The gap between a generic AI answer and a great one is usually one line. These are the six worth memorizing. Each works in Claude, ChatGPT, or Gemini.
You don't need a longer prompt or a clever system. Most of the time you need one instruction that points the AI at what you actually want. The six below do exactly that, and you can paste them as they are.
Read them, pick the one that fixes your most common frustration, and use it until it's a habit.
When an answer goes over your head, or a topic is new to you, put ELI10 in front of your question.
Why it works. It tells the AI to drop the jargon and explain simply. The fastest way to understand something without feeling behind.
ELI10 [your question]
Before the AI answers anything that needs your details, make it ask you.
Why it works. A vague ask gets a generic answer. The questions pull in your real situation, and they often surface what you forgot to say.
Before you answer, ask me 5 clarifying questions.
Instead of describing the style you want, show the AI one you like and tell it to match.
Why it works. One real example carries more than a paragraph of instructions. It copies the shape, tone, and level you point at.
Here's an example of what great looks like: [paste it]. Match it.
After a good conversation, have the AI save what matters about you so you stop re-explaining yourself.
Why it works. It carries that context into future chats, so it starts already knowing your business instead of meeting you cold every time.
Based on this conversation, what are the most important things you should remember about me and my goals? Save them to memory.
AI tends to agree with you. Point it the other way.
Why it works. One line turns the AI from a cheerleader into a second set of eyes, and that's where the useful feedback lives.
Find my blind spots and weak assumptions.
For a task with a clear finish line, tell the AI the target and don't let it stop short.
Why it works. It keeps the AI working and checking its own result against your standard, instead of handing you a first draft and calling it done.
Don't stop until [outcome], verified by [a check the AI can run].
These are the power-user versions. A little more setup, a lot more payoff, worth it once the first six feel easy.
If you keep giving the AI the same instructions, save them once as a skill.
Why it works. A skill stores your way of doing a task so the AI follows it every time, with no re-explaining. It's the step that turns a one-off into a system you reuse.
Create a skill so I don't have to repeat what I said.
The real power version of number 6. Give the AI the actual rules and make it check every one before it answers.
Why it works. Spelled-out rules plus a self-check means the AI does the proofreading you'd normally do, and you get the version that already passes.
Don't stop until [your rules, like: under 150 words, opens with the client's problem, one proof point, a clear next step]. Check every rule before showing me.
Push past "find my blind spots" into a real test of your idea.
Why it works. It turns vague doubt into a checklist. Now you know exactly what to shore up before you commit.
What would you need to see to be convinced this plan works?
AI can flatter its own work. Add a grader.
Why it works. A grading pass catches what one pass misses. Even better, run it in a fresh chat so it judges the work cold, not as the same chat that just wrote it.
Grade this answer from 1 to 10 against [my goal], list what's missing, then fix it until it scores 9 or higher.
If answers feel generic, start with number 2 or number 3. If you want the AI to push back instead of agree, number 5. Use it every day for a week until it's automatic, then add the next. One line at a time is how this actually sticks.