*Thisarticle is based on a YouTube video by Sophie Hundertmark, an expert in the use of artificial intelligence with a focus on chatbots and strategic AI applications in companies and public institutions. Sophie is a researcher and lecturer at the Lucerne University of Applied Sciences and Arts and is doing her doctorate in Conversational AI at the University of Fribourg. The blog text was created using a custom GPT model that was trained on Sophie’s video content, language style and expertise. The result is well-founded, up-to-date articles based on Sophie Hundertmark‘s own expertise.
You can find the link to the video at the end of this article.
Artificial intelligence can provide answers to almost any question in a matter of seconds. But just because an answer sounds convincing doesn’t mean it’s necessarily correct. AI models can hallucinate, misinterpret information, or reproduce outdated content.
That’s why the same rule applies to fact-checking with AI: Don’t blindly trust the results. Use the answers as a starting point, and double-check important information with reliable sources or consult experts if necessary.
In this post, you’ll find three prompt templates that you can copy directly and use in ChatGPT, Claude, Gemini, or any other AI tool.
Why Fact-Checking with AI Makes Sense
Even if an AI performs a fact-check, that doesn’t automatically mean the result is correct. Language models can also hallucinate or misclassify information.
However, with good prompts, you can get the AI to work more transparently, disclose its uncertainties, and structure its responses more effectively. That’s exactly what the following three templates are designed to help you with.
Prompt 1: The Structured Fact-Check
Using this prompt, the AI breaks down a statement into individual sub-statements. She then evaluates each point individually, explains the basis for her assessment, and clarifies where uncertainties exist.
Prompt for copying
Prüfe die folgende Aussage auf ihren Wahrheitsgehalt:
„[AUSSAGE EINFÜGEN]“
Gehe dabei so vor:
1. Zerlege die Aussage in einzelne überprüfbare Teilbehauptungen.
2. Bewerte jede Teilbehauptung als:
- korrekt
- teilweise korrekt
- falsch
- nicht überprüfbar
3. Begründe jede Einschätzung kurz.
4. Markiere ausdrücklich, wo Unsicherheiten bestehen oder keine verlässlichen Informationen vorliegen.
5. Nenne, falls möglich, welche Art von Quelle zur Bestätigung geeignet wäre.
Practical example from the video
In the video, I test the prompt with a statement I was unsure about myself:
“Orange seeds are edible and can be used to make jam.”
The AI first breaks this statement down into individual sub-statements and then evaluates them separately. This quickly makes it clear which parts are well-supported and which statements should be treated with caution.
Prompt 2: Verifying Sources
This prompt is particularly well-suited for current topics. The AI should conduct targeted research and support its answer with multiple reputable sources.
Prompt for copying
Recherchiere die folgende Behauptung und belege deine Antwort mit aktuellen, vertrauenswürdigen Quellen:
„[AUSSAGE EINFÜGEN]“
Suche nach mindestens zwei unabhängigen, seriösen Quellen.
Gib jeweils Quelle und Veröffentlichungsdatum an.
Sag mir klar, ob die Quellen die Behauptung:
- stützen,
- widerlegen oder
- uneinheitlich sind.
Wenn du keine verlässliche Quelle findest, sag das offen, statt zu raten.
Here, too, I made the same comment about the orange seeds in the video. The result was largely comparable to that of the first prompt and also showed which pieces of information could be corroborated by sources.
Prompt 3: The Hallucination Stress Test
With this prompt, you ask the AI to critically evaluate its own response. This clearly distinguishes between speculation and verified information.
Prompt for copying
Ich gebe dir gleich eine Aussage. Bevor du antwortest, frage dich selbst:
"Weiß ich das wirklich oder klingt es nur plausibel?"
Aussage:
„[AUSSAGE EINFÜGEN]“
Antworte in drei Spalten:
| Was ich mit hoher Sicherheit weiß | Was ich vermute, aber nicht belegen kann | Was ich nicht weiß |
Erfinde keine Zahlen, Namen oder Quellen.
Lieber "Das weiß ich nicht" als eine erfundene Antwort.
In the video, the answer is clearly presented in a table. This allows you to immediately identify which information is reliable and where there are uncertainties.
Critical thinking remains essential
These three prompts will help you better understand AI responses. Nevertheless, they are no substitute for a thorough review.
Especially when making important decisions, you should also compare the information with reliable sources or consult experts. AI is a valuable tool—but the responsibility for evaluating the information remains with humans.
Conclusion
With the right prompts, you can significantly improve the quality of AI responses. These three templates help you systematically evaluate claims, incorporate sources, and spot hallucinations more quickly.
But never forget: Even a fact-check by AI isn’t automatically proof of the truth. Therefore, use the results as a guide and examine them critically.
Any further questions?
Do you have any questions? I am happy to support you, act as a sparring partner and answer your questions. I am always happy to receive your messages, preferably by WhatsApp message or e-mail.
