Quick Cheat Sheet
Claude 4.5
- Use full context.
- Give examples of style.
- Break long tasks into steps.
- Tell it the audience and reason.
GPT-5.1
- Use personas.
- Give structure and length rules.
- Set tool logic.
- Ask for planning before execution.
Gemini 3.0
- Declare input type.
- Give one clear example.
- Define format and constraints.
- Separate context from task.
Claude 4.5
Anthropic wants you to be clear, specific and thorough. Claude responds best when you tell it the goal, the audience and the shape of the output. It also benefits from strong examples and context that explains why the task matters.Think of Claude like a partner who expects full direction before doing great work.
Example: Rewrite a SaaS landing page
You:
Act as: a conversion writer for early-stage founders.
Goal: Improve the hero section for a tool that helps teams understand user behavior without setup headaches.
Audience: Small teams on Shopify.
Instructions:
Provide two headline options.
Provide one short subhead.
Provide a CTA.
Tone: clean and confident.
Claude gives sharper copy when you define the target and the format. It also handles context well when you explain who the audience is and why they care.
Example: Turn raw interviews into insights
You:
Role: Senior researcher.
Task: Summarize interview notes from five early users of our onboarding product.
Context:
Users struggled with first-time setup.
They liked the quick video walkthrough.
They wanted simpler wording on steps.
Instructions:
Write one page in plain language.
Add a short section called “What surprised people”.
Add another section called “What to validate next”.
Keep it tight.
Claude follows structure without drifting when you set the frame clearly.
Example: Plan a complex project
You:
You’re helping me: rebuild our help center.
Task: Create a detailed plan with phases, timelines, and key decisions.
Context: We serve founders, small support teams, and new product managers.
Rules:
Start with a short overview.
Then break the project into clear phases.
Add dependencies.
End with open questions we should answer.
Claude handles long tasks well when the steps are spelled out. Source: https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-4-best-practices
GPT-5.1
OpenAI wants users to give GPT-5.1 a clear persona, a structure to follow, and limits on length. It performs best with sections that guide tone, format, and initiative. Treat it like a flexible agent that follows rules.
Example: Market brief for founders
System:
You are: a SaaS market analyst.
Objective: Produce short research founders use to make decisions fast.
Tone: neutral and simple.
Structure:
Start with a short snapshot.
Then list key players.
Then describe gaps worth exploring.
Finish with risks.
Length: five to eight short paragraphs.
User: Give me a brief on the state of time-tracking tools for small agencies.
GPT-5.1 sticks to structure when you give it clear sections and limits.Example: Social content creation
System:
You are: a content partner for early-stage SaaS founders.
Task: Produce ten post ideas for LinkedIn.
Rules:
Each idea must include a short title and a short angle.
No repeated terms.
No clichés.
Keep everything tight.
User:
Create ideas for a founder running a tool that helps manage async check-ins.
GPT-5.1 follows strict formatting when those rules live in the system message.Example: Support assistant
System:
You are: a support triage assistant.
Task: Review user messages and decide if we respond from support or escalate to engineering.
Rules:
If the message mentions data loss, security, or outages, escalate.
If it’s about setup, UI, or billing, give the next step.
When things are unclear, ask one follow-up question.
User:
A user says their workspace is missing a week of data after an app update. What should we do?
GPT-5.1 performs well when boundaries and decisions are defined upfront. Source: https://cookbook.openai.com/examples/gpt-5/gpt-5-1_prompting_guide
Gemini 3.0
Google recommends stating the input type, giving the exact task, showing a clean example, and defining the output format. Gemini responds strongly to structured instructions and concrete constraints.
Example: Turn messy feedback into product themes
You:
Task: Convert this long user message into three product themes.
Themes allowed: onboarding, speed, reporting, integrations, billing.
Only return themes that match.
Example input: “I want faster load times and a simpler first-time setup.”
Example output: onboarding: simpler first-time setup. speed: faster load times
Now process this message: “Exports are slow and connecting Stripe took too many steps.”
Gemini handles structured extraction well when you show an example and define valid fields.Example: Hiring brief
You:
Task: Create a hiring brief for a product designer.
Length: one page.
Tone: simple and clear.
Sections:
Role overview.
What the person will work on.
What good looks like.
What to share in the application.
Do not add extra sections.
Gemini follows format rules closely when they’re written next to the task.
Example: Competitive summary
You:
Here is context: [paste notes]
Task: Produce a short competitive summary.
Format: One short intro, Then a paragraph on strengths. Then a paragraph on weaknesses. No lists. No extra commentary.
Gemini respects the shape when you separate context from instructions. Source: https://ai.google.dev/gemini-api/docs/prompting-strategies
Extra resources
To understand AI in a simple, non-technical way, watch this video:
https://www.youtube.com/watch?v=7xTGNNLPyMI
To learn how to build strong AI agents that handle real work at scale, read this guide from Anthropic:
https://www.anthropic.com/engineering/effective-harnesses-for-long-running-agents