Why ChatGPT and Claude Give You Weak Answers: Context and Examples Are Missing
By Anis Hammouche·October 19, 2026
A weak answer almost always comes from a prompt that is too short. The model does not know your trade, your client, the tone you want or the expected form. It fills those gaps with the most likely answer for someone who had not said anything. Add context, a precise goal and an example, and the quality changes clearly.
The model fills the gaps
Anthropic describes Claude as a brilliant but new employee who does not know your standards or habits. The more precisely you explain what you want, the better the result. The image holds for ChatGPT and other assistants too.
When you write "write a description for my offer", the AI does not know whether you are a graphic designer, a coach or a plumber, or whether you address a small business or a private individual. So it produces a text that could suit anyone and suits nobody.
The three gaps
| Gap | What the AI fills in for you | What to add |
|---|---|---|
| Context | An average reader, a generic trade | Your trade, your client, your situation |
| Goal | A "complete" text that acts on nothing | What the text must do: convince, inform, decide |
| Example | A neutral style, bland wording | A text you like, or a model of the format |
Before and after
Here is a typical prompt, followed by its corrected version.
Before:
Write a description for my website creation offer.
After:
I am a freelance graphic designer. I help local craftspeople get a
simple, fast website that appears on Google. My clients are not
comfortable with digital tools and mainly want to be understood.
Write the description of my "showcase website" offer for my home page:
- 3 sentences maximum
- direct tone, informal "you"
- start with the client's problem, not with my service
- end with one clear action
Here is an example of the tone I like (do not copy it):
"Is your phone quiet? People are looking for you and cannot find you."
The second version gives the AI a reader, a goal, a length constraint and an example of tone. The result will be shorter and more precise, and closer to what you would write yourself.
A second example, for research
The same problem appears when you ask for research. A prompt like "give me the market trends" does not say which market, which area or what you will use it for. The result will be broad and of little use.
Before:
What are the trends in the joinery market?
After:
I am a joiner in Lille, mostly making custom kitchens for private
clients. I want to know what is changing in my client requests, so I can
adjust my quotes and my website.
List 5 changes observed, separating what comes from sourced figures
from what is your own impression. For each change, give the source
I should check.
The second version states who you are, what you will do with the result, and what you expect for each point. It also asks for a separation between facts and impressions, which avoids a common mistake (see the article on hallucinations). To see what still gets read line by line even with a good prompt, read how to check every price and every condition in a quote.
How to apply the method
- Run the colleague test. Show your prompt to someone who does not know your situation and ask them to follow it. If they are lost, the AI will be too. Anthropic recommends this check in its documentation.
- Give the context before the task. Who you are, who you write for, what has already been done.
- Say what you want to get. A text, a list, a table, a decision. Specify the format and the length.
- Add an example. Examples steer format, tone and structure. One is often enough to start. Add more if the result stays off.
- Iterate. If the answer is wrong, say exactly what is wrong: "too formal", "you invented this figure". Each correction becomes a reusable instruction.
The same documentation notes that explaining why an instruction matters helps the model apply it better. "Stay under 3 sentences, because the text appears under a photo on mobile" gives more to work with than "be short".
Frequently asked questions
Do you need very long prompts to get a good answer?
No. You need precise prompts. A few lines of context and a clear goal are worth more than a page of vague instructions. Useful length is the length that removes assumptions.
Can an example be copied as it is?
It can, and that is often the risk. State clearly that the example is a tone reference, not text to reuse. Otherwise the AI copies the words, not the spirit.
Do you need to rewrite the prompt for every conversation?
For a recurring task, keep your context in a document and paste it at the start. You save time and get stable quality. See also AI as an intern you need to brief.
Stuck on a specific case? Write to me via the contact page.
