Generative AI took the world by storm in 2022, sparking debates about possible concerns related to its usage. This includes unethical generation of texts by students and content creators. Together with it came the problem of AI detection, and consequently, enthusiasts started working out ways to bypass the detectors by making generated texts appear human-like.
If you want to know how to make AI-generated text undetectable, you’re in the right place. Read on to understand the workings of ChatGPT content detection and learn workable ways of humanizing texts.
🤖 What Is AI-Generated Text?
As the term suggests, AI-generated text is any piece of content created by an advanced algorithm, not a human being. Generative software of previous generations could only produce a few sentences of garbled text. Meanwhile, current AI models are capable of making impressive, human-like content.
The most famous AI tools for content generation are ChatGPT, Gemini/Bard, Perpexity.ai, and Jasper.ai. The latter two options are paid, but the rest are free to use, with ChatGPT widely considered the best free software.
🔎 2 Things You Should Know About How AI Detectors Work
AI generation goes hand in hand with AI detection. Since the dawn of ChatGPT, shrewd content writers and students have started presenting texts by algorithms as their own intellectual products. Thus, the community of content buyers, professors, and publishers has become interested in AI detection software that may help spot unethical conduct.
The first efforts in this domain were GLTR, GPTZero, Originality AI, Crossplag, and others. Currently, there are many more free and paid tools, and you can choose among them based on your goals.
Regardless of the checkers you select, here’s what you need to know about their workings and the parameters they use in assessments.
1. AI Detectors and Generators Are Based on Similar Language Models
Generative AI and detection software are based on a similar language model – in most cases, it’s a GPT model that uses natural language processing technology for text creation.
Detectors use data from AI texts to check whether the sentences in the analyzed content are constructed in compliance with the GPT logic. Based on this feature, they determine the text’s origin.
2. Perplexity & Burstiness
Human writing is distinguished from the robotic and cliché-ridden generated text by two major parameters: perplexity and burstiness.
- Perplexity is the degree of unpredictability in word choice and syntax. It may also be attributable to a person’s individual writing style and preference for particular phrases.
- Burstiness is the degree of variability in the syntax that only humans can employ to diversify their content and make it more engaging for the readers.
🤩 4 Ways to Make AI Text Undetectable
So, which methods can you use as soon as you have an AI-generated text and need it to bypass the detectors? Here are 4 proven techniques that helped make hundreds of AI texts more human-like:
- Tweaking the text.
This approach is the most reliable. It involves rewriting the text at least partially by changing the words and sentence structures but retaining the content. This method will also help you add a personal touch and style to your content. - Applying the same AI software you used for text generation.
You may write a prompt asking the algorithm to make generated text more human-like. - Using automated rephrasing tools.
This is a new category of assistive software that can rephrase sentences and change the syntax to disguise the AI origin of your content. - Utilizing text humanizers.
These advanced and modified paraphrasing tools target AI-detected chunks of text and make tailored changes to help your text look human.
We’ll discuss the effectiveness of automated tools and humanizers in the following sections.
Disclaimer: Submitting generated texts as your own violates academic integrity. The same goes for humanized AI content. The techniques described in this article should only be for ethical purposes, such as improving generated essay examples.
🚀 6 Ways to Make AI Undetectable Automatically: Our Experiment
Here is a detailed breakdown of our experiment using AI-generated text produced by ChatGPT. To begin with, we asked the algorithm to generate a 150-word text on developmental psychology. It yielded the following output:
Next, we checked this text with GPTZero and IvyPanda’s AI detector for essays. The former tool analyzes word use and sentence structure, while the latter is sensitive to text’s overall quality.
The results were as follows:
- GPTZero – 98% likelihood that the text was AI-generated.
- IvyPanda GPT checker – high risk of AI-generated text.
After that, we tried to modify it using 2 chatbots, 2 paraphrasing tools, and 2 humanizers and ran the check again.
According to the results, this is how we would rank the tools we’ve tested:
- Phrasely.ai ⭐⭐⭐⭐⭐
Provides the best-quality results but doesn’t have a free version. - Undetectable.ai ⭐⭐⭐⭐
Makes AI texts more human-like while maintaining readability but doesn’t have a free version. - IvyPanda Paraphraser ⭐⭐⭐
Helps to avoid AI detection but requires additional manual editing. - Quillbot ⭐⭐
Creates high-quality paraphrases but doesn’t make AI undetectable. - Gemini ⭐
Offers helpful comments but fails at making AI texts undetectable. - ChatGPT ⭐
Completely unsuitable for humanizing AI texts.
Based on our experiment, the IvyPanda paraphrasing software is the best free tool for avoiding detection. Still, it requires additional input on your part. Phrasely.ai is the most efficient, though paid, humanizer available to date.
Read on to learn more about the results of our investigation.
ChatGPT
Since ChatGPT is the most popular and accessible AI tool, we’ve decided to check how well it can humanize its own texts.
Option 1: Simple Prompt
We pasted our generated text on developmental psychology into the dialog box and added a prompt:
We checked the resulting text, and this is what we’ve got:
- GPTZero – 98% likelihood of AI-generated text.
- IvyPanda GPT checker – medium risk of AI-generated text.
This first attempt was pretty disappointing.
Option 2: Using Perplexity & Burstiness
Next, we chose to do it differently. We started by explaining to ChatGPT what parameters AI detectors use to scan the texts so that it could introduce more nuanced changes to its content.
We fed the info about perplexity and burstiness into the bot and asked it to perform changes in the AI-generated text with these parameters in mind:
The resulting output was 234 words long, which is x1.5 larger than the original text. A detector check gave us the following:
- GPTZero – 98% likelihood of AI-generated text.
- IvyPanda GPT Checker – medium risk of AI-generated text.
Our Takeaways
Benefits
- None.
Drawbacks
- There’s no sense in using ChatGPT for text humanization: it doesn’t cope with the task at all.
As you can see, even though ChatGPT is a high-quality generative AI, reworking and humanizing its texts is better done with the help of other tools.
💡 Pro tip
Did you know that the chatbot can effectively refine and improve texts? Read our guide on using ChatGPT to edit essays and learn everything about it!
Gemini (ex-Google Bard)
Gemini works similarly to ChatGPT but uses a different language model developed by Google. Given that most AI detectors are geared toward GPT model detection, we assumed they may fail to spot Gemini texts as AI-generated.
Results
We used the same prompts for text humanization as we did with ChatGPT, and the results for both were as follows:
- GPTZero – 98% likelihood of AI-generated text.
- IvyPanda GPT checker – medium risk of AI-generated text.
Our Takeaways
Benefits
- Unlike ChatGPT, Gemini gives a report on what tweaks it introduced into the text, such as incorporating metaphors and imagery, diversifying the sentence structure, personalizing the text, or transforming the passive voice to active.
Drawbacks
- It failed to humanize the text since both checkers marked it as suspicious.
So, we can conclude that chatbots are unable to make ChatGPT essays more human-like.
Quillbot
Now, let’s see how to make AI-generated texts undetectable with the help of paraphrasers.
Quillbot is one of the best-known paraphrasers that enjoys a good reputation among users. Its free version has only a few options. For our experiment, we used mode fluency and applied the maximum percentage of paraphrasing.
Results
The rewritten text appeared coherent, readable, and overall pretty impressive. Yet, the free version limits the amount of text to 125 words only, and not all input ends up paraphrased.
The AI scan of the resulting content was disappointing:
- GPTZero – 94% likelihood of AI-generated text.
- IvyPanda GPT checker – low risk of AI-generated text.
Our Takeaways
Benefits
- High-quality paraphrase with readable results.
Drawbacks
- Advanced checkers like GPTZero can still mark the paraphrased text as generated.
- The tool’s free version is very limited regarding word count and functionality.
Overall, Quillbot is a good paraphraser but unsuitable for AI texts’ humanization, especially if you’re a non-premium user.
IvyPanda Paraphraser
IvyPanda’s paraphrasing tool has a convenient option for text customization. You can edit the output directly on the webpage by replacing each highlighted word or phrase with a more suitable synonym or leaving it as is. We used two options here:
Option 1: Fully Automated Paraphrasing
We used the 40% paraphrasing option, and here’s the outcome of the AI detection check:
- GPTZero – 30% likelihood of AI-generated text.
- Ivypanda GPT checker – low risk of AI-generated text.
However, looking at the text attentively, you will see that some replaced words and expressions do not fit well into the context. For instance, the tool changed the term “developmental psychology” into “enlightening therapy,” which makes little sense.
To make the results readable, you can use low levels of paraphrasing depth (20% or 40%) and manually check all the changed words (there is an option of choosing the most suitable synonym). This way, the text will look good not only for an AI detector but also for a picky reader.
Option 2: Additional Manual Editing
Next, we’ve tweaked the inconsistencies in the results manually. We aimed to retain the effect of paraphrasing but make the text more readable.
The outcomes of the AI check were impressive:
- GPTZero – 50% likelihood of AI-generated text.
- IvyPanda GPT checker – low risk of AI-generated text.
Our Takeaways
Benefits
- Free usage with a sizeable character limit (5,000 characters equals around 700-1000 words, which is an average-size essay.)
- More effective for humanizing than chatbots and other paraphrasers.
Drawbacks
- You should check the output carefully and rework if necessary to improve readability.
💡 Pro tip
Want to learn how to tweak and improve texts automatically? Check out our article on AI tools for writing and editing.
Undetectable.ai
The next type of tools we checked was humanizers. These are advanced AI-based paraphrasing tools that specialize in making AI content appear human-written.
The most popular product in this category is Undetectable.ai. It’s a paid humanizer that gives new users 50 credits for free to test its capabilities.
Results
Once a text is paraphrased, it is automatically rated by 8 widely used AI detection tools, including Writer, Copyleaks, and CrossPlag. While it’s a handy option, it has a significant problem: the ratings are incorrect. For instance, it may show that Writer and CrossPlag detectors mark the checked text as human-written, but if you run a separate check via each of these tools, they will mark the text as AI-generated.
The “humanize” option of Undetectable.ai works well, and the resulting text was impressive, aside from several punctuation errors. The AI scan showed the following:
- GPTZero – the text is likely written by a human.
- IvyPanda GPT checker – low risk of AI-generated text.
The second check after the removal of punctuation errors showed worse results:
- GPTZero – 50% likelihood of AI-generated text.
- IvyPanda GPT checker – medium risk of AI-generated text.
Our Takeaways
Benefits
- High-quality paraphrasing.
- The text actually became more human-like.
Drawbacks
- You can humanize only 250 words for free; more capacity is available for a $9.99 subscription (10,000 words per month.)
- The results are riddled with punctuation errors. As soon as we corrected them, the text became more AI-like.
- The AI detection feature is misleading, with false results for different AI detectors.
Phrasely.ai
Phrasely.ai works similarly to Undetectable.ai, having an AI detection and humanization feature.
Results
The initial AI check of our text with the Phrasely.ai detector correctly indicated that it was a generated piece. We then ran it through the humanizer. While the tool provided a readable and grammatically correct rephrased text, the output was considerably longer than the original version.
Here are the AI scan’s results after humanization:
- GPTZero – the text is written by a human.
- IvyPanda GPT checker – low risk of AI-generated text.
Our Takeaways
Benefits
- High quality of paraphrasing.
- No errors in the humanized text.
- The results were undetected by two AI checkers.
Drawbacks
- There’s no free version.
- The text became longer.
We can confidently say that Phrasely.ai has provided the best quality results. Unfortunately, using this rewriter costs a $14.99 subscription for an unlimited word count. If you’re using humanizers to improve essay samples, paying for a subscription is simply not worth it.
💡 Pro tip
Want to access the functions of a high-quality rewriter tool for free? Try IvyPanda’s online AI humanizer. This tool is an excellent alternative to popular paid humanizers: it will effectively reshape your text without compromising readability. The results will be checked automatically, and you will see how much more natural your text has become.
✍️ 4 Tips to Make AI Undetectable Manually
In addition to the software at your service, you can use various techniques to humanize your content yourself. This approach will also come in handy if your own text is being falsely flagged as AI-written and you want to improve it.
Here’s how to make AI undetectable even if your text is 100% GPT-written.
1. Rework the Syntax
GPT models mostly use standard, unnaturally correct sentence structures. That’s why adding non-trivial syntax is an excellent way to hack the system and surprise AI detectors with original sentence structure choices.
2. Make Vocabulary Adjustments
Synonyms, literary means, and diverse phrasing are a great remedy against AI detection. It often comes that AI detectors give false positives for texts written by humans due to the use of plain wording. That’s why a varied and sophisticated vocabulary is your key to bypassing the AI check.
3. Try Shortening the Sentences
As a rule, AI writers don’t use short sentences. They tend to string together as many words as possible using clichéd syntactic forms. Thus, you may break down long sentences into a couple of smaller ones, and the AI detector will change its mind about your text’s origin.
4. Add Stylistic Tweaks
A bit of personal touch will always do good for your text’s quality and origin assessment. Read through the AI-generated text and think of what you would have said differently. Add these tweaks, and the text will appear more like yours.
While the manual approach is more time-consuming than simply using an automatic humanizer, it yields the best results. This way, you will increase your text’s burstiness and perplexity, which will significantly improve its score. You will also have complete control over the output without the danger of turning your text into gibberish.
💡 Pro tip
Wondering how to make AI essays undetectable more efficiently using a manual approach? Try our smart tools:
- Plagiarism checker will ensure your essay is 100% original.
- Essay reader can read your text out loud and help you ensure it sounds natural.
- Chat GPT essay maker will generate additional samples on your topic.
- Essay extender will let you easily reach the desired word count.
And if you want to improve your paper even further, the following tools are just what you need:
- Hook generator will suggest excellent examples of attention-getters.
- Research introduction maker will make a background section for your paper.
- Free question generator from text will assist in developing research questions.
- Quote explainer and rhetorical device finder will assist in writing a rhetorical analysis.
- Key points maker will momentarily extract keywords from any text.
👋 Final Word
Now, you know everything about how to make AI texts undetectable. If you’ve found this material helpful and have someone in mind who may also benefit from these tips, feel free to share our article. Good luck in your studies and interactions with your chosen AI assistants!
We also recommend checking out our comprehensive guides on using ChatGPT for research paper writing and making it write longer essays.
🔗 References
- How to Detect AI-Generated Content: TechTarget
- How to Spot AI-Generated Text: MIT Technology Review
- AI Detection: How to Pinpoint AI Generated Text and Imagery [+ Detection Tools]: HubSpot
- 10 Best Undetectable AI Tools to Bypass AI Detectors: Geeks for Geeks
- Tips on Detecting & Improving AI-Generated Text: University of Wisconsin–Madison
- AI Writing Detection: Red Flags: Montclair State University
