Using ChatGPT to fill information gaps

—

dilligence

When information is difficult to find, scattered across different sources or only indirectly available, many people turn to ChatGPT. A recent report by OpenAI illustrates this using salary information as an example. According to the report, users in the United States submit around three million salary-related queries to the system every day.

The report shows that such queries are particularly common in situations where salaries are difficult to compare, depend heavily on negotiation or play an important role in career decisions. In general, the greater the uncertainty and the more important the decision, the more likely people are to use ChatGPT for guidance, to fill information gaps and obtain an initial assessment. Instead of carrying out the more time-consuming process of consulting multiple sources that may provide ambiguous results, users receive a direct, consolidated answer from the AI.

What does this mean in practice for the use of AI in professional research?

Salary research illustrates that LLM-based chat systems are used particularly often for questions where reliable public information is difficult to obtain. Yet this is also where a key limitation becomes apparent: both conventional search engines and AI models reach their limits when the information being sought is unavailable, incomplete or unreliable. The lower the availability and quality of the underlying information, the more cautiously the generated output should be assessed.

This is especially important for queries where very little information exists or where only a small number of reliable sources are available, such as discussion forums. Results need to be assessed critically and should not be accepted without verification. In such cases, ChatGPT can provide useful ideas and initial guidance, but it does not replace thorough research. Ultimately, some questions simply do not have satisfactory answers because too little data or information is publicly available.

By Ida Kandler