We Tested 5 AI Detectors on the Same Text (Results)

According to research by Digitaloft, 25% of e-commerce businesses use AI to help write product descriptions. At the same time, 93% of marketers still edit AI-generated text before publishing. That makes AI detection part of the workflow for many content teams. Before you begin editing, it’s important to distinguish between robotic style and human-written text. The market is currently full of AI detectors. Some tools flag almost every sentence, while others miss obvious AI patterns. 

Our team selected 5 popular detectors and tested them on the same text. We also monitored how each platform handled structure, wording, and editing traces. The results were more mixed than expected. Read on to find out which detectors remained lenient and which were really strict.

The We-Right Team is keeping this comparison simple and straightforward. The results are not affected by some links that may appear in this article. We used the same text to test each detector and our notes were made based on actual scans, not on promo claims.

What Text Did We Use For AI Detectors?

We picked an informational SEO article about “What Is Ahrefs Domain Rating” for this AI detector test. We selected this topic because SEO content falls in the middle ground between AI writing and manual editing. Many marketers leverage AI to generate drafts and then edit the text to improve the copy before publishing. This helps make this format useful for detector testing

The article contained 6 H2 headings, 3 H3 headings, 15 paragraphs, and was 720 words long. The text used here has been partially generated with the help of OpenAI ChatGPT 5.2. The other half of it was written and edited by a copywriter. Additionally, we made several manual edits to the text to create a more realistic publishing scenario.

How Did We Conduct the AI Detectors Testing?

Once we chose an article, we ran a Google search and identified the 5 most common AI detectors in the search results and on review pages. We looked at AI detectors that could be used by regular users, without paying for plans or waiting long periods to get started: 

• Copyleaks AI Detector;
• ZeroGPT;
• Undetectable.ai;
• Originality.ai;
• Turnitin Checker AI.

All 5 AI detector tools scanned the same article – the text length remained constant. Nothing was rewritten or edited between the scans.

We compared the following parameters during testing: AI scores, human scores, scanning speed, interface structure, and the language markers identified by each detector. The aim was not to locate one “perfect” detector. We wished to know just how different these systems were from each other when working on the same material.

Test #1: Copyleaks AI Detector

We initially used Copyleaks AI Detector for the text checking. The tool displays the scan in a split view, with the text on the left and flagged parts highlighted in purple. The score is on the right side – you may see a circle chart and a word count split between AI text and human text. 

Here are the results:

• AI probability score: 27%
• Human-written score: 73%
• Scanning speed: Fast, under 10 seconds

Copyleaks identified 197 words as AI-generated and 533 as human-written. It also discovered 17 AI phrases. The majority of the flags were found in short SEO-type lines, step-by-step instructions, and general terms such as “visit”, “enter”, and “display its DR score”. The tool appeared to be sensitive to clean structure and repeated phrases. 

Our team liked the clear layout, quick scan, and word-level highlighting. The primary problem is that Copyleaks sometimes has a tendency to “mistake” well-written English for AI-created content. The score is helpful, but it is still necessary to read the parts marked yourself.

Test #2: ZeroGPT AI Detector

We tested the same text using ZeroGPT. It displays the verdict first, then shows a score meter. It identified the text as “most likely human-written” in our scan, and generated some AI or GPT content. Under the score, ZeroGPT marks lines that are likely plagiarized in yellow to make it easy for you to identify the sections written by AI.

The results are as follows:

• AI probability score: 21.4%
• Human-written score: 78.6%
• Scanning speed: Fast, around 5–10 seconds

Several clear patterns were identified in the text by ZeroGPT. It highlighted sections “What Does Domain Rating Evaluate?” and “Ahrefs Domain Rating measures the overall authority of a website based on its backlinks.” It also marked phrases like “The Scale of Domain Rating”, “measured on a 0-100 scale”, “the scale is logarithmic”, and “anything above 70 suggests a strong backlink presence.” The tool was also sensitive to the phraseology used, such as “Follow these simple steps” and “Click Check Authority”. 

We liked the clean score view and the easy yellow highlights. The minus is the clutter on the pages from ads and promo bars. 

Test #3: Undetectable.ai AI Detector

Next, we checked the same text with Undetectable.ai. It provides you with a verdict at the top and a big AI score meter. Under this, the text is coloured – green means human, pink means AI.

The exact results are as follows:

• AI probability score: 75%
• Human-written score: 25%
• Scanning speed: Fast, around 5–10 seconds

Undetectable.ai detected more text as AI than the first two tools. It highlighted “What Is Ahrefs Domain Rating,” “a backlink authority metric,” “measures the strength,” and “points ranging from 0-100”. It was also the first time that terms such as “Quantity of Backlinks,” “high-quality connection,” “stronger impact,” “Backlink Distribution,” and “well-distributed profile” were flagged as AI text.

It probably marked these sections as it sounded like a stock SEO article. The phrasing is regular, term-oriented, and is structured around definition-style lines, which may appear machine-written to more rigorous detectors. 

Our team liked the color split, and the score was clear. The great advantage is being able to quickly scan each marked line. The minus is the strict scoring – it considers a lot of the key terms used in SEO to be AI-related. 

Test #4: Originality.ai AI Detector

We used Originality.ai to check the same text. The tool displays the pasted text on the left, and the score panel on the right. It indicates risk areas in pink, red, and yellow. The result card states that the scan was very strict because it said: “100% Confident That’s AI“. 

Read also our article: How to Fix Low SEO Scores in Surfer SEO (Real Examples)

These are the outcomes: 

• AI probability score: 100%
• Human-written score: 0%
• Scanning speed: Fast, around 10 seconds

Originality.ai identified entire paragraphs as AI text, rather than just isolated phrases. It highlighted the whole sections under the H2 “What Is Ahrefs Domain Rating” and under the H2 “What Does Domain Rating Evaluate?”. We think the tool probably regarded the text as “generated” because the language is highly uniform. The sentences are well-optimized for SEO, very succinct, and use repeated structures to make the content easier to read, which is why Originality.ai deemed the copy too formal and structured. 

We really liked the visual marks, and the scorecard was clear. The downside is the bad result – 0% human text despite multiple edits.

Test #5: Turnitin Checker AI

Turnitin Checker AI was used to check the same text. The tool displays an analysis report in a pop-up window rather than marking the lines. It provides 2 score cards – a brief summary and pattern tags. The design is simple, and the primary result is received quickly.

Let’s take a look at the results of our text:

• AI probability score: 80%
• Human-written score: 60% originality score
• Scanning speed: Fast, around 10 seconds

Turnitin Checker AI did not display the exact highlighted words in the text. Instead – it is called the patterns. The tool identified the following: “uniform sentence length,” “low perplexity,” “repetitive transitions” and “informative but generic content.” In its summary, it also highlighted a “well-structured” article on Ahrefs Domain Rating”. The copy was said to have a generic voice.

We found the short report and pattern labels good. They can assist you to detect the problem quickly. The downside is that there’s no highlighting of sentences so you have to guess which lines should be humanized to improve the result!

Comparison of Tested AI Detectors 

Every AI detector we tested had its own level of logic. There were some tools that were soft, and detected AI in just a section of the text. Others considered the same article to be largely AI-generated. So, the selection of one tool will depend on the need. We’ve created a quick comparison table below.

Main Takeaways From the Test

The biggest surprise from this experiment was how inconsistent the results were. The exact same article received AI scores ranging from 21% to 100% depending on the detector.

We also noticed that most tools reacted strongly to:

• highly structured SEO wording;
• repetitive sentence patterns;
• definition-style paragraphs;
• and predictable transitions between sections.

At the same time, manually edited sections still triggered AI flags in several detectors. This suggests that many tools focus more on writing patterns and formatting consistency than on whether a text was actually written by a human.

Another important takeaway is that no detector behaved fully consistently across the entire article. Some tools flagged only isolated phrases, while others treated almost the entire text as AI-generated despite heavy editing.

For content teams, this highlights an important point: AI detectors should be treated as a reference tool, not as a final verdict on content quality or originality.

Our Final Thoughts

The test revealed one thing – AI detectors don’t assess text in the same manner. Low scores on one tool may be high scores on another. These tools are only a guideline, not a final decision. Always check flagged parts on your own.

FAQ