The best AI for selecting a research topic is not ChatGPT, although the OpenAI Language Model is one of the most used AI models across the globe. In research conducted by Stanford University’s Human-Centred Artificial Intelligence, titled “The 2025 AI Index Report,” US-based AI models like ChatGPT still lead the AI race in terms of popularity and usage, but does that mean ChatGPT is the best for researchers, especially beginners who are just starting their research journey?
.jpg)
A colleague researcher once stated, “If you want to make any AI model drunk, give the wine to the one using it.” After a careful review, I think he is right! In most cases, AI mistakes stem from the user, though we cannot rule out system inefficiencies, especially in an era when AI is still in its infancy. But it takes the knowledge and expertise of a user to fully harness the power of any AI model. In almost every case, using AI for research with no or limited knowledge about what research entails will always lead to many errors and inaccuracies. However, some AI models are advanced and can sometimes self-correct mistakes that emanate from users. As a user of both ChatGPT (from 3.5 to 5.2) and Google Gemini (from Bard to Gemini 3.0 Pro), I can confidently say (although subjectively) that Gemini is the best for users beginning their research journey, especially when it comes to selecting a research topic.
Help me formulate a research topic on postpartum sleep deprivation
Immediately, ChatGPT provided about 10 topics based on criteria such as mental health and psychological outcomes, the maternal and infant health nexus, and social, environmental, and cultural dimensions. Across the topics provided by ChatGPT, almost all were separated by a colon (:). For example, the first topic was “Postpartum Sleep Deprivation and Maternal Mental Health: Examining Its Effects on Depression, Anxiety, and Emotional Well-Being. Whereas the topic is clear and direct, the use of colons is becoming excessive, robotic, and easy to spot as AI-generated. Also, the topic was broad (many variables), and the constructs were overlapping (“Depression”, “anxiety,” and “emotional well-being” are closely related. Emotional well-being typically includes depression and anxiety, so listing them separately may feel redundant). And did you see that ChatGPT did not specify the kind of mothers? (e.g. first-time mothers, mothers of multiples, mothers with C-sections).
 |
| Image from Storyset |
Overall, experience and experimentation tell us that Gemini is clearly ahead of ChatGPT in topic selection. However, one must not substitute research knowledge for AI assistance.