Let's be honest: We can't (and don't want to) live without ChatGPT anymore. In our agency's daily routine, AI has greatly simplified processes, given us inspiration, and answered complex questions. But when we navigate the web and read blog posts, social media posts, or landing pages, we notice more and more often: No human wrote this for sure. Why? Because ChatGPT-generated content follows certain patterns that regular users of the AI immediately recognize. Complete presentations, emails, tickets – even proposals – are now generated with the tool. As helpful as that is: Sometimes it's annoying to only read computer-generated content. When I research a specific topic and am looking for expert advice, I don't want to constantly encounter untrustworthy tips and unchecked half-truths. Ultimately, it has become much more time-consuming to identify whether the advice xy really involves real experiences or whether the lying machinery ChatGPT was at work here again. Because, and we've all probably understood this by now, ChatGPT tends to hallucinate. For some reason, the tool has a big problem not having information and admitting when it doesn't have the faintest idea which answer is the right one to our question.

Exposing ChatGPT Content – Typical Phrases

Text Structure

ChatGPT-generated texts can sometimes be spotted at first glance: They often follow similar patterns. The AI loves lists. That's why it particularly often creates lists where the technical term/list item is bolded, then followed by a colon and then an explanation appears. Also typical: Posts with many (sometimes too many) subheadings, under which only a three-liner of text follows. Headings are normally incorporated in content creation to separate topics from each other or to structure the article neatly. ChatGPT takes this approach to the extreme and ensures that texts are flooded with subheadings and broken down into small pieces, even though the differentiation is often not necessary. And if all that doesn't stand out: Pay attention to how the paragraphs are formed. The AI uses breaks less often and instead integrates complete paragraphs.

In the Text

Content-wise, there are also clear indications of whether it's pure AI text. Here are some examples that strongly suggest ChatGPT:

  • Use of quotation marks only at the top
  • Phrases like "next level"
  • Formulations like "comprehensive guide"
  • Terms like "extensive guide"
  • Frequent sentence beginnings with "This/These"
  • Constructions like "not only, but also"
  • Use of "Konversion" instead of "Conversion"
  • General expressions like "in today's world/landscape/in today's age" etc.
  • "Let's" and "Let's dive in"
  • indirect address
  • Formulations in the subjunctive

...and so on!

Of course, such formulations occur from time to time: After all, the AI was trained based on our way of expression. However, if I notice when reading an article that these phrases accumulate, I immediately leave the page. Then I can just ask ChatGPT directly – and at least know what was prompted.

How to do it better?

Without mincing words: "Shit in, shit out". The worse or less specific the prompt is, the worse the result. When we write AI texts, we therefore make sure that we provide as specific information as possible – sometimes also in the form of example or reference texts. We make specific statements about the address, target group, the manner of expression (e.g. formal or casual) and bring in our own ideas. "Write a blog post about xy" usually goes wrong. It's better if ChatGPT first makes a suggestion for the structure and you then work on the post together with the AI piece by piece and paragraph by paragraph. Insert valid sources, provide texts from credible references and revise the content that the tool spits out.

Conclusion

We're all excited about the new possibilities that ChatGPT and Co. have brought us. In the end, however, the tool is simply just an assistant, not a replacement. The result can only be as good as the information we've provided. When a text is uploaded or published, ChatGPT is not "liable" – but you as the website operator or profile owner. YOUR name is ultimately associated with the text that was typed there. With this awareness, you should create content and make it available to the internet. If you can represent the result well and without a guilty conscience, there is no reason to forgo the support of language models.

Do you need support with the professional and efficient creation of high-quality website content? Our online marketing team is happy to be at your service.