Shoutout to the dev who told me my AI prompts were too vague

I was building a simple chatbot for my small biz and kept getting nonsense replies. A friend in a coding group said my prompts were like asking a chef to cook something without saying what ingredients I have. So I started adding specific examples and constraints to every prompt. The output went from gibberish to actually useful in just three tries. Has anyone else had a big shift from just tweaking how you write prompts?

3 comments

3 Comments

logan_mitchell1mo ago

You said you gave it examples, were those fake ones you made up or did you copy paste real customer convos from your logs? That made way more difference for me once I started using actual data instead of guessing.

logan5251mo ago

Mine were real logs - figured if the AI was gonna learn my customer's weird phrasing, might as well show it the real deal instead of my terrible made-up examples.

reese_hayes711mo ago

Real logs? Seems like overkill to me. @logan525, how much difference did it actually make in the end? I just scribbled a few fake examples and my chatbot still works fine for what it does. People overthink this stuff way too much. It's not like the customer is gonna notice if the bot misses a weird phrase once in a while. Unless your logs were full of wild typos and slang that broke everything, I doubt it mattered that much. Maybe my bar is just lower though.