11 TOOLS FOR DETECTING AI-GENERATED CONTENT

Large language models of artificial intelligence, such as ChatGPT, can be effective assistants for professionals across many fields. Copywriters, marketers, and business owners use neural networks for text writing. Currently, AI-generated content is often published on social media and websites. With AI, you can even create a book or develop a complete marketing strategy. But will such materials be of high quality?

Article Outline

• How to Identify Machine-Generated Content?
• Google’s Approach to AI-Written Texts
• Why Can AI-Generated Content Be Problematic?
• Tools for Checking Texts for Neural Network Usage
  • ZeroGPT
  • GPTZero
  • AI Content Detector
  • GPTRadar
  • Hive Moderation
  • Illuminarty
  • Crossplag
  • AI Text Classifier
  • Writefull GPT Detector
  • AI Plagiarism Checker
  • GPTKit
• Conclusions

How to Identify Machine-Generated Content?

Undoubtedly, artificial intelligence is a valuable tool for saving time and generating ideas. With its help, you can continuously create informative and engaging content for your target audience. However, today, there are many ways to check texts for the use of ChatGPT and other neural networks.

Machine-generated texts may appear artificial, fail to resonate with readers, or even negatively impact webpage rankings in search results. To avoid this, it’s essential to understand the nuances of such content and learn to refine it manually before publication. The human touch is crucial here, as only a human author can express personalized opinions within each specific context, highlight key points, and provide examples from personal business or life.

The ability to give clear commands and prompts is necessary, as well as the constant refinement of texts to maintain the brand’s desired tone of voice, avoid plagiarism or copyright issues, and ensure that each material aligns with the company’s business goals or the author’s personal objectives.

There are numerous tools available to detect AI-generated content, and we’ll share a collection of such services later in this article. However, it’s possible to identify machine-generated content even without specialized scanning tools. AI content’s artificial nature can be recognized by several typical signs.

• Language models struggle with complex topics and biographical material, often producing lower-quality texts filled with fabricated facts, fictitious quotes, and nonsensical passages.
• Usually, automatically generated content appears formulaic. Common indicators include frequent repetition of words and phrases and similar opening phrases in list points.
• AI generates texts of a general nature that lack audience targeting and contextual awareness. Even if you provide a very detailed prompt for the chatbot, certain nuances may still be missed, and the model itself cannot independently improve the results.
• A lack of creativity and emotion makes machine-generated texts dull, while inaccuracies in statistical data and weak arguments diminish their credibility. In some cases, such texts may work for a personal social media blog, but readers of an expert online publication may raise questions.
• Few lines per paragraph and short sentences are typical features of AI-generated content. You’ll also notice fewer punctuation marks in machine-written texts than in human-authored ones. Humans tend to use question marks, colons, semicolons, and exclamation marks more frequently.

Even if you manage to obtain an emotionally nuanced text without factual errors, the writing style will still differ from yours. So, if you used to write for your blog yourself and suddenly decide to delegate this task to ChatGPT, loyal readers might notice the change. A professional copywriter, familiar with your previous texts, is more likely to replicate the required style.

This can be compared to a customer support service: if customers regularly communicate by phone or chat with real managers and are later only connected with chatbots and voice bots, audience loyalty may sharply decline.

Google’s Approach to AI-Written Texts

It is known that Google can identify AI-generated content. However, such high-quality content will still be ranked. Removal only occurs if the content violates Google’s spam policy. Additionally, sanctions may apply to websites where articles on pages do not align with the principles of E-E-A-T: 

• Expertise
• Experience
• Authoritativeness
• Trustworthiness

Links to authoritative sources and author photos with background information serve as proof of expertise and authority. It is essential to highlight the experience, credentials, and achievements of content creators. Trustworthiness is confirmed by reliable sources and by including concrete information without excessive “fluff”.

AI-generated content should be readable, understandable, and valuable to readers. Content that does not meet the search engine’s requirements could be generated by AI, a copywriter, or a business owner, so a thorough review by an expert editor is crucial.

With AI’s evolution, identifying AI-created content is becoming increasingly difficult. The demand for prompt engineers and editors capable of enhancing AI-generated text and “humanizing” it is growing.

AI-generated content is not always appropriate. For instance, when writing news, it’s often better to avoid using language models entirely. Otherwise, there is a risk to the brand’s reputation due to potentially false information. However, using AI for summarization can save time and resources. ChatGPT, for example, can condense large volumes of text into brief summaries, which is highly useful for research, planning, and strategy development.

AI can also process large data volumes and convert them into tables, charts, and diagrams. AI tools can assist in growing organic traffic. For instance, GPT can help find relevant keywords and distribute them evenly throughout the text. Google does not prohibit this, provided that SEO requirements (keyword density, article structure, subheadings, etc.) are met.

Certain fields, such as medicine and finance, are areas where AI can negatively impact E-E-A-T principles and undermine the trust of potential clients. The risk of inaccuracies in such articles can harm readers, affecting their safety or health. AI-generated medical advice is considered unethical, as are AI-generated recommendations on addressing psychological issues.

This also applies to political content, where chatbots may produce biased material. A human author considers nuances of the current political situation, the mood of potential readers, their views, and expectations. For an independent source read nationwide, any advocacy or criticism could lead to significant negative reactions and misunderstandings. An author knowledgeable in the subject can avoid these issues and create more appropriate and objective articles or news posts.

According to Google’s updated policy, publishing AI-generated texts on web resources is not prohibited. Unfortunately, AI-written content often lacks sufficient quality, which can lower rankings even for well-optimized and promoted websites.

Why Can AI-Generated Content Be Problematic?

Research from Cornell University indicates that 66% of surveyed internet users cannot identify fake news generated by artificial intelligence. This is partly because copywriters also often use keyword repetition, which can make texts seem unnatural — particularly when more focus is placed on SEO (search engine optimization) rather than LSI (latent semantic indexing).

AI-generated content is considered problematic because it resides in the “gray area” of search engines’ focus. Without proper editing before publication, using such content puts sites at risk of sanctions. The primary issue is not a lack of authorship or low uniqueness but the potential disregard for the E-E-A-T formula.

Legal considerations must also be taken into account. Since Ukraine is aiming to join the EU, it is essential to follow European legislation at this stage, specifically the GDPR (General Data Protection Regulation), which is the standard for protecting personal data, including AI-generated materials.

Approaches to copyright protection and AI-created content vary by region. For instance, in Italy, ChatGPT is officially banned due to GDPR violations regarding the confidentiality of conversations in its paid version.

China does not ban AI-generated materials but thoroughly reviews each service before granting permission for use. In the United States, writers and other creatives are urging Congress to prohibit copyright protection for AI-generated art. Conversely, the United Kingdom protects AI-created work under copyright law.

According to the updated Law of Ukraine “On Copyright and Related Rights”, AI-created works are not considered original but are still protected.

“…at the World Economic Forum in Geneva (WEF), the risks of job loss for 26 million people due to advancing technologies were highlighted. For this study, the WEF surveyed over 800 corporations from different countries, representing a workforce of 11.3 million” — “The Application of Artificial Intelligence in Business Processes: A Legal Aspect”, by Yuliya Ryabets, Business Support Lawyer at Juscutum.

One primary concern for creators of original content and art is that artificial intelligence technologies learn from human-made works. They can partially replicate others’ images (as with Midjourney) or mimic speech styles and borrow and modify text fragments (common in ChatGPT). AI may produce results that contain copyrighted material, posing a high risk of intellectual property infringement.

Another issue is the potential misinformation of internet users due to the inaccuracy of AI-generated materials. This could involve images created much faster than with photo editors, as well as text — especially news or biographical content.

Ideally, the use of artificial intelligence should be noted upon publication. For informational or entertainment texts, AI-generated content should be reviewed by a knowledgeable human editor who can correct factual errors and tailor the content as needed (for example, by simplifying terminology or making the article more formal and professional).

The European Parliament is actively working on regulations for AI usage. European Parliament members believe that responsibility for the misuse of AI programs should fall on developers like OpenAI, rather than on smaller companies that are effectively clients of these corporations.

“According to proposals from the European Parliament, developers of generative AI models like ChatGPT will be required to label AI-generated content and publish summaries of copyrighted data used for training purposes so that creators can be compensated for the use of their work”. “The European Parliament Has Prepared Strict Rules for AI Usage“, by Anita Prasad

Language models can be quite “cunning”: they are known to distort factual information, figures, and biographical details of famous individuals. There have even been cases of fabricated links to fake sources. Not every professional can work effectively with AI, as it requires specific skills and knowledge in various fields. If you need to assess the readiness of AI-generated text for publication, consider using specialized services.

Tools for Checking Texts for AI Usage

There are numerous tools available to check texts for AI-generated content. However, it’s essential to understand that not all of them are effective. Similar to plagiarism detection services, AI content detectors vary in terms of accuracy.

To select the best methods for assessing AI-generated content, we created a text using ChatGPT and scanned it with each tool.

The prompt was “write a post about the benefits of native advertising”. Commands given in English usually yield better results, as AI produces text with minimal semantic and spelling errors in this case.

ZeroGPT

The website highlights features such as batch file processing in the background, bulk file uploads, a high-precision AI detection model, automatically generated PDF reports, and multilingual plagiarism detection.

The tool successfully identified 95.98% AI-generated content. For some reason, only the final list items were marked as unique, but this didn’t significantly impact the overall score.

GPTZero

“More than an AI detector” — this is how the service, created for teachers, students, and writers, positions itself. Users can choose from several products: Chrome Origin extension, human writing reports, handwritten text and Google Doc reports, API and Canvas integration, as well as a Microsoft Word extension. The description of the technologies used for text analysis is impressive, including analysis of AI writing patterns, online material searches, and a sentence classification model that identifies mixed text.

After a simple and very quick scan, we received this result: “This text was likely written by AI. There is a 98% probability that this text was fully AI-written. Sentences likely written by AI are highlighted. 30/30 sentences were likely generated by AI”.

AI Content Detector

The tool supports analysis of different text formats, not just lengthy prose. Text may be highlighted in varying shades depending on the ratio of fully AI-written sentences to fragments created by a human.

The service offers a Chrome extension. The developers claim to have tested over 20,000 human-written documents, with a false positive rate of 0.2%, the lowest among competitors. Other products are available as well, such as code verification, grammar checks, and more.

The text was highlighted in red, but the lack of specific scoring and commentary was somewhat disappointing.

GPTRadar

The algorithm uses token probability and analyzes each sentence separately, displaying results on a chart. However, this didn’t allow the scanner to accurately identify artificial content. The result was: “Likely human-created. 81% accuracy”. The text scored 6 on the perplexity scale.

Hive Moderation

This tool is designed for checking both text and images. A free demo shows which text segment is most likely AI-generated. The Hive AI Detector Chrome extension is also available. Almost the entire text was highlighted as artificially generated, indicating the tool’s accuracy.

Illuminarty

NLP algorithms help detect texts generated by neural networks, while computer vision identifies digital images created with AI. According to the AI analysis, the likelihood of AI content in the examined text is 1.5%, which is far from accurate.

Crossplag

To use this service, registration on the website is required, with the option to log in via Google or LinkedIn. We used the free version for our analysis.

“This text is mainly AI-written”, — the detector’s response was satisfactory. Along with the score, a visual AI Content Index scale is provided, offering a convenient way to see the “artificial” level. The service also includes a user-friendly dashboard and online consultant.

AI Text Classifier

A completely free but ultimately ineffective online tool. According to the developers, it helps identify how much of the text is AI-written versus human-written. However, only 3% of our content was detected as AI-generated.

Writefull GPT Detector

This tool is free but has a daily usage limit. The service can be used for writing and editing academic texts, creating summaries, rephrasing, and more. For scanning texts for AI content, we do not recommend Writefull GPT Detector, as it identified our text as human-written, with only 1% AI-like content.

AI Plagiarism Checker

This service combines plagiarism detection and ChatGPT’s AI detector. The AI checker assesses creativity and predictability to identify AI content.

Potential AI content was found in 53.7% of the analyzed text. Better than 0%, but not perfect.

GPTKit

To bypass text length restrictions, registration on the website is required. However, only the first 2,048 characters per request are free. The checker uses six different methods to detect AI content and generates authenticity reports.

67% of the native advertising text was identified as real (not AI-generated), while 33% was flagged as fake.

Frequently Asked Questions

Conclusions

Out of the 11 tools we analyzed for detecting AI-generated content, we can genuinely recommend only four: ZeroGPT, GPTZero, Hive Moderation, and Crossplag. These services effectively distinguish natural, human-authored content from AI-generated material. Other checkers should be used cautiously. There are also many paid detectors available that offer broader functionality.

If you lack the time or inspiration to create your own content, using AI is an option, but don’t rely on it entirely. Make sure to thoroughly review each text, adding your own thoughts, ideas, and quotes. You could also delegate writing to copywriters or hire a prompt engineer and editor.

Remember that in certain fields, particularly medicine, political news, and financial advice, using artificial intelligence is inappropriate and even risky. In other cases, publishing AI-created texts on your website is acceptable, provided they comply with E-E-A-T principles.

However, regardless of your business area or content marketing goals, large language models can be beneficial for other tasks, such as creating content plans, developing marketing strategies, or researching and clustering keywords for site promotion.