Prompting mistakes
Fix keyword prompts and overloaded messages, show examples instead of adjectives, and correct a weak first answer in the same chat.
-
What Do You Do When ChatGPT's First Answer Isn't Good? Don't accept it, and don't start over. Level 1: Basic prompts
- ContinueThe result meets the brief
- RedirectGive one focused correction
- Pause and decideAccess or a decision is missing
-
Why Does Showing ChatGPT an Example Work Better Than Describing? An example shows length, structure and tone all at once, which adjectives can't. Level 1: Basic promptsAdjectivesopen to interpretation
- "Punchy"
- "Professional but friendly"
- "Short"
Examplessomething to match- Two past emails that worked
- The exact table layout you want
- One thing they share, named by you
-
Why Does ChatGPT Ignore Parts of a Long Prompt? Usually you've given it too many jobs and rules at once, so some get less attention. Level 1: Basic prompts
- 1Comparison table with sources
- 2You check it
- 3200-word email from the table
- 4Then the translation
-
Why Is ChatGPT Bad at Keyword Searches? Keywords don't tell ChatGPT what you're trying to do, so it has to guess and give a generic overview. Level 1: Basic promptsSearch habitkeywords
- "best crm small business"
- No situation or goal
- A generic list comes back
Assistant habitsituation plus a task- Team size, budget, tools you use
- A decision you need to make
- A recommendation with sources
Too little and too much
Prompting mistakes come in two opposite kinds. Too little is the keyword-style request with no situation, or a description of what you want made of adjectives. Too much is ten jobs in one message. Both leave ChatGPT without a clear target. OpenAI's Prompting guide suggests starting with the result you want, adding only the context that could change it, and keeping to the one or two boundaries that matter most.
What happens after the answer counts too
The fourth mistake here isn't really in the prompt at all. It's what you do with the reply. Many people either accept the first answer or give up and start over, when one specific follow-up is often all it takes. These four cards are meant to change what you do in your very next chat, so pick the one you recognize, and try it on something real today.
What is the most common ChatGPT prompting mistake?
Leaving out your situation. A request with no audience, purpose or constraints gets a generic answer. Add who it's for and how you'll use it. OpenAI's guide says that telling ChatGPT how you plan to use a result helps set its length, detail and organization.
How long should a ChatGPT prompt be?
As long as it needs to be for one clear task. A long source document is fine, but several unrelated jobs in one message aren't. OpenAI's Academy suggests breaking multi-part requests into smaller steps.
Should I give ChatGPT examples?
Yes, when you care about format or voice. One or two real samples show length, structure and tone more clearly than adjectives can. Say what to copy from them and what to leave out, so their facts don't appear in the new text.
Should I edit my prompt or reply with a correction?
Reply with a correction, in most cases. OpenAI's guide says your first prompt doesn't need to be perfect, and to ask for the specific change you want. Start a new chat if it misunderstood the whole task or keeps going in circles.