Gemini
Google's assistant: huge context, strong with images and video, wired into Gmail and Docs.
Best for
- large files and long PDFs
- understanding images and video
- Gmail and Docs work
- answers that need the live web
Signature move
Hand it the whole file and ask what it would cut.
How it wants to be spoken to
Name the files and say “read all of it before answering”.
Number the steps on anything complex — it follows a list far better than a paragraph.
Put the most important rule last; it weights the end of the prompt heavily.
Ask it to quote the line it's relying on.
Why it glitches
None of this is you doing it wrong. These are the shapes it reliably falls into — and the fix beside each.
Follows the last instruction best → put your non-negotiable at the end.
Mixes up which document you mean → name each one and ask it to say which it used.
Confidently vague on niche facts → ask for sources and check two.
Summarises so hard the nuance vanishes → ask for “what the file does not cover”.
A prompt that works there
Attached is [file]. Read all of it before answering. What I need: [task] Rules: quote the exact line you're relying on for each point. If the file doesn't cover something, write “not in the file”.
Bring Tay your actual task and she'll shape it for Gemini.
Practice with Tay