What's the problem
You open ChatGPT or Claude, write 'write a post about motivation' — you get three paragraphs about 'the importance of goals' and 'small steps to your dream.' You write 'give sales advice' — you get a list of eight points that any blog could have written in 2018.
This is not a bug. It's math. The language model predicts the next token based on what has appeared most frequently in training data. A vague request activates the most common patterns. And the most common patterns on the internet are clichés.
Good news: this can be fixed. You don't need to learn 'prompting' in an academic sense. Just understand five mechanisms — and the quality of responses will increase exponentially.
Reason 1. No context about you
When you write 'write a text about investments,' the model doesn't know: are you writing for beginners or professionals? Are you selling a course or running an educational blog? Do you want provocation or dry analytics?
Without this, the model chooses the 'average' — what will suit everyone and won't offend anyone.
Reason 2. No role and voice
The language model is a chameleon. It adapts to any voice, but without instruction, it chooses a neutral 'formal' style. This is where 'Certainly, here are some key aspects' comes from.
The solution is simple: assign a role at the beginning of the request. Not 'help me,' but 'you are a marketer with 10 years of experience in B2C.' This is a default instruction that filters out template constructions.
Want to write prompts that work the first time?
Prompting Mastery — a system of 50+ templates for real tasks: texts, analysis, code, marketing. Stop guessing — start getting consistent results.
Learn Prompting Mastery →Reason 3. Request is too broad
'How to improve sales' — a dissertation question. 'How to formulate an offer for a cold email to the director of a small manufacturing company so that he responds' — that's a task.
The narrower and more specific the question — the more precise the answer. Don't be afraid of long requests. Length doesn't scare the model; it helps narrow down the space of possible answers.
Reason 4. No example of an undesirable response
One of the most underrated techniques is to show the model what you don't want to receive. This works more powerfully than describing the desired outcome because it cuts off entire classes of templates.
Reason 5. No iteration
Most people ask a question once and accept the first answer. But the first answer is a draft. The model doesn't know what's more important to you: length, tone, structure, or specific details.
After the first answer, say: 'Make it a third shorter and add one specific example from real practice.' This is how good prompt engineers work — iteration, not trying to guess the perfect request on the first try.
Algorithm for breaking out of template responses
- Assign a role — who speaks, with what experience, in what style
- Describe the audience — who reads, what they already know, what they fear
- Narrow down the task — one specific situation, not 'generally about X'
- Prohibit templates — literally write what is not needed
- Iterate — ask to rewrite with specific adjustments
Example of a complete prompt
This request generates an email that can be sent immediately. The model knows who is writing, who it is writing to, and what is absolutely not allowed — templates have no place here.
FAQ
Why does ChatGPT always give the same answers?
The model is trained on patterns from the internet. Without specific context, it provides an averaged response — the one that appeared most frequently in the training data. Add context — you'll get specificity.
How to make AI give specific answers?
Add context: who you are, who you are writing for, what tone is needed, what you've already tried, what you don't want to receive. The more precise the request — the more specific the answer.
How does Claude differ from ChatGPT in the quality of answers?
Claude maintains context better in long dialogues and less frequently falls into templates when working with detailed system prompts. However, the issue of banality exists for both models without proper prompting.
What is a system prompt and why is it needed?
A system prompt is an instruction you give the model before the main request. It describes the role, tone, and limitations. This has a stronger impact on quality than the formulation of the question itself.
How to check if a prompt works well?
If the answer can be inserted into any other task without changes — the prompt is weak. A good answer contains details specific to your situation and audience.
50+ prompt templates for real tasks
Prompting Mastery — breakdowns of prompts for texts, marketing, data analysis, code, and strategy. Take what's ready, adapt it to yourself, and achieve results.
Open Prompting Mastery →