Whether you’re proofreading essays, reviewing job applications, or simply browsing the web and scrolling through social media, it’s a sign of our times that you might be wondering: did an AI write this?

There are supposedly telltale signs of AI writing, including the script. There are also several tools available that compare writing to combinations of these AI ticks to tell you whether or not something was written by a human.

However, can AI-produced text be identified as regularly as these tools claim? We decided to put them in text and test them: five different tokens against a selection of samples written by me, ChatGPT, Gemini or Claude.

I selected a selection of introductions from articles I recently wrote (all without AI help, by the way), then had the AI ​​bots write their own 150-word versions based on the title and premise of each article.

Then it was turned to the AI ​​detectors to see if they could tell the difference, and here are the results.

pangram

Pangram promises to be “an AI detector that really works.” You can run up to four queries a day for free, with much higher limits and extras like plagiarism detection available with a subscription starting at $20 a month.

The good news for me was that both samples of my own writing turned out to be 100 percent written by a human, and Pangram declared a “high” level of confidence in each decision. Whatever my writing style is, it’s apparently not that of AI.

I gave Pangram some AI-written intros from ChatGPT and Claude, and it correctly identified them both as 100 percent AI-written – top marks for Pangram. He also explained some of the gifts, such as the phrase “from the moment when…”.

4/4 correct

grammatically

Next, Grammar. It’s been around for years as a spelling and grammar aid, and now also does AI and AI detection. You get three free AI checks per day, with paid plans starting at $12 per month that offer much higher limits (and a host of other features).

Once again, the results were reassuring because of how different my writing is from AI: both of my writing samples got a perfect score for being written without a machine. The text samples returned 0 percent AI text or AI text patterns.

The AI ​​samples I tested Grammarly on were from Claude and Gemini, and were marked as 68 percent and 66 percent written by AI, respectively. Grammarly caught on to the characteristics of AI, but wasn’t completely convinced.

4/4 correct

GPTZero

GPTZero’s stated goal is to “preserve what is human,” and that includes being able to detect AI-crafted phrases. On a free account, you can scan up to 10,000 words each month, while paid plans start at $23.99 per month, allowing you to scan many more words and access many more features.

At this stage, I felt confident that my writing was free of whatever these detectors were looking for, and sure enough, GPTZero marked both of my samples as written by a person: “We are very confident that this text is completely human,” they told me.

When encountered with writing samples from Gemini and ChatGPT, GPTZero correctly identified them as AI-generated text, and even flagged the most AI-like sentences (although, honestly, I had a hard time detecting patterns in these flagged phrases).

4/4 correct

notary

Our fourth AI screening tool is Scribbr, which also offers a variety of other editing and proofreading services. Their detection tool is completely free to use and you don’t even need to register an account to start checking text.

My own samples were approved as 100 percent written by a human. Although Scribbr has a disclaimer that AI detectors are not always completely reliable, it was certainly accurate when evaluating my own writing.

However, the samples I entered from ChatGPT and Claude were also marked as completely AI-free, which was not the case, and Scribbr seemed very confident in their conclusions. Whatever you’re doing to detect machine-created text, you need to improve your methods.

2/4 correct

Copies

The fifth and final verifier I tried was Copyleaks, which also has tools for detecting AI images and videos. I was able to scan four samples for free, although paid plans, starting at $16.99 a month, give you access to more detailed reports and additional features.

Like the other AI writing detectors on this list, Copyleaks gave my writing samples the thumbs up: Both were marked as 100 percent AI-free. With these tools, at least, I don’t seem to be accused of writing with AI anytime soon.

When I tried writing samples from Claude and Gemini, the former was marked as 0 percent AI written, and the latter was marked as 100 percent AI written, so I was half correct. Like Scribbr, there seems to be something wrong with the AI ​​detection calibration.

3/4 correct

the verdict

I wasn’t quite sure what to expect from these AI handwriting detectors, and that was one of the reasons I wanted to try them. Personally, I feel like I have a pretty good idea of ​​what AI writing looks like, but it’s hard to identify specific patterns and signals (well, except “it’s not x, it’s y”).

According to my tests, these AI detectors actually work, or at least most of them, most of the time. Text written by humans was identified as such in all cases in all of these tools, perhaps showing that it is easier to tell what has not been written by AI than what has.

AI detection wasn’t as foolproof, although it depends on which verifier you’re using. If you really want to detect AI writing, my advice would be to use two or three of these checkers together, along with other checks, and perhaps upgrade to levels where you get detailed reports on the detector’s reasoning.

Disclosure: Ziff Davis, the parent company of Popular Science, filed a lawsuit against OpenAI in April 2025, alleging that it infringed Ziff Davis’ copyrights by training and operating its AI systems.

2025 PopSci Best of what’s new

The 50 most important innovations of the year