The most important skill you’ve never been taught:
Prompting.
10 dead-simple tips to get better results from AI (worth saving):
Prompting.
10 dead-simple tips to get better results from AI (worth saving):
1. Cut the fluff
Most prompts are vague, rambly, or full of filler.
AI doesn’t need your backstory it needs clarity.
✗ “Hey can you maybe explain AI a little bit, like in a way someone could understand?”
→ “Explain AI in 3 bullet points as if I’m 12 years old.”
Most prompts are vague, rambly, or full of filler.
AI doesn’t need your backstory it needs clarity.
✗ “Hey can you maybe explain AI a little bit, like in a way someone could understand?”
→ “Explain AI in 3 bullet points as if I’m 12 years old.”
2. Assign roles + memory
Don’t just say “write this.” Give the AI a job identity and context anchor.
Example:
“You are a McKinsey consultant. Act as if you’ve advised Fortune 500 CEOs for 10 years. Create a 3-phase GTM plan for an AI SaaS startup.”
Output quality jumps instantly because the model maps to patterns in that role.
Don’t just say “write this.” Give the AI a job identity and context anchor.
Example:
“You are a McKinsey consultant. Act as if you’ve advised Fortune 500 CEOs for 10 years. Create a 3-phase GTM plan for an AI SaaS startup.”
Output quality jumps instantly because the model maps to patterns in that role.
AI is shaping your future - discover the key developments, challenges, and opportunities in The AI Foreground.
aiforeground.beehiiv.com
aiforeground.beehiiv.com
3. Control reasoning depth
AI can “think shallow” or “think deep.” Guide it.
Example:
“Before answering, write your reasoning in 3 steps. Then provide the final conclusion.”
vs.
“Skip reasoning, give me the most likely answer in 1 sentence.”
You decide: speed vs depth.
AI can “think shallow” or “think deep.” Guide it.
Example:
“Before answering, write your reasoning in 3 steps. Then provide the final conclusion.”
vs.
“Skip reasoning, give me the most likely answer in 1 sentence.”
You decide: speed vs depth.
4. Force structure with scaffolding
Instead of asking open-endedly, set up frameworks/templates.
Example:
“Answer in this JSON schema:
{‘Hook’: , ‘Body’: , ‘CTA’: }”
or
“Use the STAR method (Situation, Task, Action, Result) to rewrite this story.”
This locks the model into useful shape.
Instead of asking open-endedly, set up frameworks/templates.
Example:
“Answer in this JSON schema:
{‘Hook’: , ‘Body’: , ‘CTA’: }”
or
“Use the STAR method (Situation, Task, Action, Result) to rewrite this story.”
This locks the model into useful shape.
5. Inject constraints that mimic reality
AI shines under pressure. Add rules like word counts, personas, or tradeoffs.
Example:
“Explain transformers in under 150 words, as if you’re pitching an investor who hates jargon.”
It forces clarity.
AI shines under pressure. Add rules like word counts, personas, or tradeoffs.
Example:
“Explain transformers in under 150 words, as if you’re pitching an investor who hates jargon.”
It forces clarity.
6. Use negative prompting
Don’t just ask what you want - tell it what to avoid.
Example:
“Write a blog post on AI agents. Avoid buzzwords like ‘revolutionary’ or ‘cutting-edge.’ No metaphors.”
The absence rules sharpen style.
Don’t just ask what you want - tell it what to avoid.
Example:
“Write a blog post on AI agents. Avoid buzzwords like ‘revolutionary’ or ‘cutting-edge.’ No metaphors.”
The absence rules sharpen style.
7. Layer prompts iteratively (Chain-of-thought steering)
Treat outputs as drafts. Improve through deliberate re-prompts.
Example:
Prompt 1: “Give me 10 startup ideas in AI.”
Prompt 2: “Rank them by potential market size.”
Prompt 3: “Now expand the top 2 into 500-word business memos.”
This simulates senior-level reasoning.
Treat outputs as drafts. Improve through deliberate re-prompts.
Example:
Prompt 1: “Give me 10 startup ideas in AI.”
Prompt 2: “Rank them by potential market size.”
Prompt 3: “Now expand the top 2 into 500-word business memos.”
This simulates senior-level reasoning.
8. Mix models strategically
Different models excel at different tasks. Route accordingly.
Example:
– Use Claude for summarization.
– GPT-5 for reasoning.
– MidJourney for visuals.
– Runway for video.
Orchestrate them like an ensemble, not one model doing all.
Different models excel at different tasks. Route accordingly.
Example:
– Use Claude for summarization.
– GPT-5 for reasoning.
– MidJourney for visuals.
– Runway for video.
Orchestrate them like an ensemble, not one model doing all.
9. Prompt as negotiation, not command
AI isn’t a vending machine. You steer it like a dialogue.
Example:
First ask: “Draft a 3-paragraph op-ed on AI safety.”
Then refine: “Paragraph 2 is too soft. Make it more contrarian.”
Then polish: “Rewrite the closing line as a tweetable quote.”
This iterative shaping = expert prompting.
AI isn’t a vending machine. You steer it like a dialogue.
Example:
First ask: “Draft a 3-paragraph op-ed on AI safety.”
Then refine: “Paragraph 2 is too soft. Make it more contrarian.”
Then polish: “Rewrite the closing line as a tweetable quote.”
This iterative shaping = expert prompting.
10. Build reusable prompt libraries
Experts don’t reinvent each time. They keep modular prompts they can adapt.
Example:
– Template: “Explain {topic} in {format} for {audience} with {tone}.”
– Plug variables: “Explain diffusion models in a 60-second TikTok script for high schoolers, funny tone.”
Reusable, scalable, efficient.
Experts don’t reinvent each time. They keep modular prompts they can adapt.
Example:
– Template: “Explain {topic} in {format} for {audience} with {tone}.”
– Plug variables: “Explain diffusion models in a 60-second TikTok script for high schoolers, funny tone.”
Reusable, scalable, efficient.
I hope you've found this thread helpful.
1. Follow me @eyishazyer for more.
2. Like/Repost the thread if you can.
1. Follow me @eyishazyer for more.
2. Like/Repost the thread if you can.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
