Deep dive · Jump here anytime when you want more on this idea
Your First Productive AI Conversation
Stop testing. Start asking real questions
After this, you'll be able to turn a real task into a specific request with context and a success criterion, and read the response critically before refining it once.
Before you start
Before diving in, complete AI Is Not a Search Engine so you have a clear mental model of prediction versus search.
The idea
Most people's first AI session is a test. They type something vague to see if it's smart, get a vague answer, and conclude it's overrated. This is backwards. AI tools are completion engines. They predict the most useful continuation of what you give them. Give them nothing specific, get nothing specific back.
| Before | After | |
|---|---|---|
| Habit | Guess from a loose request | Use the lesson move |
| Work move | Skip Your First Productive AI Conversation | Apply Your First Productive AI Conversation |
| Check | No clear proof | Pass the lesson check |
The after column is the lesson target.
The prompt isn't the hard part. Knowing what you actually want to ask is. Spend 30 seconds writing down your actual goal before you open the chat.
Here is the before and after: A marketing manager opened Claude and typed 'help with Q3 report.' The response was a generic outline. She closed the tab. Two days later she tried again, this time spending 30 seconds writing down what she actually needed: a one-page summary for her VP, focused on customer acquisition, with a confident tone. That second session produced something she sent the same afternoon. The model did not change. Her clarity did.
What makes a good first question: it's specific, it has context, and it has a success criterion. 'Help me write an email' is a test. 'Help me write a 3-sentence email declining a meeting request from a vendor, keeping it warm but firm' is a real question. You'll know you've crossed from L0 to L1 when you stop testing and start using.
Try it (5 min)
Watch out for
- Typing search-engine keywords ('email decline meeting') instead of a real sentence. Keywords get keyword-quality answers.
- Skipping the success criterion. Without it, Claude guesses at length, tone, and format and almost always guesses wrong.
- Treating the first response as final. The first answer is a draft. One specific follow-up beats starting over.
- Testing the tool with throwaway prompts instead of using it on a task you actually need to do today.
Paste this into Claude
I need help with [one real task you have today]. Here's the context: I am [your role or who you are]. The situation is [what's happening, why this matters now, who the audience is]. A good result for me would be [describe what success looks like: length, tone, format]. Please draft it, then ask me one clarifying question if anything is unclear before you finalize.
What good looks like
- Your prompt names the task, your role, the audience, and what 'good' looks like in concrete terms (length, tone, or format)
- The response is something you could actually use or share with one round of light editing, not a generic template
- You read the output critically and named at least one thing that was right and one thing that was off before refining
When this breaks
- Breaks when you have not decided what you actually want before opening the chat, because the prompt becomes a vague test instead of a real request and the model has nothing concrete to predict against.
- Breaks when the task depends on information only you have (calendar, inbox, internal context) and you have not pasted it in, because the model cannot read your mind or your other tabs.
AI can help with this
Use AI to apply this lesson to your current work. Share your situation, ask for one concrete next step, and check the answer against this test: Write one prompt for a real task today that includes role, situation, and a clear success criterion, then use the response with no more than one round of refinement.
You can now
Write one prompt for a real task today that includes role, situation, and a clear success criterion, then use the response with no more than one round of refinement.
Key takeaways
AI completes your thought, so give it a real one. Specificity plus context plus a success criterion turns a test into a useful answer.
- AI completes your thought. Give it a real one, not a search-engine keyword.
- Specificity plus context equals useful output. State your role, the situation, and what good looks like.
- One real task beats ten test prompts. Use the tool on something you actually need to ship today.
- The first response is a draft. Read it like an editor and refine once before walking away.