The problem with AI-generated texts
ChatGPT, DeepL and Gemini are good enough to produce usable texts. The trouble is, “usable” means the text sounds convincing enough that people stop questioning it. It does not mean the text is correct.
Here is what regularly goes wrong:
- Hallucinations—AI invents facts, quotations and sources. You do not spot it right away, because the text reads so smoothly.
- Stylistic sameness—AI texts give themselves away through monotonous sentence lengths, excessive politeness and the same stock phrases in every other paragraph.
- Compliance gaps—AI knows neither your industry regulations nor your internal communication guidelines.
- Breaks in tone—A premium brand that sounds like a chatbot damages itself.
AI literacy under the EU AI Act
Article 4 of the EU AI Act has applied since February 2, 2025. Providers and deployers of AI systems are expected to ensure that employees and other people using AI on their behalf have a sufficient level of AI literacy. In its AI literacy FAQ, the European Commission explicitly mentions cases such as ChatGPT for advertising copy or translations, and points to risks such as hallucinations. I do not replace legal advice. I support the language-quality side: facts, sources, tone, style, clarity and publication-ready copy.
Why work with me
At Jung von Matt I checked campaigns for BMW and Sparkasse against quality criteria, and at Goodgame Studios and InnoGames I checked software releases. Software or language: the process is the same.
What I check
- Factual accuracy—are the facts right, or has the AI hallucinated?
- Sources and evidence—do the cited sources support the claim?
- Style and tone—does the text suit the brand and the target audience?
- Terminology—consistent, technically correct and easy to understand?
- Compliance-related passages—precise enough, without false legal or service promises?
- Accessibility requirements under the BFSG and BITV 2.0