We’re living in the middle of an AI wave. From chatbots to automated content, AI quickly transforms how we communicate and create.
But as AI tools become more common, the line between human-written and AI-generated content is starting to blur.
So, how can you tell if a piece of text was crafted by a person or generated by a machine?
Let’s dive into AI text detectors. We conducted tests using fully human-written content, entirely AI-generated content, and AI-generated content with human revisions. Let’s begin!
- What are AI detectors?
- How do AI detectors work?
- Top AI detectors at a glance
- AI detectors tested
- How does Google View AI-Generated Content?
- Can AI Content Really Be Detected?
- Should you or should you not use AI content detectors?
What are AI detectors?
AI detectors are tools that figure out whether a piece of text was written by a human or generated by artificial intelligence, like ChatGPT, Claude or Gemini.
They do this by looking for certain clues in the writing—things like repetitive structures, overly complex phrases, or patterns that are common in AI-written content.
These detectors are used in places like schools, businesses, and content platforms to ensure authenticity, helping people verify that the writing they’re seeing or submitting is genuinely human-created when needed.
How do AI detectors work?
AI detectors work by scanning text for signs that it was written by artificial intelligence rather than a human.
They rely on machine learning algorithms that have been trained on tons of data, both human-written and AI-generated. Here’s how they typically operate:
- Pattern Recognition: AI detectors look for repetitive phrases or overly structured sentences, common in AI writing.
- Complexity Analysis: AI text can sometimes overuse complex language or lack natural flow, which detectors flag.
- Training Models: The detectors are trained on methods and structures used by AI models like ChatGPT or the alikes.
In short, they’re like trained readers that spot the subtle differences between human and AI writing. That being said, no two AI detectors are programmed exactly the same, so you need to keep that in mind.
Top AI detectors at a glance
Here is a comparison table of the four tools we tested.
| Tool Name | Accuracy (# of tests passed) | Features | Pricing |
| CopyLeaks | 2/3 | Plagiarism detection, omit options, scan destination options. Multiple detection modes | Starts at $7.99/mo billed annually |
| GPTZero | 3/3 | Scan options, batch uploads, statistics | Starts at $10 per month billed annually |
| Scribbr | 2/3 | Multi-lingual, percentage based results | Free forever |
AI detectors tested
CopyLeaks
CopyLeaks is a flexible tool for handling more than just AI detection. It also checks for plagiarism, scans code with CodeLeaks, improves text through a writing assistant, and offers governance tools to monitor AI-generated content.
Plus, it includes an AI grader to assess the quality of writing. You can easily add CopyLeaks as a Chrome extension, and the convenience doesn’t stop there—you can even use it without logging in for quick, hassle-free checks!
Key features
CopyLeaks gives you the following features when detecting AI content with it:
- Plagiarism Detection: This toggle enables detection of plagiarism within the document, highlighting copied or paraphrased content.
- Omit Settings: Users can choose to omit certain aspects, such as references, quotes, citations, and the table of contents, during the detection process.
- Scan Destination: The user can choose where the scan results are saved, such as “My Scans” in the image.
- Multiple Detection Modes: Users have control over enabling or disabling both AI and plagiarism detection modes.
Putting CopyLeaks to the test
When you log into CopyLeaks, you can paste your text into the box. You’ve got a couple of editing options like font sizes, alignment and font types. There’s no word limit but free users get only five credits after which they need to pay.
We tested the tool with 100% human-written content, 100% AI generated content and AI generated with human edits.
Here are the results we got with CopyLeaks:
- 100% human-written
CopyLeaks accurately identified the text as 100% human with 0% AI content. This result shows that the tool correctly flagged the submission as authentic.
- 100% AI-generated
Prompt used: Generate a piece on “How does bad customer service affect your business?”
The target audience is: Customer Support Teams
Use the following keywords: bad customer service, customer experience, and customer support.
The AI detector again accurately flags the entire content as AI generated by giving a 100% AI generated rating.
- Partially AI generated with human edits
Despite human intervention, the detector still identified AI-generated elements throughout the text.
The purple highlights indicate sections that were recognized as AI-generated, even though they were edited, showing the CopyLeak’s limitations in accurately differentiating between AI and human-refined text
Pricing
- Plagiarism detector: $8.99/mo, billed annually
- AI detector: $7.99/mo, billed annually
- AI + Plagiarism detector: $13.99/mo, billed annually
- Enterprise: Contact sales
GPTZero
GPTZero is another popular AI content detection tool. Free users get a total of 10k credits and at once you can upload up to 5000 characters.
GPTZero is designed to identify content generated by models like GPT-3, GPT-4, Gemini, Claude, and Llama. On their website, GPTZero mentions that they use a seven-layer detection model to identify content created by AI.
Users can directly paste text when they open the website, however, the user interface feels better when you log in. Once logged in, you can see options to upload files or paste text.
Key features
Here are some features GPTZero offers:
- Scan Options: Includes Basic and Deep AI detection, Plagiarism Scan, Writing Feedback, and Multilingual Detection (Spanish and French).
- Batch Upload: Allows uploading multiple documents for analysis at once, ideal for larger projects.
- Statistics: View scanning analytics and track platform usage, including the number and types of scans.
- Create a Team: Allows users to collaborate and share work, ideal for group projects or content verification.
Putting GPTZero to the test
For this test, we have pasted three versions of our topic: How Does Bad Customer Service Affect Your Business? There are no formatting options or other editing options here.
Here are the results given by GPTZero:
100% human-written
This AI detector has given the results as 97% human written content. It tells us that only 3% text is AI generated. While this is not true, the results are close to accurate.
GPTZero also highlights the sentences that maybe AI generated and gives you the number at the end. In this case, the number of sentences is zero.
100% AI-generated
When I pasted the ChatGPT generated text, the results were 100% AI generated. This is a 100% accurate result.
For a second verification, I put text generated by Claude and the results were the same. This goes out to show that GPTZero has so far done a better job than the others.
In terms of the number of sentences, the results were again accurate.
Partially AI generated with human edits
Next up, I put AI generated and human edited text into GPTZero. To my surprise, the result was again an accurate one. It is 76% for AI content and 24% for human content.
Now, this was interesting. While, GPTZero claimed 24% content to be human-written, it still flagged all sentences as AI.
Pricing
- Essential: $10 per month billed annually
- Premium: $16 per month billed annually
- Professional: $23 per month billed annually
Scribbr
Scribbr’s AI Content Detector is a handy tool that helps users identify AI-generated content from sources like ChatGPT, Gemini, and CoPilot.
What makes it stand out is that it offers unlimited free AI checks—so you can run as many scans as you need, all without any cost. Each submission can include up to 1,200 words. It provides results in the form of percentages for AI-generated, AI-generated and AI-refined, Human-written and AI-refined and Human-written.
Key features
There’s not much to Scribbr features, but here are some:
- Multi-language Support: Detects AI-generated content in multiple languages, including English, Spanish, German, and French.
- Percentage-Based Results: Provides a percentage breakdown indicating the likelihood of the text being AI-generated or human-written.
- Simple Interface: Clean, user-friendly design with a clear “Analyze text” button to initiate the scan.
Putting Scribbr to the test
You need to directly paste the text into Scribbr and run a scan. Unlike CopyLeaks, there are no editing options.
Here’s how Scribbr performed in the content detection test:
100% human-written
Scribbr gives an accurate result by showing a 0% AI-generated text, which is great.
100% AI-generated
For fully AI-generated text, the results are quite accurate. It highlights all AI-generated sentences in yellow.
Partially AI generated with human edits
Scribbr struggles when dealing with text that has been partially AI-generated and human-edited. For such content, it inaccurately reports 0% AI detection, indicating that it performs best with either fully AI-generated or fully human-written text. When the text is a mix, the detection tool tends to fall short.
This highlights the limitations of Scribbr’s results, particularly when it comes to evaluating AI-refined or human-refined text, making those percentage-based readings less reliable in these cases.
Pricing
Free forever
How does Google View AI-Generated Content?
Google’s stance on AI-generated content focuses on quality. Whether the content is human or AI-created, it must:
- Be useful, original, and meet user needs.
- Follow a “people-first” approach, providing real value instead of manipulating search rankings.
Google doesn’t have an issue with AI tools; the problem arises when AI is used for spammy or low-quality content, which can hurt rankings.
Content that meets Google’s E-E-A-T standards (Expertise, Experience, Authoritativeness, Trustworthiness) will be rewarded, regardless of its origin. Proper use of AI can enhance content creation without breaking Google’s guidelines.
Screenshot of a blog post from Google about AI generated content
Can AI Content Really Be Detected?
The question of whether AI content can truly be detected is still up for debate.
Companies like OpenAI, which created AI systems like ChatGPT, have admitted the difficulty in reliably detecting AI-generated content. In fact, OpenAI recently discontinued its AI writing detector due to its low accuracy.
Tools meant to identify AI content often have high error rates, sometimes even flagging human-written content as AI-generated, especially when “trigger words” are used. We’ve seen the mess AI detectors can create from the tests above.
This brings into question the overall reliability of AI detectors, and disclaimers frequently highlight these limitations.
Should you or should you not use AI content detectors?
When deciding whether to use AI content detectors, it’s important to understand their limitations. While they can be helpful in flagging AI-generated content, they’re far from perfect. AI detectors can misidentify human-written content as AI-generated, and mixed AI-human content often confuses them.
Given their inaccuracy, they should be used with caution rather than relied upon entirely.
Human oversight remains crucial, as AI detectors are simply tools to aid judgment, not definitive measures of content authenticity or quality. Use them as a guideline, not the final verdict.