed clear improvement in virtually all the cases except its own AI content detector tool.

Ironic.

It has performed one of the best so far.

Test 9: Using Undetectable.ai

The final method on this list is using a relatively new tool that I found called undetectable.ai.

It functions more or less like Quillbot, but it actually rephrases to pass AI detection.

All you got to do is drop the text you wanna rephrase in the text field and click “Humanize”.

Here are the results:

Here is a summary of all the topics:

Verdict

Ding ding ding. We have a winner!

Undetectable AI has produced near-perfect results for all topics.

Unfortunately, I didn’t have enough text for topics 2 and 3 to run through the Open AI text classifier.

But it’s safe to assume it would produce unlikely or very unlikely.

What Is The Best Way To Make AI Content Undetectable?

The best way to make AI content undetectable from all my tests is to use Undetectable.ai.

You get some credits on the free plan, but they have very affordable pricing should you need more.

And they have made a very bold claim of refunding you the cost of humanization if their tool produces anything that gets flagged AI.

What Is The Best Free Way To Make AI Content Undetectable?

The best free way to make AI content undetectable is simply asking ChatGPT to rewrite the previous output. It performed the best out of all the free methods listed above.

Why Bother Making AI Content Undetectable?

Making AI content undetectable should be worth considering for the following reasons:

How To Make AI Content Undetectable (FAQs)

To Sum Up

From the above tests, AI-written texts can no doubt be edited or modified to evade detection.

But I’ll always advocate not relying solely on gimmicks or prompts to make your content undetectable as AI-generated.

The absolute best way is always to add the human touch and use these tools as helpful assistants to supercharge your content marketing strategy.

Hope you enjoyed this article.

Feel free to quote this article in any of your content.

Share other tips that I can test in the comments below.