Recognizing AI text: detect whether content was written by a computer

With the rise of ChatGPT and other AI writing tools, it is becoming increasingly important to be able to tell whether text was written by a human or by artificial intelligence. Whether you are a teacher, an employer or simply curious: in this article you'll learn how to recognize AI-generated text and which tools can help you do so.

Why is recognizing AI text important?

Detecting AI-written text has become relevant for several reasons. In education, teachers want to know whether students did their own work. Employers want to receive authentic cover letters. And news readers want to know whether articles were written by real journalists.

In addition, AI text can be misleading when it is presented as human work. Think of fake reviews, automated social media posts or plagiarized academic work. Being able to recognize AI content helps preserve authenticity and trust.

Characteristics of AI-written text

Before we look at detection tools, it is useful to know which characteristics AI text often shows. Note: these characteristics are no guarantee, but they can be clues.

A noticeably flat writing style

AI text is often correct but dull. The sentences are grammatically sound, but lack the personal touch that human writers often have. There is little variation in tone or emotion, and the text sometimes feels mechanical.

Repetitive structures and words

AI models tend to repeat certain words and sentence structures. If you notice that certain words or phrases come back remarkably often, this can be a signal. Typical AI words in Dutch include, for example, "echter" (however), "daarnaast" (in addition), "bovendien" (moreover) and "het is belangrijk om op te merken dat" (it is important to note that).

Lack of personal experiences

Human writers often refer to personal experiences, anecdotes or specific examples from their lives. AI text rarely contains these elements, because an AI has no real experiences to share.

Perfectly structured but superficial

AI text is often neatly organized with clear headings and lists, but the content can be superficial. The text covers topics broadly but rarely in depth, and can contain general information without unique insights.

No up-to-date information

AI models have a knowledge cutoff (the date up to which they were trained). If a text is about recent events but contains no current details, or if dates and facts remain vague, this can point to AI generation.

Online tools to detect AI text

Several tools are available that can help detect AI-written text. Here we discuss the most reliable options.

GPTZero

GPTZero is one of the best-known AI detection tools and is widely used in education. The tool analyzes text on two main characteristics: perplexity (how predictable the text is) and burstiness (variation in sentence length and complexity).

Here's how to use GPTZero:

  1. Go to gptzero.me
  2. Paste the text you want to check into the text field
  3. Click the button to analyze it
  4. View the results with percentages for the likelihood of AI

GPTZero offers a free version with limitations and paid plans for more extensive use.

Originality.ai

Originality.ai was developed for content marketers and publishers. The tool combines plagiarism detection with AI detection and is particularly accurate. It is a paid service, but it delivers professional results.

Copyleaks AI Detector

Copyleaks, originally a plagiarism detection tool, has now added AI detection as well. The tool supports multiple languages, including Dutch, and is available as a browser extension for easy use.

ZeroGPT

ZeroGPT is a free online tool that quickly gives an indication of whether text was written by AI. The interface is simple: paste your text and receive a score within seconds. Although it is not the most accurate option, it is handy for a quick first check.

Writer AI Content Detector

Writer offers a free AI detection tool that you can use without registering. You can check text of up to 1,500 words and receive a percentage indicating how much of the text was probably written by AI.

Limitations of AI detection tools

It is important to understand that no AI detection tool is 100% accurate. There are several limitations you need to take into account.

False positives

Sometimes human-written text is wrongly flagged as AI-generated. This happens especially with text that is formal or technical in nature, or with writers who have a very structured writing style.

False negatives

Conversely, AI text that has been edited or rewritten by a human often cannot be detected. Text generated with specific instructions to sound more human can also slip past detection.

Language differences

Most AI detection tools are trained on English-language text. For Dutch text, the results can be less reliable. If possible, choose a tool that explicitly supports Dutch.

Continuous improvements on both sides

Both AI writing tools and detection tools are continuously being improved. What works today may already be outdated tomorrow. So always treat detection results as an indication, not as conclusive proof.

Manual techniques to recognize AI text

Besides automated tools, you can also look out for certain things yourself.

Ask probing questions

If you can question the writer, ask probing questions about the content. Someone who wrote the text themselves can explain why certain choices were made. AI users without knowledge of the subject will struggle with this.

Check facts and sources

AI models can "hallucinate" and present incorrect information as fact. Check specific claims, dates, quotes and source references. If these are wrong or cannot be verified, this can point to AI generation.

Pay attention to context

Consider whether the text fits the supposed writer. Does the language level match? Is it consistent with this person's earlier work? Sudden changes in writing style or quality can be suspicious.

Analyze the metadata

With digital documents, you can sometimes view metadata that gives clues about how the document was created. However, this is not a reliable method, because metadata can easily be changed.

The future of AI detection

The battle between AI writing tools and detection tools will probably continue. As AI models get better at producing human-sounding text, detection tools will have to keep pace.

Some experts argue for digital watermarks in AI-generated content, so that it always remains recognizable. Others emphasize the importance of transparency: that users of AI tools are honest about when they have used AI.

Frequently asked questions

How reliable are AI detection tools?

AI detection tools are useful aids, but they are not infallible. Accuracy varies between 60% and 90%, depending on the tool and the type of text. Use detection results as supporting evidence, not as the only criterion for your assessment. Preferably combine multiple tools and methods for a more complete picture.

Can I check for free whether text was written by AI?

Yes, there are several free options. GPTZero offers a limited free version, ZeroGPT is completely free, and Writer has a free detector for texts of up to 1,500 words. For regular or professional use, paid tools are often more accurate and more extensive.

Do AI detection tools also work for Dutch text?

Most tools are optimized for English, but they also work for Dutch with mixed results. Copyleaks explicitly supports multiple languages. For the best results with Dutch text, you can try several tools and compare the results.

Is using AI for writing always wrong?

No, using AI as a writing aid is not wrong by definition. It depends on the context. As a tool for brainstorming, structuring or improving your own text, AI can be valuable. It becomes problematic when AI text is presented as entirely your own work in situations where originality is expected, such as academic assignments or job applications.