When I first examined whether it is possible to fight back against AI-borne literary theft and how this approach can work, it was January 2023-a few months about awareness about the generosity of the world.
Also: Top 20 AI tools of 2025 – and #1 to remember the cheese when you use them
This is a completely updated version of that original January 2023 article. When I first tested GPT detectors, the best result was one to 66% of the three available checkers. In February 2025, my most recent test, 10 checkers were used – and three of them had the right score. This time, a few months later, five did it.
For whom am i testing and how am I doing it
Before I go ahead, let’s discuss literary theft and how it is related to our problem. Merium-Wabster defines “PlagiarizeTo steal and pass “as” thoughts or words) as their own; Use the source without creating (production of another). ,
This definition fits the well-made material. While someone is using AI tool Perception AI Or Puffy The material is not stealing, if the person does not credits the words coming from AI and claims them as their own, it still fulfills the dictionary of literary theft.
Also: The Dead Giveway’s chat wrote your content – and how to work around it
To test AI detectors, I am using five blocks of the lesson. Two were written by me and three were written by Chatp. To test the material detector, I feed each block separately to the detector and record the result. If the detector is correct, I consider passing the test; If this is wrong, I consider it a failure.
When a detector provides one percent, I consider anything above 70% as a strong possibility-in favor of human-written or AI-written material-and consider the answer to the detector. If you want to test a material detector yourself using the same text blocks, you can pull them this document,
Overall result
To evaluate AI detectors, I re-prepare my five-observe series in 10 detectors. In other words, I cut and glued 50 individual tests (I had a Very Of coffee).
I tested detectors Brandwell, Copilix, GPT -2 output detector, Gptzero, Grammatically, Monica, Originality., Quilbot, Undetectable.ai, Author.comAnd Zero.
Too: 3 clever chat tricks that prove it is still to defeat AI
For this update, I added Copilix and Monica. I dropped from my tests in writing because it Close Its GPT detector. Material mentor Requested inclusion, but I did not hear back on time to test accounts.
This table shows overall results. As you can see, five detectors correctly recognize human and AI text in all tests.
I tried to find out if there was a solid pattern of improvement over time, so I created a chart compared to five-testing set over time. So far, I have run this series six times, but there is no strong tendency. I increased the number of tested detectors and swapped some, but the only consistent result is that the test 5 was retired firmly as human in detectors and dates.
I will continue to test over time, and hopefully I will constantly see the trend of reliability upwards.
While some are score, I do not recommend fully relying on these devices to validate human-written materials. As shown, writing from non-country speakers is often rated as being generated by AI.
Even though my hand-prepared content was mostly rated as a human-written, a detector (GPTZERO) declared himself very uncertain for the judge, and the other declared it to be AI-written. The results are inconsistent in the system.
Also: Best AI Chatbots: Chat, Copilot, and notable options
Bottom Line: I will advocate caution before relying on any of these devices – or the results of all.
How did each AI material detector perform
Now, let’s look at each individual test tool, listed in the alphabet.
Brandwell AI material detection (accuracy 40%)
This device was originally manufactured by the AI Content Generation Firm, Content at Scale. Later it left Brandwell.ai, A new name for the AI-centric marketing services company.
Too: AI -Janit Pictures are a legal mess – and still a very human process
Unfortunately, its accuracy was low. This device was unable to explain whether the AI-related material in Test 2 was human or AI, as shown in this screenshot:
Copilix (80%accuracy)
I think it’s entertaining Copilix When the tested detectors performed better more than half of the detectors, declares themselves “the most accurate AI detector with more than 99% accuracy”. But marketing people will be in the form of marketing people – superlatives are equally difficult for them, which is for my dog to resist in a squirrel (and fedex truck, and all neighboring children).
Too: 5 Quick Ways Apple’s AI tool can fix your writing on fly
The company’s primary offering is a literary theft checker sold to educational institutions, publishers and enterprises, which is willing to ensure the originality of the material and maintain educational integrity.
GPT -2 output detector (accuracy 60%)
This device was made using a machine-learning hub managed by New York-based AI Company. Throat faceWhile the company has received $ 40 million in funding to develop its natural language library, GPT -2 detector Hugging face transformer seems to be a user-made tool using the library.
GPTZERO (80%accuracy)
Gptzero It is clearly growing. When I first tested it, the site was bare-baked-it was also not clear whether GPTZERO was a company or just a passion project of someone. Now, the company has a complete team, with a full team with a mission of “protecting human beings”. It offers AI verification tools and a literary theft checker.
Too: The most popular AI device of 2025 (and what does it mean)
Unfortunately, the performance has declined. In my previous two runs, GPTZERO correctly recognized my text as a human being. This time, it declared the same text as AI-Janit.
Grammar (accuracy 40%)
Grammatically Writers are well known to help produce grammatically correct material – it’s not what I am testing here. Grammarli can check literary theft and AI material. In grammar checkers, the lower-right corner has a literary theft and AI text check button:
I am not measuring literary theft checker accuracy here, but even though Gramarli’s AI-check accuracy was poor, the site previously published the test text correctly.
Monica (accuracy 100%)
Monica There is a new entry. It provides an all-in-one AI accessory with a wide range of services services. Users can choose from different large language models.
Too: 5 ways can help you write essays chatgate
The company calls Monica “Best AI Detector Online”, but it seems that it runs the content through other detectors including zirogpet, giptzero and copilix. Funny, both GPtzero and Copilix did not perform well in my tests, but Monica – and Sherogpt – did.
We are giving it 100% because it has earned that rating, but I will see how it stands in future tests.
Originality .AI (accuracy 100%)
Originality. There is a commercial service that bills itself as AI and literary theft checkers. The company uses the use credit: I used 30 credits for this article. They sell 2,000 credits per month for $ 12.95. I pumped 1,400 words through the system and used only 1.5% of my monthly allocation.
Quilbot (accuracy 100%)
Last time i tested QuilbotThe results were wildly inconsistent – many of the same text had a different score. This time, however, this rock was solid and 100% correct. So I am winning it. I will see back to see in a few months whether it is on this performance.
Undetectable.ai (accuracy 100%)
Undetectable.ai’s The big claim is that it can “humanito” the AI-borne text so that the detector does not flag it. I have not tested that feature – it bothers me as a professional writer and teacher, as it sounds like cheating.
Too: Why you should ignore 99% AI tools – and what four I use every day
However, the company also has an AI detector, which was very high at the point.
The AI detector passed all five tests. Notice indicators showing flags for other detectors. The company said, “We have developed several detector algorithms to provide a federed and unanimous approach after those major detectors. They do not feed in the model listed directly; rather, models are trained on the basis of the results that they have produced. It is based on the models that we have created and the models that we have created, it is based on the models that we have created and the models that have been destroyed. Are based on the algorithm made. ”
Too: How to use chat: a starting guide for the most popular AI Chatboat
I have a question about openi flag, since In 2023, the content detector of Openai was discontinued due to low accuracyNevertheless, Undetectable.ai explored all five tests, earning a full 100%.
Author.com AI Content Detector (40%)
Author.com There is a service that produces AI writing for corporate teams. Its AI Material Detector The equipment can scan for the ingredients generated. Unfortunately, its accuracy was low. It recognized each text block as a human-written, even though three out of six tests were written by CHATGPT.
Zero (accuracy 100%)
Zero Since I have last evaluated it, it has matured. Then, the name of any company was not listed, and the site was paired with Google advertisements and lacking clear mugration. Seva worked quite well but it seemed sketch.
Too: Will AI destroy human creativity? No – and why is here
He has realized the sketch. Zerogpt now presents as a specific mother -in -law service, with pricing, company name and contact information. It also increased its accuracy: last time it was 80%; This time it scored 5 out of 5.
Is it human, or is it AI?
How are you? Have you tried AI content detectors such as Copilix, Monica or Zerogpt? How accurate are they in your experience? Have you used these devices to protect educational or editorial integrity? Have you faced situations where human-written tasks were accidentally marked as AI? Are there detectors that you trust more than others to evaluate originality? Let us know in the comments below.
Get top stories of morning with us in your inbox every day Tech Today Newsletter.
You can follow my day-to-day project updates on social media. Be sure to subscribe to My weekly update newsletterAnd follow me on Twitter/X @DavidgewirtzOn Facebook Facebook.com/davidgewirtzOn Instagram Instagram.com/davidgewirtzOn blue @Davidgewirtz.comAnd on youtube Youtube.com/davidgewirtztv,
