Abstract prompting advice is easy to nod along to and hard to apply. Here’s what actually changes, side by side.
Writing a product description
Weak: “Write a product description for my mug.” Better: “Write a 50-word description for a $40 ceramic mug, sold to home-office workers who care about minimalist design. Warm but not cutesy. No exclamation points.” The second version hits on the first try instead of guessing at length, audience, and tone.
Debugging code
Weak: “This doesn’t work, fix it.” Better: “This should return the average of the list, but returns None for empty lists instead of 0. Here’s the function and a failing test.” Naming the exact expected versus actual behavior turns a vague ask into something the model can actually reason through.
Summarizing a long document
Weak: “Summarize this.” Better: “Summarize this for an executive deciding whether to approve the budget in it. Five bullets max, lead with the number that matters most.” Naming who it’s for tells the model what to emphasize.
Matching a specific tone
Weak: “Make this sound professional but friendly.” Better: paste two sentences you actually like and say “match this.” A concrete example beats an adjective every time.
Same move each time: swap the vague ask for concrete constraints. See our full prompt engineering guide for the template behind it.
See Anthropic’s own prompt engineering documentation for more depth.
