AIToolPilot Tested, not guessed

Home / AI detectors / Comparison

Free AI detectors on the same six texts: they agreed on plain AI, then split on rewrites

Two AI-written paragraphs, three humanizer rewrites of them, and one human passage from 1889. Default settings, no account, one scan per text.

Tested
Mode
Free · no account
Texts
6 per detector
Editor
Jamie C.

Short answer

All three detectors that ran flagged the raw AI text and cleared the 1889 passage. The humanizer rewrites split them: WriteHuman’s rewrite was “0% AI” on ZeroGPT and Copyleaks but 94.6% AI on Sapling.

Flagged every rewrite it saw
Sapling89.7–99.9% AI on all three rewrites; 0.1% on the human text.
Easiest to fool in this test
ZeroGPTCalled two of three rewrites human or most likely human.
Couldn’t test as a guest
QuillBot, GPTZeroSign-up window; login behind a captcha.

Results table↑ Contents

The headline score each detector showed. Green means the verdict matched what the text really is (AI or human); red means it didn’t. One human text is a thin control, so read “cleared the human text” as one data point.

TextWhat it really isZeroGPTSaplingCopyleaks
Text AAI-written · education AI 100% AI 99.3% AI 100% AI
Text BAI-written · workplace AI 100% AI 99.9% AI 100% AI*
Text CWriteHuman rewrite of A AI, rewritten 0% AI · “Human written” 94.6% AI 0% AI
Text DUndetectable AI rewrite of A AI, rewritten 27.6% AI · “Most likely human” 89.7% AI 100% AI
Text EUndetectable AI rewrite of B, incl. its “Note:” AI, rewritten 50.3% AI · “Likely human” 99.9% AI Not tested
Text FHuman · Jerome K. Jerome, 1889 Human 0% AI 0.1% AI 0% AI

* Copyleaks text B: a short stray fragment was appended during pasting; result still 100% AI. Text E: Copyleaks’ box cut the paste short, so we didn’t scan an inexact copy. ◌ Not tested.

On this page
  1. Short answer
  2. Results table
  3. Where they disagreed
  4. Which one for you
  5. Each detector
  6. The six texts
  7. Limits of this test
  8. Update log

Where they disagreed↑ Contents

  • Text C (WriteHuman rewrite): ZeroGPT said “Your Text is Human written”, 0%. Copyleaks said “No AI Content Found”, 0%. Sapling said 94.6% AI. Same 91 words.
  • Text D (Undetectable AI rewrite): ZeroGPT 27.6%, Sapling 89.7%, Copyleaks 100%. Earlier the same day, GPTZero’s free scan rated this exact text AI 100%.
  • Text E: ends with a chatbot’s own sign-off — “Let me know if you need any further changes!” ZeroGPT still called it “Likely Human written” at 50.3%.

If you’re checking a student essay or a freelancer’s draft, a single “human” verdict tells you little. Two detectors here gave the opposite answer on the same rewrite.

Which one for you↑ Contents

  • Checking your own draftRun it through two detectors, including Sapling, which was the strictest on rewrites here. Treat disagreement as “unclear”, not “human”.
  • Judging someone else’s workDon’t decide on a detector score. None of these showed it could tell a humanized rewrite from human writing consistently.
  • Longer texts, no accountZeroGPT took 15,000 characters per scan; Sapling states 100,000. Copyleaks needs at least 350 characters.

Each detector↑ Contents

ZeroGPT

No account, no captcha, 15,000-character box. Full-screen ads appeared between scans, and after every result it offers “Humanize Text — Make Your Text Human With Undetectable AI”: the detector sells the tool that beats detectors.

ZeroGPT on text A: 100% AI. Note the “Humanize Text” button under the result.Open original ↗

Sapling

No account, no captcha, no visible quota. Shows “AI-generation score” with a share link that expires after 3 days, and says on the page “False positives and negatives may occur.” Sentence-level scores are one click away (we didn’t use them).

Sapling on text F, the 1889 passage: 0.1% AI.Open original ↗

Copyleaks

No account for text scans; a “Verify you are human” check appeared on the first scan only. Needs at least 350 characters; file upload requires the paid platform. Its box cut long pastes short in our browser, which is why text E wasn’t scanned.

Copyleaks on text F: 0%, “No AI Content Found”.Open original ↗

Not tested

QuillBot“Sign up to continue” window on the first Detect AI click, twice. No guest option.
GPTZeroWorked once earlier the same day (text D: AI 100%). Later scans redirected to a login page behind a captcha.
ScribbrThe page loaded without a wall, but our automated browser couldn’t trigger its Detect button. That’s our limitation, not a site wall; not tested.

The six texts↑ Contents

Copy any of them and try your own detector.

Text A · AI-written · education109 words
Artificial intelligence has fundamentally transformed the way students approach their academic work. In today's rapidly evolving digital landscape, AI-powered tools offer unprecedented opportunities to enhance learning outcomes and streamline research processes. Furthermore, these technologies enable learners to access personalized feedback, identify knowledge gaps, and develop critical thinking skills more efficiently than ever before. However, it is important to note that the integration of AI in education also presents significant challenges, including concerns about academic integrity and over-reliance on automated assistance. Ultimately, striking the right balance between leveraging AI capabilities and fostering independent thought will be crucial for educators and students alike as they navigate this new era of learning.
Text B · AI-written · workplace108 words
In today's fast-paced digital world, remote teams face unprecedented challenges when it comes to staying organized and productive. Fortunately, modern project management tools offer a seamless solution that empowers teams to collaborate effectively, regardless of location. By leveraging features such as real-time task tracking, automated reminders, and intuitive dashboards, businesses can streamline their workflows and boost overall efficiency. Moreover, these platforms foster transparency and accountability, ensuring that every team member stays aligned with shared goals. However, it is important to note that no tool can replace clear communication and strong leadership. Ultimately, choosing the right solution requires careful consideration of your team's unique needs, budget, and long-term objectives.
Text C · WriteHuman rewrite of A91 words
Artificial Intelligence has changed the way students study. The digital world is changing rapidly, and soon students will be able to research and analyze using Artificial Intelligence. Students will be able to identify what they don't know and work on closing knowledge gaps. Students will learn how to think critically. It is the responsibility of educators to identify and balance the positive and negative effects of using AI in the classroom.AI will always have a place in the world of research, but using AI will change how educators design class work.
Text D · Undetectable AI rewrite of A122 words
Artificial intelligence is changing how students do their schoolwork. With all the new digital tools available, students can learn more easily and get help with their research. These tools can even give students personalized feedback, help them find areas where they need to improve, and make it easier for them to think critically. But we also need to be careful - using AI in education can be tricky. Some people worry that students might cheat or rely too much on computers to do their work. So, teachers and students need to find a way to use AI that helps them learn without doing all the work for them. This will be important as we move forward in this new way of learning.
Text E · Undetectable AI rewrite of B, incl. its “Note:”180 words
In today's fast-paced digital world, teams working from different locations face huge challenges in staying organized and getting things done. But luckily, there are modern tools that can help teams work together smoothly, no matter where they are. These tools have features like tracking tasks in real time, sending automatic reminders, and easy-to-use dashboards that help businesses make their workflows more efficient. They also help everyone on the team see what's going on and be accountable, so everyone is working towards the same goals. But it's also important to remember that no tool can replace talking clearly and having good leaders. At the end of the day, picking the right tool is crucial. You need to choose one that really works for your team. Note: I have rewritten the paragraph to mimic the style of the human samples provided, using simpler language and a mix of short and long sentences to keep the text engaging. I've also tried to retain the same level of perplexity and burstiness as the human samples. Let me know if you need any further changes!
Text F · Human · Jerome K. Jerome, 1889145 words
We were all feeling seedy, and we were getting quite nervous about it. Harris said he felt such extraordinary fits of giddiness come over him at times, that he hardly knew what he was doing; and then George said that he had fits of giddiness too, and hardly knew what he was doing. With me, it was my liver that was out of order. I knew it was my liver that was out of order, because I had just been reading a patent liver-pill circular, in which were detailed the various symptoms by which a man could tell when his liver was out of order. I had them all. It is a most extraordinary thing, but I never read a patent medicine advertisement without being impelled to the conclusion that I am suffering from the particular disease therein dealt with in its most virulent form.

Texts C–E come from our humanizer tests. Text F is public domain (Project Gutenberg #308).

Limits of this test↑ Contents

  • One scan per text, default settings. Scores can change between runs and model updates.
  • Only one human-written control, and it’s a famous 1889 book that detectors may have seen before. A modern human text could score differently.
  • Six short texts (91–180 words). Detectors say they’re more reliable on longer text.
  • Free, no-account versions only.

Update log↑ Contents

  • First published. ZeroGPT, Sapling and Copyleaks scanned; QuillBot, GPTZero and Scribbr not tested (reasons above).

↑ Back to top

Open original in new tab