AI Writing
Twixify: Can This AI Rewriter Really Trick AI Detectors?
We put AI rewriter Twixify to the test to see if it can make AI content undetectable by AI detectors like Originality.ai, Writer, and ZeroGPT.
Twixify is yet another tool in the long list of AI “humanizers” that purports to make your ChatGPT (or other AI-generated) text “undetectable” by AI detectors.
Can it outwit the Originality.ai AI detector or other AI detection tools? We put Twixify to the test to see how it performs.
Launched in July of 2024, Twixify claims to not only be able to make your AI-generated writing undetectable by modern AI writing detectors but also make sure that the output matches your unique writing style.
Their blog offers a number of suggestions and tips on how to use Twixify or ChatGPT directly to reformulate and humanize articles.
Twixify offers a free trial, limited to 150 words per cycle and three requests per day. Then, its prices start at $8/month and move up to $27/month for higher limits.
Given that the site recently launched in just the past few months, its brand isn’t as well-known as some other popular AI rewriters.
However, this tool, in particular, has been catching on in various writing circles, particularly among freelancers who are looking for a way to humanize their AI-crafted content.
Check out our reviews of other AI rewriting tools in our guide on undetectable AI.
Twixify recently introduced a brand-new version of its platform, so we put it to the test!
Step 1: We generated a brief article with GPT-4o according to the prompt below.
Step 2: We added the generated article directly into Twixify.
Step 3: Next, we then ran a separate test according to guidelines published on their blog, which we’ll refer to as the burstiness/perplexity version.
Step 4: We then ran the generated results through several AI detectors, including Originality.ai Turbo and compiled the results below.
Here is our prompt: Write me a short article on the development of the coffee bean. Make it sound as human as possible.
Original content (GPT-4o generated): 100% confidence the text is Likely AI
Out of all four AI detectors, the 3.0 Turbo version of Originality.ai was the only one to accurately determine that all three versions of the article (the original GPT-4o generated content, the “Twixified” version, and the “burstiness/perplexity” version) were Likely AI with 100% confidence.
Twixify version: 100% confidence the text is Likely AI
Even after processing the text through Twixify, the Originality.ai AI detector continued to identify it as Likely AI with 100% confidence — even though there was extensive paraphrasing and rewriting.
Burstiness/perplexity version: 100% confidence the text is Likely AI
Then, for a further test, we applied Twixify’s prompt recommendations for making AI text ‘undetectable’ using ChatGPT. Even with the guidelines, Originality.ai AI checker continued to showcase exceptional performance in detecting the text as Likely AI with 100% confidence.
Original content (GPT-4o generated): 85.55% AI
ZeroGPT, at first, noted that the original text was AI-generated with an over 85% likelihood.
From there, however, the results shifted dramatically.
Twixify version: 18.83% AI
With the paraphrasing, reworking, and rewriting via the “Twixified” version, the text became “Most Likely Human Written.”
What about the version created with Twixify’s burstiness/perplexity prompt?
Burstiness/perplexity version: 15.52% AI
The burstiness/perplexity version got an even higher score as human-written.
After running each piece through ZeroGPT, we then turned to Writer, which boasts of a similarly reliable AI detector as part of its suite of services.
How did it do when we put it to the test?
Writer Score:
Original content (GPT-4o generated): 77% Human-Generated
The original was noted as 77% human-generated by Writer’s AI detector, even when it was produced by AI.
Twixify version: 89% Human-Generated
The humanization notably increased the ‘human-generated’ score for Writer’s AI detector, but what about the burstiness/perplexity prompt?
Burstiness/perplexity version: 91% Human-Generated
The burstiness/perplexity prompt clocked in at 91% human-generated.
We put the popular AI detector and plagiarism scanner offered by Copyleaks to the test to see if it could uncover AI in plain sight, as well as a more humanized version.
Rather than give a percentage, CopyLeaks only reports (in the free version) as to whether or not the piece is human text.
Original content (GPT-4o generated): Human text
The original ChatGPT-generated version was highlighted as human.
Twixify version: Human text
The Twixify version got a similar human score.
Burstiness/perplexity version: Human text
The burstiness and perplexity prompt was also marked as human by Copyleaks.
The results speak for themselves; the Originality.ai AI detector was able to identify the GPT-4o, Twixfy humanized, and ‘burstiness/perplexity’ versions of AI-generated text as Likely AI with 100% confidence.
Looking to learn more about AI detection? Check out our guide on AI detector accuracy as well as a meta-analysis of AI detection studies from third parties.
Twixify, launched in July of 2024, as an AI humanizer tool that claims to use a variety of techniques to make AI-generated text undetectable. It also offers suggestions on how to bypass AI detection using its service as well as ChatGPT-specific prompts.
From the results of our brief study, Originality.ai effectively detected Twixify’s AI-humanized text as AI-generated. So, Twixify AI text is detectable by Originality.ai.
