The rise of AI-driven search engines

In the fall of last year, both Google and OpenAI launched their latest innovation: search engines that generate answers using advanced language models. These smart chatbots, linked to powerful search engines, are now also available to users in the Netherlands and Belgium. Let's see how they perform in practice.

How do these AI search engines work?

To gain insight into their way of working, we asked the chatbots themselves for an explanation. Both Gemini “grounded” and ChatGPT Search indicate that they first analyze the user's question. When no additional information is needed, they draw from their own knowledge base. When additional or current information is required, they enable their search function.

Gemini uses Google Search for this, while ChatGPT uses its own integrated search engine. After collecting information, they formulate an answer. Gemini combines knowledge of the AI model with search results, resulting in an answer that is “anchored” in reality. ChatGPT, on the other hand, provides its responses with links to sources so that users can access the original information themselves. Gemini largely lacks this source citation, which makes it difficult to verify the accuracy of the information.

Practical performance

To test the effectiveness of Gemini and ChatGPT as search engines, we confronted them with various queries. We also included traditional search engines such as Google and DuckDuckGo in the comparison.

The results were mixed. ChatGPT Search proved to be excellent at answering specific questions, such as the latest service price index or the date of Tweakers's founding date. For more complex questions, such as looking for free streaming options for a movie, Gemini provided more relevant suggestions.

Both AI chatbots showed their strength in answering detailed questions and dealing with complex topics. For example, they were able to correctly indicate how late a specific supermarket was open and provided clear explanations of current economic issues.

Challenges and Limitations

However, we also encountered the limits of these AI search engines. In very specific searches, such as finding the cheapest curved monitor during Black Friday, or determining the exact position of the Netherlands in the global economy, none of the tested search engines were able to provide a satisfactory answer.

Future perspective

After a few weeks of use, it appears that AI-driven search engines have both strengths and weaknesses. They perform well with simple queries and are sometimes even better at answering complex queries than traditional search engines. A major stumbling block, however, remains the phenomenon of “hallucinating”, where inaccurate information is presented convincingly.

Despite these challenges, AI search engines certainly have potential. In many cases, they can answer questions more quickly. In addition, there appears to be a growing dissatisfaction with traditional search engines, offering an opportunity for these new AI variants to attract a significant audience.

While there is still room for improvement, these AI-driven search engines undoubtedly mark a new phase in how we search and find information on the internet. The coming years will tell how they continue to develop and whether they will fundamentally change how we use the web.