21.07.2026 - It is becoming increasingly difficult to identify text and websites that have been generated using artificial intelligence (AI). Reports to the NCSC involving emails and websites created with the help of AI continue to increase. As AI models keep improving, many of the warning signs that used to give them away are disappearing. That makes it all the more important to know what to look out for. In the second part of its summer series, the NCSC explains how to spot AI-generated texts and websites.
It is becoming increasingly difficult to identify scam emails and websites based on writing style alone. In the past, a badly written email or a website full of spelling mistakes was often a clear sign of phishing or fraud. Today, that is no longer something you can rely on. AI can produce flawless texts in any language, making fraudulent content appear legitimate. Scammers also use AI because it can create professional-looking websites very quickly. But, AI isn't perfect: it tends to fall back on familiar patterns, and these patterns can provide clues.
Long dashes
A common feature is the long dash, also known as an em dash (“—"). This dash is very versatile, and AI-generated texts often use it to connect ideas or add nuances and asides. While dashes are common in English, they can be a sign of AI-generated text if they are used unusually often or in a way that feels formulaic.
Overuse of bold text
To improve readability, AI tools often highlight key points by putting sections of text in bold, making as much information as possible visible at a glance.
The German Eszett (“ß”)
Because AI models that generate German are largely trained on Standard German texts and follow Standard German spelling rules, their output often contains the Eszett, “ß”. In Swiss Standard German, however, this letter isn't used; words that would contain “ß” in Germany or Austria are written with "ss" instead. The use of "ß" is therefore incorrect in Swiss Standard German and would be especially unusual in texts produced by official organisations.
Even sentence length and structure
AI-generated texts are now often free from errors and appear correct. The style is usually neutral and even, lacking strong emotion. Sentences are often structured similarly and can feel unusually smooth.
Single-page websites: All information on one page
Scam websites often only remain online for a short time, so they have to be created very quickly. For simplicity, these sites are often designed as single-page sites with only a few small subpages for the registration process. A website consisting of one long page listing all the information one section after another is a strong indication that it may have been generated using AI.
Exaggeration without substance
AI tends to produce long texts that don't actually say much and contain exaggerations. This is not due to the original training data, but rather subsequent training with human feedback. People often rate detailed, confidently worded answers as more helpful and credible, so models have learned to produce this kind of text. In fraudulent texts, this often manifests as an overuse of superlatives and repetitive sentence structures, while the actual information conveyed is minimal.
Animation
Another sign that AI may have been used to create a website is the excessive use of animations and other moving elements. AI may have learned to associate these features with good web design and may therefore overuse them, even where they do nothing to make the website easier to read.
Inconsistent images and exaggerated Swiss references
Another issue with AI-generated websites is that the text may not fit the images used, or the images themselves may not form a coherent whole. For example, the backgrounds or views from windows may not match, showing different landscapes or surroundings that couldn't plausibly be seen from the same place.
Another clue can be an excessive number of references to Switzerland. To make an AI-generated website look Swiss, scammers usually have to include this connection explicitly in the prompt, i.e. the instruction given to the AI system. The AI may then overdo it, adding too many Swiss crosses or other Swiss landmarks.
Conclusion
AI-generated website texts tend to be well structured and grammatically correct, as well as being neutral in tone. However, they often lack specificity and have a very uniform style. You usually need to look at several clues together to tell whether a text or a website may have been generated by AI. Inconsistencies are not the only thing to watch out for – the usual safety rules still apply: read reviews carefully before using a website, and never enter passwords or credit card details on pages that seem suspicious or that you opened via a link in an email or text message.
Current statistics
Last week's reports by category:
Last modification 21.07.2026
