Which AI chatbot is best? A Singapore writer tests four and finds it depends on you

A simple test with four chatbots

As Her World Singapore reported, writer Karen Fong wanted to know whether people prefer particular AI chatbots out of habit or because some are really better at certain jobs. She put the same prompts to four large language models (LLMs, the text-generating systems behind chatbots): ChatGPT, Claude, DeepSeek and Gemini.

She is upfront that this is not a scientific benchmark. It is one person applying the close-reading habits she learned in junior college literature class to compare the answers.

Travel plans: close, but not identical

The first prompt asked for a 3-day itinerary in Seminyak, Bali, for three female friends travelling at the end of August, mixing beach clubs, restaurants and a spa. Fong found all four answers broadly similar, and says the choice would barely matter for a question like this beyond how you like to be spoken to.

The differences were in style. She found ChatGPT too brief and light on reasons for its picks, though it was the one that added rough cost estimates. DeepSeek and Gemini struck her as concise but explained their choices, and DeepSeek added a quick list of top restaurants and spas.

Gemini was her favourite. It gave sources she could check, links for booking and a map of the locations.

Emotional advice and the confidence trap

The second prompt asked for advice on making friends as a 30-something Singaporean. Fong noted that all four were sympathetic, but because she mostly uses AI for research, the emotional answers sounded somewhat fake to her. DeepSeek felt the most approachable.

Assistant Professor Lee Yi-Chieh of the National University of Singapore used this to point to what he calls AI-mediated relationships. People use AI to rehearse difficult conversations or for companionship, which can help, for example by giving a lonely person comfort or a safe place to practise. But he said research also points to risks such as weaker social skills and, for heavy users, a decline in well-being.

The third prompt asked each chatbot which of the four is most accurate. Each largely claimed the top spot. Claude stood out by acknowledging its own bias and suggesting a crowd-sourced AI evaluation platform and a hallucination leaderboard. A hallucination is when a model states something false or unsupported as if it were true.

What the experts say about choosing

Dr. Grandee Lee of the Singapore University of Social Sciences thinks the gaps between models are quickly narrowing. He describes Claude as noted for longer, more nuanced reasoning and coding, and Gemini as benefiting from tight links to Google's ecosystem and search.

Professor Lee Yi-Chieh says the practical differences matter more in the moment: whether it can search the web, what software it connects to, whether it handles images, audio and video, how much it can read at once, how fast and costly it is, and what happens to your data. He adds that a model's tone also decides whether people keep using it.

On hallucinations, Dr. Grandee Lee puts much of the blame on users who give too little context. A model that lacks enough to go on may fill the gaps with plausible but inaccurate material. He also warns that the most confident model is not necessarily the most reliable, since people trust a confident tone too readily.

His test for the best answer is that it shows its work, separates fact from interpretation and links to original sources. He mentions retrieval-augmented generation, where a system first looks up relevant documents and answers from them instead of from memory. Such techniques reduce the problem but do not solve it.

Use it as a coach, not a crutch

Fong's takeaway is to treat AI like a capable assistant who still needs very clear, specific instructions. Professor Lee says some of his students use AI to organise their studies, plan activities, track deadlines and even choose where to eat.

He also worries about students who use AI to generate ideas or finish work without understanding how the answer was reached. In his view AI works well as a coach or coordinator, but people should avoid outsourcing their judgment, relationships or sense of self to it.