We recently launched our new AI detector. For almost any webpage, you can analyze the likelihood that AI was used to create it, highlight AI-generated sections of text, and even identify the specific AI models used to create the text.
There are tons of AI detectors available. And, while ours is the only one that can also show you any webpage’s backlinks and estimated search traffic, how the page content has changed over time, and how it performs relative to other pages…
…we still wanted to know how Ahrefs’ AI detector compares to other popular detectors. So we tested it.
Interestingly, false positives were not a big issue. Only 2/24 tests run on human-written content incorrectly flagged the text sample as AI-generated. All the AI detectors struggled the most with the hybrid human/AI content (for reasons explained below).
In the table below, you can see the actual AI content of each test article, followed by each tool’s analysis:
Actual AI% | Ahrefs | Copyleaks | GPTZero | Originality.ai | Scribbr | ZeroGPT | Grammarly | Writer |
---|---|---|---|---|---|---|---|---|
100% | 100% | 100% | 92% | 100% | 94% | 99.62% | 59% | 24% |
100% | 100% | 100% | 100% | 100% | 100% | 98.99% | 57% | 23% |
100% | 86% | 83.8% | 52% | 64% | 0% | 67.76% | 7% | 15% |
0% | 2% | 0% | 0% | 0% | 0% | 0% | 0% | 2% |
0% | 6% | 0% | 0% | 0% | 0% | 2.76% | 0% | 2% |
0% | 0% | 0% | 4% | 0% | 0% | 91.69% | 0% | 27% |
50% | 13% | 46.4% | 42% | 46% | 0% | 39.06% | 0% | 0% |
50% | 32% | 0% | 28% | 4% | 0% | 29.4% | 0% | 5% |
50% | 70% | 100% | 86% | 4% | 83% | 83.21% | 0% | 6% |
Score | 13 | 13 | 12 | 12 | 10 | 9 | 6 | 4 |
Ahrefs’ AI Detector
Score: 13/18
URL: https://ahrefs.com/writing-tools/ai-content-detector
Ahrefs AI Detector showed good accuracy across the board, with no false positives for human-written content, and good detection for AI content. It also managed to detect the specific model types used for the AI content: GPT-4o and Meta’s Llama.
By my criteria, it failed one out of nine tests, struggling to identify AI content that was blended with human writing. This is a known limitation of all AI detection tools (more on that below), and was true of all detection models.
Ahrefs AI detector is based on our own proprietary detection model, trained on a huge amount of web content. It’s unique among AI detectors because you can use it in conjunction with tons of other Ahrefs data points to see how content actually performs.
You can use Ahrefs’ AI detector to see:
- Which AI models are the best for creating high-quality content.
- How often your competitors publish AI content, and which models they use.
- How much AI content is present in a particular SERP, and how much effort you might need to invest to rank.
- How organic performance metrics correlate with different levels of AI use, like search traffic, keyword rankings, and backlinks.
- Whether AI content use correlates with traffic drops on particular pages or in particular subfolders.
How to use Ahrefs’ AI detector
Start by heading to Site Explorer and pasting the URL you want to analyse. From there:
- Click the Page inspect report tab in the left sidebar.
- Choose the AI Detector tab.
- See your AI content level report in the right sidebar. Text that our model has detected as likely AI-generated will be color-coded according to the colors in the pie chart
In this example, our AI detector has found a section of my blog post that I used ChatGPT’s GPT-4o model to generate:
From there, you can also see how the page content has changed over time, how many backlinks it has earned, how many keywords it ranks for, how much estimated organic traffic it receives…
You can also test out the model via our free AI content detector page. We’re also adding bulk AI content detection to the Top pages report in Site Explorer (coming soon).
Copyleaks
Score: 13/18
URL: https://copyleaks.com/ai-content-detector
Copyleaks matched the top score, showing solid detection ability across both extremes of AI content. It proved especially effective at catching obvious AI writing, though it occasionally faltered in mixed or borderline passages.
GPTZero
Score: 12/18
URL: https://gptzero.me
GPTZero offered reliable results overall, with a clear strength in catching high-percentage AI content. However, it sometimes hesitated in assigning confident AI probabilities to mid-range or hybrid examples, slightly affecting its total accuracy.
Originality.ai
Score: 12/18
URL: https://originality.ai/ai-checker
Originality.ai performed well in most cases, accurately flagging AI-heavy text but showing a tendency to overestimate human authorship when faced with subtle or well-edited AI-generated material.
Scribbr
Score: 10/18
URL: https://www.scribbr.com/ai-detector/
Scribbr landed in the middle of the pack, handling clear-cut AI content reasonably well but displaying a drop in performance on more nuanced pieces, where its predictions tended to be inconsistent or overly cautious.
ZeroGPT
Score: 9/18
URL: https://zerogpt.com
ZeroGPT’s performance was uneven—it occasionally nailed high-AI content but frequently misclassified partial-AI and low-AI samples. The tool’s sensitivity seemed skewed toward extremes, resulting in a less balanced profile.
Grammarly
Score: 6/18
URL: https://www.grammarly.com/ai-detector
Grammarly’s free AI detector struggled with accurate AI detection, offering low-confidence or inaccurate predictions in many cases. It often failed to recognize clear signs of AI authorship and was unreliable on mixed or borderline content.
Writer
Score: 4/18
URL: https://writer.com/ai-content-detector/
Writer’s free AI detector scored the lowest, frequently misidentifying or entirely missing AI-generated material. It lacked precision across the board and provided little useful signal even when dealing with content that was 100% AI-written.
Further reading
It’s clear that AI content isn’t going away, so it’s a good idea to use a tool like Ahrefs’ AI detector to understand how AI content impacts your website performance. To get started, head to Site Explorer.
Similar Posts
Accelerate Sustainability Targets with VMware
Hosting providers are becoming more competitive in their electricity consumption. This is measured by the amount of data centre power consumption versus how much of the total Kwh is being used by the actual computing power of the data centre, known as the PUE (Power Usage Effectiveness). At a recent customer supplier day I learnt…
My Exact 7-Step Framework for Brand SEO (With Templates)
Branding wasn’t something SEOs traditionally thought much about. The real wins were in non-branded keywords, where the traffic and conversions lived. However, that changed when Google and OpenAI turned most of these queries into zero-click searches. For the remaining queries, search platforms directly reward authoritative and popular brands, so branding can no longer be ignored…
Best WordPress Maintenance Mode Plugin in 2024
Your WordPress website needs regular updates and maintenance. And sometimes, when you’re fixing your site, it results in downtime. In situations like this, you can switch on maintenance mode, which displays a temporary page to your site visitors, letting them know that it’s being worked upon and is not broken.
Scattered Spider: Understanding Help Desk Scams and How to Defend Your Organization
In the wake of high-profile attacks on UK retailers Marks & Spencer and Co-op, Scattered Spider has been all over the media, with coverage spilling over into the mainstream news due to the severity of the disruption caused — currently looking like hundreds of millions in lost profits for M&S alone. This coverage is extremely…
Deepfake Defense in the Age of AI
May 13, 2025The Hacker NewsAI Security / Zero Trust The cybersecurity landscape has been dramatically reshaped by the advent of generative AI. Attackers now leverage large language models (LLMs) to impersonate trusted individuals and automate these social engineering tactics at scale. Let’s review the status of these rising attacks, what’s fueling them, and how to…
Understanding the Difference Between Cache Keys and Cache Tags in Drupal
I recently got this this question via the feedback form on the bottom of the Cache API Overview tutorial: “What is the difference between cache keys and cache tags? Why do I need a tag for node:5 when I can just invalidate it using a partial key like [‘node’, 5]?” After digging into the topic…