Practice10 min read

When AI Isn't Working (Diagnostic Patterns)

The four reasons prompts fail

When AI gives you a bad answer, it's almost always one of four things. Knowing which one is the difference between fixing it in 30 seconds and giving up.

Vague instruction — AI doesn't know what you actually want

Missing context — AI doesn't have what it needs to give a good answer

Wrong format requested — AI is producing the right content in the wrong shape

Outside its reliable zone — you're asking about specific facts, recent events, or niche domains where AI hallucinates

The 30-second diagnostic

When a prompt fails, run this checklist before rewriting:

Reread the prompt as if you're seeing it for the first time. Is what you want actually clear? Or did you assume AI would know something it can't?

Did you specify format? Length, structure, tone? If not, that's usually the first fix.

Did you give enough context? AI defaults to generic when it's missing your specific situation.

Are you in a trust-low zone? (Specific facts, current events, citations.) If yes, source-grounding or verification is the move, not better prompting.

Most prompts fail for reasons 1-3, which are fixable. Reason 4 is a tool-limit problem, not a prompt problem.

Save your failures. When a prompt didn't work the first time and you figured out the fix, write down both. A library of "what didn't work and why" trains your judgment faster than any course can — including this one.

Practice Exercise

Here's a prompt that produced disappointing results: > "Help me write a difficult email to a vendor." The AI's response was a generic email template with placeholder language like "I hope this message finds you well" and "please don't hesitate to reach out." You actually need to email a vendor whose work quality has dropped recently and tell them you're considering ending the contract — without burning the relationship in case things improve. Diagnose what was wrong with the original prompt and write a properly-improved version that would have worked on the first try.

No pressure — just give it your best shot! Write a prompt for the scenario above and our AI will give you friendly, specific feedback on how to improve.

0 / 5,000

(try writing your own prompt first!)

You did it.

You've finished "Advanced Prompting Techniques." Here's what you can now do that you couldn't 65 minutes ago:

Tell signal from word count — advanced prompting is precision, not length

Invoke chain-of-thought when you need AI to show its reasoning

Use few-shot examples to anchor style or pattern that's hard to describe

Iterate effectively — refine, focus, or reset depending on what's actually wrong

Set up system prompts that turn AI from generalist into something tuned to your work

Build a template library for the prompts you write more than once

Diagnose failed prompts in 30 seconds instead of giving up

The most important thing now: pick one technique and use it tomorrow. Don't try to use all six at once. Start with whichever one matches the prompts you're writing most often — chain-of-thought if you make multi-step decisions, few-shot if you write style-sensitive copy, templates if your AI work is repetitive. The point of an advanced course isn't to use everything at once. It's to add the right tool to your hand at the right moment.