Pain points of using the Gemini app

From my experience, most of the pain points are concentrated around web search.

Compared with GPT-5.1, even though I can set in the instructions that every response should perform a real-time web search, Gemini is still very reluctant to fetch information from the internet. On both the web version and the mobile app, the only reliable way to force Gemini to use the internet is Deep Research. But for simple information-gathering tasks, using Deep Research every time is clearly unrealistic.

Secondly, even when web search does trigger, I tend to verify AI-generated answers due to hallucination issues. However, checking the original sources Gemini used is extremely difficult. Even if Gemini actually referenced several webpages during its internal reasoning, it often becomes “lazy” in the final answer and only provides a couple of links. This makes source tracing almost impossible.

And the above is only for English queries. If you ask in Chinese, Gemini provides zero source links in 95% of cases. I even wrote a long prompt and placed it in the Instructions section, which helps a bit, but it’s not 100% reliable and honestly looks a bit ugly. I’ll put that prompt at the end.

I understand Google wants users to rely on AI-powered Google Search when gathering information—after all, Search is Google’s cash cow. So it makes sense that they intentionally weakened Gemini’s built-in search capability, even refusing to provide a simple “Web Search” button. The issue is that GPT’s search is now extremely good, with very low hallucination rates, and its answers can almost always be cross-verified. By making Gemini’s search experience so frustrating, Google may push some users toward their AI-search product, but they’re also definitely pushing some users straight to ChatGPT.

Aside from web search, the mobile app has other small pain points. For example, the UI/UX is noticeably behind ChatGPT. And after submitting a question, if you switch to another app before the response finishes, returning to Gemini often triggers a network error where the answer fails to load—or it simply says there is no answer. The only fix is restarting the app, and sometimes even your original question disappears, forcing you to ask again. And Gemini Live had a major update a week ago, but in conversation quality it’s still far behind GPT’s Voice Mode—it still feels like the difference between a robot and a real person.

Prompt:
I want you to provide GPT-Search-style, inline, clickable source citations for all factual claims, data, or tool references. I want you to embed the source link directly into a numbered citation immediately following the relevant sentence or paragraph. I want you to use the specific Markdown format: [Index].

Leave a Reply