Walter Writes Review: Can It Bypass AI Detectors As It Claims?

Many AI text humanizers claim that they can remove traces of AI from AI-generated text and bypass AI detectors. But most of them cannot do that fully, and even if they do, the original text loses some of its meaning and structure.

Researchers from the University of Maryland and the Max Planck Institute, publishing at ICLR 2024, had human volunteers deliberately rewrite AI-generated text with the explicit goal of stripping any trace of its origin.

Despite their best efforts, the watermark remained detectable with high confidence (at a false-positive rate of just 1 in 100,000).

Today, I’ll be reviewing another AI humanizer called Walter Writes that claims to bypass major AI detectors.

I’ll be generating text with AI, humanizing it with Walter Writes, and scanning both samples with seven major AI detectors.

Stick around to find out whether Walter Writes manages to bypass them.

What is Walter Writes AI?

Walter Writes AI launched in mid 2024 as an AI humanizer. Then, in late 2024, it added an AI detector to the same workflow so humanized text gets checked for AI immediately.

Side Note: Most AI humanizer websites that also offer AI detectors do not have a high accuracy rate. They offer AI detectors to give users peace of mind that the text they humanize won't be flagged as AI. But in my experience reviewing AI humanizers like WriteHuman AI, Grubby AI, and many more, I have found that built-in AI detectors of these tools are highly inaccurate and biased towards the platform. 

Today, Walter Writes AI still bills itself as a tool for humanizing AI-generated content, despite adding more AI-powered tools over time.

For instance, Walter Writes now offers a paraphraser, grammar checker, and resume writer. 

The AI humanizer can be used for free even without creating an account. It’s not clear how many words you can humanize per request on the free version, but considering the 750 words/request on the basic paid plan, the limit on the free plan must be lower than that. 

However, I have been able to humanize over 1,000 words on the free plan, so it’s not clear if Walter Writes enforces these limits. 

There are two modes you can choose from. The Lite mode is available for free, while the Standard mode unlocks with paid plans. 

Walter Writes claims that AI-text humanized by its AI humanizer can bypass major AI detectors like GPTZero, Turnitin, Proofademic, and Originality. It says that the humanized text can achieve up to a 100% human score on these four AI detectors.

But does it actually bypass major AI detectors? Let’s find that out in my hands-on testing of Walter Writes AI against 7 AI detectors. 

How I Tested The AI Humanizer by Walter Writes AI 

I asked ChatGPT to generate 200 words on any random topic. The generated text had a heavy, typical AI-writing style that serves as an ideal input for an AI humanizer.

The screenshot below shows the text generated by ChatGPT:

I ran the ChatGPT-generated text through the following AI detectors:

• GPTZero
• Originality AI
• ZeroGPT
• Copyleaks
• Proofademic
• Pangram
• Sapling AI

Then I fed the text to Walter Writes AI humanizer, and it returned around the same number of words after humanizing. You can see the changes it made highlighted in a different color in the screenshot below. Notice how it barely replaced a few words. There were no sentence-level changes. 

Walter Writes reported a 15% AI score for the humanized text.

I ran the humanized text through the same AI detectors to see if they report the same. 

Here are the results: 

Results: Can Walter Writes AI Bypass AI Detection?

The text humanized by Walter Writes couldn’t fully bypass any AI detector, except for Proofademic and ZeroGPT.

The table below shows the AI scores for the original and humanized text reported by the 7 AI detectors:

Read on for further details and screenshots of the AI scans.

GPTZero

Walter Writes AI badly failed against GPTZero. It was the most accurate of all the detectors in this test.

GPTZero was able to detect both the original and humanized text as 100% AI. But it could also distinguish which of the two was raw AI text and which was paraphrased by AI.

The screenshot below shows GPTZero’s results for the original ChatGPT-generated text: 

GPTZero flagged this as 100% AI and marked it as AI-generated. 

The screenshot below shows how GPTZero scored the humanized text by Walter Writes: 

GPTZero also gave a 100% AI score to the humanized text, but with a different verdict. To quote GPTZero, “We are highly confident this text was originally AI, but rewritten by AI or a human.”

Despite very minor changes made by Walter Writes, GPTZero was able to tell that the text was originally raw AI and later altered with an AI tool or by a human.

Verdict: Walter Writes AI cannot bypass GPTZero.

Originality AI

Originality AI now reports its results in a confusing way if you don’t have a paid plan.

There are two percentages, one showing how much of the text appears to be AI and the other showing its confidence in that estimate.

The ChatGPT-generated text was declared more than 40% AI with 100% confidence. It cannot be said exactly how much of the text is AI beyond 40% because Originality doesn’t report like that anymore. But since the whole text was highlighted in red (likely AI), it’s safe to assume that Originality flagged 100% of the content as AI.

Then I scanned the humanized text with it, and the scores were similar. Originality flagged more than 40% of the text as AI, with 90% confidence.

The exact verdict is “AI use appears to exceed 40%.”

But how much does it exceed 40%? 45% and 95% are both greater than 40%, but they are drastically different.

This confusing reporting system doesn’t let us tell exactly how much Walter Writes bypassed Originality.

Verdict: Walter Writes cannot bypass Originality AI.

ZeroGPT

ZeroGPT isn’t named by Walter Writes, but I still added it to the test because it’s a popular tool. 

Walter Writes was somewhat successful against ZeroGPT, especially when compared with other AI detectors like GPTZero, Pangram, etc.

ZeroGPT scored the original AI text as 100% AI.

But the AI score for humanized text dropped significantly. It could only detect 62.9% of the text as AI.

And the verdict was “Your Text is Likely Human Written, may include parts generated by AI/GPT.”

It’s no surprise since ZeroGPT has shown mediocre performance previously as well in my detailed ZeroGPT AI content detector review.

After this verdict, it’s safe to say that Walter Writes managed to bypass ZeroGPT, albeit not fully.

Verdict: Walter Writes can bypass ZeroGPT.

Copyleaks

Copyleaks was able to detect both the original and humanized text as 100% AI.

The screenshot below shows the AI detection results for the original ChatGPT sample:

And here’s the AI detection report for the humanized sample: 

Verdict: WalterWrites cannot bypass Copyleaks.

Proofademic

Walter Writes claims to bypass Proofademic’s AI detector, so I tried running our ChatGPT text sample and the humanized version of Walter Writes through it.

Initially, I couldn’t sign up on the platform after multiple tries, which I figured was a bug. I tried again after a couple of hours, and it worked.

Then I tried running the Original ChatGPT text sample through Proofademic’s AI detector, and to my surprise, unlike other AI detectors on this list, it couldn’t flag the content as AI and said there’s a 74% probability that the content is human-generated.

Their verdict was “We are highly confident this text was human written,” although the text was straight from ChatGPT without any humanization done on it. 

After seeing its wrong call on raw AI text, I had a feeling that it would perform the same on the humanized sample as well, and I was right. The humanized version of the text that came from Walter Writes was identified as human by Proofademic.

So for Proofademic, Walter Writes AI’s claim proved to be technically true, but the fact that Proofademic could not even detect raw AI text says a lot about its accuracy.

Verdict: Walter Writes AI can bypass Proofademic’s AI detector.

Pangram

Walter Writes AI also fully failed against Pangram. 

Pangram reported a 100% AI score to the original ChatGPT text: 

The entire humanized text was also scored 100% AI by Pangram: 

Verdict: Walter Writes cannot bypass Pangram.

Sapling AI

Sapling AI succeeded in detecting the humanized text as fully AI-generated.

Its AI score for the original AI text was 98.4%:

The humanized text by Walter Writes could bring down the AI score by less than 1 percentage point. The AI score for the humanized text was 97.7%:

Verdict: Walter Writes AI cannot bypass Sapling AI.

What Does Walter Writes AI Cost

The Lite mode of Walter Writes makes small changes to your text.

The Standard mode is reserved for paid users only. 

Here’s what the four paid plans cost:

These plans are also available for yearly subscription for a lower per-month rate.

One thing worth noting in the plans is that the max words/request limit is 2,000. In other words, you cannot humanize more than 2,000 words in one attempt, even with the most expensive plan of Walter Write. 

Use GPTZero to Catch AI Humanizers

Walter Writes AI couldn’t fully bypass a single AI detector, except for Proofademic. It was caught either fully or partially by 6 out of the 7 AI detectors I ran the humanized output on.

But GPTZero was the only tool that truly caught it with perfectly accurate verdicts. It is the only AI detector in this test that differentiates between raw AI-written text and AI-text that has gone through alteration by humans or other AI tools such as AI humanizers.

GPTZero AI detector can do this because it was separately trained on a wide range of paraphrasing and bypassing techniques. 

It can also detect when a given text contains a mix of human and AI writing. 

You can scan 10K words with GPTZero for free every month. Our detection model is the same for both free and paid users.

Try GPTZero for free today.

FAQs

What is the most trusted AI humanizer?

AI humanizers' output varies, and there’s no single most trusted AI humanizer out there. Some AI humanizers completely change the meaning of your original text by replacing full sentences with fillers. Then there are some that keep the meaning of your text the same but change certain sentence structures and phrases to sound more human and bypass AI detectors. However, top AI detectors like GPTZero can detect whether the content is pure AI or paraphrased using AI or human help.

Is Walter Write AI free?

Walter Write AI does not have a traditional free plan, but it allows you to humanize text for free for up to 300 words at once. If you want to increase the limit of words you can humanize, you’ll have to upgrade to one of their paid plans.

Which AI writing checker is the most accurate?

GPTZero is the most accurate commercial AI writing checker out there, and it has proven it in independent benchmarks several times. Apart from GPTZero, Pangram, Originality, and Winston AI detector are also accurate.