The quality of a response depends more on the request than on the model choice. These seven rules work with any chatbot.
1. Treat the model as a new team member
It will do exactly what you asked and nothing more. Everything that should be in the result needs to be specified.
2. Picture the ideal answer in advance
Decide what it looks like: a table or text, how many points, what tone. Then you can describe it and quickly understand what's missing.
3. Provide context
The model doesn't know who you are, who the text is for, or what's already been done. The more detailed your background information, the less it speculates.
4. State the final goal
"Write a letter" and "write a letter that will make the client agree to a call" produce different results.
5. Show examples
Two or three good samples explain the format and style better than a long description.
6. Break the task into steps
Ask the model to proceed in order: first analyze the conditions, then suggest options, then pick the best one. On complex tasks, this noticeably improves accuracy.
7. Assign a role
"You're a news editor," "you're a financial analyst," "you're a real estate specialist." The role sets the vocabulary and level of detail.
Example
Before: "Write a post about our service."
After: "You're the editor of a Telegram channel for developers. Write a post up to 600 characters about a new free APIs section in the catalog. Goal — get people to click the link. Calm tone, no exclamation marks. Here are two previous posts as examples."