AI chatbots are increasingly being used to find information. But how reliable are their answers when it comes to current news? An analysis by PR analytics platform Aclipp shows that AI systems often fall back on outdated sources even when more recent information is available.
For the analysis, Aclipp asked seven widely used AI tools the same 12 questions about coalition talks in Berlin every day for a week in mid-October. The answers were then checked against more than 900 media articles and the sources used by the tools. The results showed that the tools initially reflected new developments largely correctly, but increasingly relied on older information over time. The day after a political decision, 55% of the answers still reflected the latest state of affairs; one week later, this had fallen to just 39%. By then, 61% of the answers were based on outdated information. There were also notable differences between individual tools. Perplexity, for example, cited the most sources per answer, but also relied on particularly old information, which led to incorrect responses. According to Aclipp, almost one in ten answers contained a verifiable factual error, often involving dates or the attribution of political parties.
Limited information diversity presents another challenge. Compared with conventional search engines, AI chatbots often reflect only a small proportion of the available information. This is supported by a study from Aalborg University in Denmark: the chatbots tested produced much more similar results than those found through a Google search. As a result, users are more likely to encounter the same information, while alternative perspectives become less visible. Over time, this could reinforce the prominence of already widespread content while pushing other viewpoints further into the background. The researchers suggest that this may be linked to the way language models work, identifying frequently occurring patterns and therefore giving less weight to less common information.
What does this mean for the use of AI in professional research?
We have already discussed the limited diversity of information and potential biases in AI-generated answers in previous articles. However, the Aclipp analysis highlights another problem: even when more recent and relevant sources are available and have already been used by the AI systems, they are not always given priority. Access to the internet alone is therefore no guarantee of up-to-date research results.
Especially when dealing with fast-moving topics, users should therefore not only assess whether an answer appears plausible, but also check the publication dates of the sources used and question whether the response genuinely reflects the latest developments. Key claims should also be cross-checked against current primary sources.
By Ida Kandler
