Want to make ChatGPT undetectable by AI detectors like TurnItIn and GPTZero for free?
You’re in the right place.
I used to be on the other side of the AI detection fight.
I owned a writing service, and when ChatGPT first came out, I was responsible for ensuring hundreds of writers weren’t using it to generate content.
We tested many detectors and experimented with alternate ways to detect AI usage.
In this article, I’m sharing 15 of the latest tactics on avoiding AI detection, including methods methods recommended by AI scientists, tools that will make ChatGPT writing undetectable, and much more.
Let’s get started.
What AI Detectors Does This Guide Work For?
Here’s a list of AI detectors built for the GPT model that can be tricked with the methods in this guide.
- TurnItIn
- GPTZero
- Originality AI
- Winston AI
- Hive
- Crossplag
- Writer.com
- CopyLeaks
- Sapling
- Scribbr
- ContentAtScale
And any other AI detector designed to detect GPT output.
How To Make ChatGPT Undetectable By AI Detectors
Here are 15 proven methods for making ChatGPT undetectable.
1. Make Your Content More Human
The default writing style of ChatGPT is very predictable. This predictability is what the AI detectors look for.
Therefore, making your writing more human-like and unique will decrease the chances of getting caught.
There are many ways you can do this. You can learn my top 15 methods in this guide: How To Make ChatGPT Write Like a Human
2. Train ChatGPT To Write Like You
Of all the ways to make ChatGPT undetectable, this is my favorite.
This method will make your content more human and avoid arousing the suspicions of your teacher, boss, or anyone else who is checking your content.
Avoiding triggering an AI detector is only part of the battle. If you fool the detector by generating text you clearly didn’t write, whoever is reviewing your work may notice that your submitted work doesn’t match your other writing samples.
Training ChatGPT to write like you solves this problem.
Use my guide to implement this: How To Make ChatGPT Write Like You
3. Use GPT-4
GPT-3.5 — the model available in the free version of ChatGPT — isn’t as sophisticated as GPT-4, which is available with a ChatGPT Plus subscription.
If you use GPT-3.5, your writing is more likely to be flagged as AI.
If you don’t want to subscribe to ChatGPT Plus, you can access GPT-4 for free by using Bing Chat. It’s built on the GPT-4 model.
4. Use a Humanizing AI Tool
There are tools specifically designed to humanize ChatGPT output and avoid triggering AI detectors.
My favorite tool is Undetectable.ai. It understands how AI detectors work, and it rewrites your text to remove the subtle linguistic patterns, sentence structure, vocabulary choices, and other tell-tale signs of AI-generated content.
You can tailor your output based on readability level, the intended purpose of the writing, and the level of humanization you want.
After it generates your text, it will tell you how likely the new text will trigger 8 of the most popular AI detectors.
5. Use A ChatGPT Alternative
Many AI detectors are actually ChatGPT detectors. There are other generative AI tools created with different training methods that are less likely to trigger AI detectors.
Here’s a list of ChatGPT alternatives you can try:
6. Provide Detailed Prompts
If you use simple prompts that only contain a couple of sentences, you’re going to get the most generic ChatGPT output possible — which is exactly what AI detectors are looking for.
Put some thought into your prompts. Here are some suggestions:
- Do preliminary research. Explain to ChatGPT what arguments, descriptions, quotes, statistics, sources, or analyses it should use.
- Share the goal of the text. Is it an essay trying to prove a thesis? A blog post that aims to be the most helpful article on the topic?
- Specify the tone and style you want: formal, casual, humorous, persuasive, etc.
- Incorporate specific keywords or phrases you’d like the AI to touch upon.
- Ask ChatGPT to structure the content in a specific manner, such as beginning with an introduction, followed by main points, and ending with a conclusion.
- Provide context or background information that will aid in generating more tailored content.
- Specify your target audience so ChatGPT can tailor its language and examples accordingly.
- Highlight any biases or misconceptions you want the AI to avoid or address.
- Ask for multiple perspectives or angles on a topic to ensure a comprehensive view.
- Encourage creativity by asking ChatGPT to think “outside the box” or to present unconventional viewpoints.
- Specify any limitations or boundaries for the content, such as word count, topic constraints, or sensitivity issues.
The more guidance you provide, the more unique ChatGPT’s output will be.
7. Use Custom Behavior
The Custom Behavior setting is an easy way to add writing guidelines that will be automatically considered in every answer ChatGPT provides.
If you use detailed guidelines that humanize ChatGPT’s output — like the ones provided in my guides mentioned above — add them to Custom Behavior to ensure everything ChatGPT.
Here’s how you can add instructions to Custom Behavior:
First, click the three dots in the lower left corner of ChatGPT dashboard, then click “Custom instructions”
Then, you can enter your custom instructions, click “Enable for new chats”, and click “Save”.
Now every new chat you create will be written using these instructions. Just remember to disable them whenever you don’t want to use them.
8. Use Complex Vocabulary and Sentence Structure
A study titled GPT Detectors are biased against non-native writers found that writing samples from non-native English speakers were more likely to falsely trigger AI detectors because they use simpler language and sentences.
Therefore, your ChatGPT-generated content should have rich vocabulary and diversity of sentence structure.
This doesn’t mean you should have a bunch of really long sentences. In fact, overly wordy sentences are a common sign of ChatGPT output.
Instead, you should mix up your sentence lengths. Some should be short. Some should be a little longer. And some should be a bit wordy when the context warrants it.
Use this prompt to add complexity to ChatGPT’s output:
Please rewrite the following text to incorporate significantly more intricate vocabulary and elaborate sentence structures, ensuring it retains a human-like style and does not appear as AI-generated. The objective is to make the content sufficiently complex to avoid triggering AI detectors. Here’s the text: [Your sample text goes here]
Review the output and ensure the new text isn’t exceedingly complicated. Even if it passes an AI detector, overly complex language is a red flag for any humans reviewing your writing.
9. Add Human Errors and Variations
Unless prompted to make mistakes, ChatGPT is perfect when it comes to grammar and spelling.
By adding intentional errors to content generated by ChatGPT, you can improve your chances of passing an AI detector.
If you take this route, I only recommend introducing minor grammar errors. They will be less noticeable than typos, and an article with obvious typos will substantially decrease its quality.
You can ask ChatGPT to add errors to the text it generates. Use this prompt to do so:
Add a small number of minor grammar mistakes to the following writing sample: [Insert writing sample here]
10. Use a Paraphrasing Tool
If you’re going to use a paraphrasing tool, I recommend one specifically designed to humanize your content like Undetectable.ai.
However, other paraphrasing tools can still be worthwhile if your free trial runs out.
My favorite paraphrasers are Quillbot, Ref-n-Write, and Paraphraser.io
11. Ask ChatGPT To Write In A Particular Style
An easy way to decrease the likelihood of getting caught using ChatGPT is to ask ChatGPT to mimic the style of a well-known writer.
For the best results, use a writer who writes the same type of writing you’re generating.
For example, if you’re writing a news article, it’s smarter to ask ChatGPT to mimic Bob Woodward than Stephen King,
If you don’t know any famous writers who do the type of writing you’re having ChatGPT generate, ask ChatGPT to give you a list of “famous [insert style of writing] writers”.
If it knows who they are, it can probably emulate their style.
12. Add Personal Anecdotes
If the style of writing allows it, adding personal anecdotes and opinions can help you pass ChatGPT detectors.
ChatGPT doesn’t tend to share experiences it’s had — partially because it’s an AI model that doesn’t have the capacity to experience anything. Adding stories from your life can humanize your text and move it further from the typical ChatGPT output.
As a bonus, this can also help put the reviewer at ease if you’re concerned they might deem the work to be AI-generated even if you pass the detector.
13. Go Into Greater Depth
ChatGPT’s output tends to be generic and surface-level, and AI detectors are aware of that.
If you take the time to add unique examples, deeper analysis, and your own original thoughts to the piece, it can both decrease your odds of getting detected and make your writing better.
You can get ChatGPT to add this depth, though you’ll need to do some prep work before prompting.
Do your own research — maybe using Bing to help speed things up — and ask ChatGPT to add the extra analysis and examples in your prompt.
14. Use a Plagiarism Detector
While passing the AI detector is your main concern, you also need to be careful about ChatGPT directly plagiarizing from other sources.
Part of why ChatGPT ”knows” so much is because it’s been trained on an enormous dataset of internet resources, including books, articles, and Wikipedia pages.
While ChatGPT is usually pretty good at creating unique sentences, I have caught it generating word-for-word plagiarism when I use it. Here’s an example from a message I shared in a Discord channel:
Many AI detectors also have plagiarism detectors built in, and anyone running your writing through an AI detector is almost certainly using a plagiarism detector as well.
Submitting plagiarized content can not only increase your chances of triggering the AI detector, but it can also trigger the plagiarism detector.
I tested plagiarism detectors extensively when I was running my writing service, and I found CopyLeaks was the most accurate option. Here are the results of my tests if you’re curious:
If you can’t afford CopyLeaks, then Copyscape is a decent alternative.
The Grammarly Premium plagiarism detector isn’t great, but it’s better than nothing if you already have Grammarly.
15. Edit ChatGPT’s Output Manually
Editing ChatGPT’s output yourself is a great way to humanize it and reduce detection chances.
Go through the text and change the wording to something you’d be more likely to use. You can even use this as an extra opportunity to polish your content with Grammarly — my favorite AI-assisted editing tool.
Extra Tips To Avoid AI Detection
1. Create a New Google Doc Before Submitting Work
If you use Google Docs to submit your work, this is an important tip.
There is a Chrome plugin made by Originality AI that can analyze how quickly you added text to your document. Here’s what it looks like:
If you’re using ChatGPT to write, you’re likely pasting in large chunks of text in a relatively short time. The graph provided by Originality will look like this if you’re using AI:
This looks suspicious to someone using this plugin, and could get you caught even if your work doesn’t trigger the ChatGPT detector.
The plugin only works if you submit the same document you wrote the piece in, though. If you make and submit a copy of the Doc when you’re finished, you won’t have to worry about this.
2. Run Your Writing Through an AI detector.
If you know which AI detector will be used to analyze your writing, you can analyze it beforehand to see if it triggers the detector. Many of them have free trials, so you can run your tests without paying anything.
In my experiments, many detectors give different results each time you run it, so you’ll want to run it a few times to be extra sure.
What To Do If Your Writing Is Flagged As AI
If your writing is flagged as created by ChatGPT despite your best efforts, don’t panic. There are a few things you can do.
Share Documented Evidence of AI detectors Being Faulty
AI detectors are notorious for throwing false positives.
Here’s a list of news articles and academic studies that highlight how inaccurate these detectors can be:
- How to prove your innocence after a false positive from Turnitin, GPTZero: This article from The Washington Post provides advice from academics, AI scientists, and students on how to deal with being accused of AI cheating by AI detectors.
- We tested Turnitin’s ChatGPT-detector for teachers. It got some wrong.: This article from The Washington Post reports the results of a test conducted by the tech columnist and five high school students on Turnitin’s AI detector. It shows that the detector flagged an innocent student’s essay as AI-generated.
- Turnitin says its AI cheating detector isn’t always reliable: This article from The Washington Post reveals that Turnitin has acknowledged a reliability problem with its AI cheating-detection software, which is used on 38 million student papers. It also cites computer scientists who warn that we may never be able to reliably detect AI.
- Turnitin admits there are some cases of higher false positives in AI detector: This article from K-12 Dive reports that Turnitin has admitted that its AI detector has a higher incidence of false positives when less than 20% of AI writing is detected in a document.
- How AI detection tool spawned a false cheating case at UC Davis: This article from USA TODAY tells the story of a student who was falsely accused of using AI to cheat on his history exam by his professor, who used GPTZero and other AI detection tools.
- AI Detection Tools Falsely Accuse International Students of Cheating: This article from The Markup reports that Turnitin’s AI detection tool was much more likely to flag international students’ writing as AI-generated, and cites a study by Stanford computer scientists that found a clear bias.
- Programs to detect AI discriminate against non-native English speakers, shows study: This article from The Guardian summarizes the findings of the same study by Stanford computer scientists, and discusses the implications for students, academics, and job applicants.
Sharing these news stories can sow rightfully deserved doubt about the detectors’ effectiveness.
Run The Reviewer’s Published Work Through An AI Detector
The best way to convince someone that AI detectors can’t be trusted is to run their own writing through one.
If the detector says it’s AI-generated — which happens more often than you might think — you can share the results and hopefully change their minds.
Run Your Content Through Other AI Detectors
Just because your work was flagged as AI in one detector doesn’t mean it will be flagged in another.
If the other detectors are giving a negative trigger, send the reviewer the results and demand they explain why their AI detector is better than the others.
Summing Up
AI detectors are frustrating to deal with, but there are many tactics you can use to avoid detection.
If you want more tips on how to use AI more effectively, use the box below to join the Future Guidebook newsletter: