Gemini Prompts

How Gemini differs from ChatGPT, Claude and Grok

Gemini is strongest where the input is mixed: a screenshot plus text, a table plus a question, a link plus a request to compare. You don't need to describe in words what you could attach as an image — and that changes a good half of the library's prompts, which assume you will paste text.

The second difference is how closely it sits next to Google search. On practical everyday tasks — planning, shortlisting, comparing — it leans on what it finds, not only on what it was trained on.

Rewriting a library prompt for Gemini

• State the output shape first. "A table with columns …", "exactly five bullets", "one paragraph". An open-ended request produces the most generic answer of the four models here; a stated format produces the most precise one.
• Attach instead of describing. A screenshot of the interface, a photo of the document, a slice of the spreadsheet all work better than a retelling. A prompt with a "paste your text" slot is often faster to run by attaching the file.
• Separate found from inferred. A line like "mark what comes from sources and what is your own judgement" turns a convenient answer into a checkable one.

Where it falls down

It blends what it found with what it assumed, without flagging the seam — the most common way to get a confident wrong fact out of Gemini. If the answer feeds into real work, always ask it to mark sources.

Long reasoning without a given structure drifts: by the end of the answer it is addressing a slightly different question. For large tasks, ask for a plan first, then expand point by point.

And it asks fewer follow-up questions than the others: where a request allows two readings, Gemini picks one and doesn't mention it. Resolve ambiguity in the prompt up front.