Robert Ta says AI outputs are mediocre because people ask without examples. Show Claude what good looks like and what to avoid, and it learns what you actually want. Include a "good examples" block with your best past work and a "bad examples" block with what flopped.
Number one mistake I see when people use AI, they don't give good or bad examples. That's called few shot prompting. In your prompt, you might say something like, I need a 30 day social media content calendar. Give me 30 ideas. If you just put that and you don't give it additional context on what makes good or what makes bad, it's just going to give you a mixture of like, okay, good and mostly bad output. These things are subjective, you have to understand that. So you have to give it what is good and what is bad. Few shot prompting is simply the act of saying, hey, give me 30 days worth of social media content. And here are some good examples of what that looks like. Write literally curly brace, good examples and curly brace and then a bunch of good examples from your highest performing content. And then put read these as bad examples. This is the stuff I don't want. Curly brace, bad examples and curly brace and a bunch of bad examples of your lowest performing content or content you see out there that is absolutely garbage. So just by using this one trick, you're going to optimize the throughput of your AI output.
