“Make this page better.”
That sounds like a request. It leaves a lot of decisions open.
Better for whom? Easier to read? Faster to load? More colorful? An assistant can change all three while missing the thing that bothered you.
Take a simple example: the contact button is hard to find on a phone.
Give it a job you can point to
Instead of:
Make the landing page look more professional.
Try:
On a 360px-wide screen, make the contact button easy to find in the first section. Keep the current text, colors and desktop layout. Change only the button's size and spacing. Show me the changes before applying anything else.
The second request names the problem, the allowed change and what should stay put. You can open the page on a phone-sized screen and judge the result.
It is a clearer assignment, not proof that a particular model will follow it. You still have to inspect the output.
A small fix should have a small finish line
For this example, the finish line is practical:
- The button is visible in the first section without sideways scrolling.
- Its label still says what happens when you press it.
- The desktop layout and the rest of the page remain as they were.
If changing only size and spacing cannot meet that goal, ask the assistant to explain the conflict before expanding the job.
“Improve everything” makes it hard to tell whether you are finished. One visible problem gives you something to accept, reject or revise.
This also works for writing. Instead of “make it smarter,” try “replace the jargon, keep one example and cut the introduction.” You can check each part without guessing what “smarter” means.
What is the smallest change you actually want from your next AI request?
Top comments (0)