Gemini responds differently depending on how you frame your input. These aren't settings you toggle in a menu — they're structural phrases and prefixes that shift Gemini's behavior. Use them at the start of a prompt, mid-conversation to redirect, or inside a system instruction if you're working via API. Bookmark this page and pull from it whenever you hit a wall.
Mode Triggers
- Think step by step: — forces Gemini to show its reasoning chain before answering, not after
- Before you answer, list your assumptions: — surfaces hidden premises Gemini is making about your prompt
- Answer only from what I've given you: — prevents Gemini from hallucinating outside the context you pasted
- Respond as if you're an expert practitioner, not a teacher: — drops the tutorial tone, gets you peer-level language
- Skip the preamble: — cuts the 'Great question!' opener and gets straight to content
Persona and Role Assignment
- You are a [role]. Stay in character for this entire conversation: — locks Gemini into a persistent persona without drifting back to assistant mode
- Act as a skeptic reviewing this idea: — flips Gemini into adversarial mode; useful for pressure-testing strategy
- Take the opposing position and argue it well: — forces a steelman of the counterargument
- You are a senior editor. Rewrite this without changing my meaning: — editing mode that preserves voice
- Respond the way a CFO would respond to this pitch: — injects a specific decision-maker's frame into feedback
Output Format Control
- Return this as a numbered list only, no prose: — strips paragraphs when you just need scannable points
- Format this as a table with columns [X], [Y], [Z]: — auto-generates structured comparisons
- Give me the answer in three sentences max: — hard cap on length; useful when you're feeding output into another prompt
- Use plain text, no markdown: — critical when copying into tools that don't render formatting
- Output only the final result, no explanation: — zero commentary, just the deliverable
Iteration and Refinement Commands
- Give me three variations of this, each with a different angle: — generates options instead of one locked answer
- Make this more direct and half as long: — tightens copy without a full rewrite prompt
- Now make it sound less formal: — tone shift without restarting the session
- Keep everything except the last paragraph, rewrite that: — surgical edit instead of a full regeneration
- What would make this stronger? List three specific changes: — turns Gemini into a collaborator, not just an executor
Context Loading Commands
- Here is background context. Do not respond yet — just confirm you've read it: [paste content] — loads information before asking a question, reduces hallucination
- I'll give you three inputs. Wait until I say 'go' to respond: — batches multiple pieces before Gemini starts generating
- Treat the following as ground truth. Do not contradict it: [paste facts] — anchors output to specific facts you control
- Reference only the document I've pasted. Do not use outside knowledge: — enforces document-grounded answers in research mode
Here is background context. Do not respond yet — just confirm you've read it:
[paste your document, transcript, or data here]
Once I say 'go,' I'll ask you questions about it.Analysis and Evaluation Commands
- Rate this on a scale of 1–10 and explain what's holding it back from a 10: — structured critique with a benchmark
- Identify the three biggest risks in this plan: — risk-first analysis instead of cheerleading
- What assumptions does this argument depend on? — exposes logical dependencies in strategy or copy
- What is this missing that a buyer would notice immediately? — buyer-perspective gap analysis
- Compare these two options and tell me which one you'd choose and why: — forces a recommendation, not just a comparison
Writing and Copy Commands
- Rewrite this in the voice of the original author, not as an AI: — reduces the synthetic tone in drafted content
- Write this as a cold email, under 100 words, no soft close: — strict copy format with hard constraints
- Turn this bullet list into a paragraph that flows naturally: — converts outlines into readable prose
- Write a hook for this that starts with a specific problem, not a question: — avoids the tired 'Have you ever…' opener
- Give me five subject lines for this email. Vary the approach — curiosity, benefit, pattern interrupt, proof, urgency: — labeled variation set for A/B testing
Write five subject lines for the email below. Label each one by approach: curiosity, direct benefit, pattern interrupt, social proof, urgency. No emojis.
[paste your email here]Research and Summarization Commands
- Summarize this in three layers: one sentence, one paragraph, five bullets: — gives you multiple zoom levels from one pass
- Extract only the claims that are verifiable, ignore the rest: — filters opinion from fact in dense documents
- What are the key decisions someone reading this needs to make? — converts information into action
- Condense this to the single most important insight: — forces prioritization when everything feels important
- List every question this document raises but does not answer: — identifies gaps in research or arguments
Conversation Management Commands
- Forget everything I said before this message. Start fresh: — resets context drift in a long session
- Summarize what you know about my goal so far: — confirms Gemini has the right model of what you're building
- Stop asking clarifying questions and make your best attempt: — cuts the back-and-forth when you need speed
- From now on in this conversation, always [rule]: — sets a session-level rule without repeating it every prompt
- End every response with one specific next step I should take: — turns outputs into action, not just information
Debugging Bad Outputs
- That was too generic. Give me a specific example with real details: — pushes past vague answers
- You repeated yourself. Remove the redundancy and resubmit: — cleans up padded responses
- You answered a different question. Here's what I actually asked: [restate it] — redirects without restarting
- You added an opinion I didn't ask for. Strip it and give me only what I asked: — keeps outputs scoped
- This sounds like an AI wrote it. Rewrite it so it doesn't: — the most-used refinement command for anyone publishing content
The fastest skill you can build with Gemini is constraint writing. Vague prompts get vague outputs. Every command above works better when you add at least one specific constraint — length, format, tone, or scope. Start there before you try anything else.