Zum Inhalt springen

Evaluating AI Output

    Artificial intelligence can generate texts, summaries, arguments, or answers in just a few seconds. However, just because an answer sounds convincing does not mean it is automatically correct. AI output should therefore always be critically evaluated.

     What should I look for?

    1. Check the factsAre names, numbers, dates, definitions, and other specific statements actually correct? AI can make mistakes or invent information, especially when it comes to specific facts.
    2. Verify sourcesIf the AI cites sources, make sure they actually exist and align with the statement. Whenever possible, check the original source: Does it really say what the AI claims? This step is particularly important for academic work.
    3. How up-to-date is the information? Information can become outdated. When dealing with current topics, statistics, legal regulations, or scientific developments, you should therefore pay particular attention to the publication date and up-to-date sources.
    4. Consider context and completenessDid the AI understand the question correctly? Are important aspects, counterarguments, or limitations missing? AI often provides a plausible perspective, but not necessarily the only one or the most comprehensive one. A careful analysis of the sources used by the AI can often help you determine whether the AI’s answer is balanced. Is it perhaps based on only a few sources that, while easy to find, are not very scientific?
    5. Recognize biasesAI can adopt existing biases or one-sided perspectives from its training data. Therefore, ask yourself: Which perspective is being presented—and which ones might be missing?
    6. Question the reasoningIs the answer logical and comprehensible? Are the conclusions sufficiently substantiated? An answer that is well-written can still contain errors in reasoning.
    ExampleAI writes:
    • „A study by Smith et al. (2023) shows that …”
    Do not simply take this statement at face value. First, check whether the study actually exists. Do the author, title, and publication year match? And most importantly: Does the study really say what the AI claims?
    AI output is a starting point—not evidence. The more important a statement is to your work, the more carefully you should verify it independently. The responsibility for what you use in an academic paper remains with you.