You type a question. ChatGPT comes back generic, off-topic, half-wrong, or way too long.
Your instinct says the model is dumb. The model is not dumb. The prompt is missing the four or five details that would have made the answer good. Each missing detail is a place the model had to guess. The guesses pile up. You blame the model.
Here are the five fixes that close most of those gaps. Each one takes 30 seconds.
Fix one: ban vague success criteria
Bad:
Make it sound professional.
"Professional" means something different to every reader. The model picks whatever the median definition is in its training data.
Good:
Use formal business language. No slang. Include technical terminology where appropriate. Active voice. End with two concrete next steps.
You can argue with each of those rules. You can also measure whether the output follows them. That is the difference.
Fix two: assign a role
Bad:
Explain quantum computing.
The model has no idea who you are or what level of detail you can handle. It picks an average.
Good:
You are a physicist explaining quantum computing to undergraduates who know basic linear algebra. Use one analogy from classical computing that lands for that audience. No more than 400 words.
The role does most of the work. Audience, depth, tone, length all start to fall into place once the model knows who is in the room.
Fix three: specify the output format
Bad:
Give me a response about AI.
You will get prose, headers, a bulleted list, or some combination. Whatever the model felt like. If you needed JSON for a downstream script, you are now reformatting by hand.
Good:
Return JSON with these fields:
- definition: max 100 words
- key_concepts: array of 5 terms with definitions
- real_world_apps: array of 3 examples with brief descriptions
- glossary: object mapping terms to definitions
The model now has a target shape. The output usually compiles on the first try.
Fix four: give the model one example
This is the single highest-leverage move on most prompts. Models do much better when shown one good example of the output you want.
Bad:
Translate these support tickets to Spanish.
Good:
Translate these support tickets to Spanish.
Example: Input "My account is locked, please help." Output "Mi cuenta está bloqueada, por favor ayuda."
Now translate the following: ...
The example is doing more than illustrating tone. It tells the model what register to use, how literal to be, whether to keep "please" as "por favor," and where to break sentences. None of that has to be written as a rule.
Fix five: stop overloading
Bad:
Write a blog post that explains AI to beginners, making sure to cover transformers, CNNs, RNNs, attention mechanisms, gradient descent, backpropagation, loss functions, learning rates, overfitting, underfitting, regularization, dropout, batch normalization, data augmentation, transfer learning, fine-tuning, inference time, training time, GPU requirements, CUDA optimization, model quantization, model compression, knowledge distillation, and ethical considerations...
The model now thinks you want a textbook. It gives you a textbook. Shallow on every topic.
Good:
Explain AI to complete beginners in 600 words. Cover only:
1. What AI is (definition)
2. How it learns (training vs inference)
3. Two real-world applications
4. One common misconception to avoid
Use simple analogies. No jargon beyond "model" and "data."
Scope matters more than coverage. A 600-word post on four ideas beats a 2,000-word post on twenty-four every time.
A starter framework
If you want a single template to use as a starting point:
ROLE: who the model should pretend to be
TASK: the single most important thing you want
INPUTS: any context the model needs
RULES (must follow):
- hard constraint 1
- hard constraint 2
- hard constraint 3
FORMAT: exact output shape
EXAMPLE OF WHAT GOOD LOOKS LIKE: one example
Most prompts that follow this structure get usable output on the first attempt. Most prompts that skip three or more sections do not.
A faster way to check
Paste a prompt into FixMyPrompt. The rubric scores all five of the fixes above and tells you which one is bleeding the most signal. The rewrite fills in whichever sections you left blank.
Three free reports per day. No signup.