← All guides · Using AI well
How to write AI prompts that get better answers
Most bad AI answers come from short questions. The model fills in everything you didn't say with the most average guess. Saying a little more, in the right places, fixes most of it. No magic words needed.
The short version
Tell it who it's for, what you already have, what a good answer looks like, and what to avoid. That's four short sentences, and it beats any "act as a world-class expert" opener.
1. Say who the answer is for
"Explain photosynthesis" gets a textbook paragraph. "Explain photosynthesis to a 12-year-old who likes football" gets something a 12-year-old will actually read. The audience decides the vocabulary, the length, and the examples, and the model can't guess it.
2. Paste what you already have
Don't describe your essay, paste the paragraph. Don't describe the error, paste the error message and the ten lines around it. The model works far better from real material than from your summary of it. (If it's code, check it for passwords and keys first; see is it safe to paste code into AI?)
3. Describe the shape of a good answer
- Length: "three bullet points", "under 100 words", "one paragraph".
- Format: "a table with columns X and Y", "a numbered checklist", "just the code, no explanation".
- Tone: "casual, like a text to a friend", "formal enough for a teacher".
If you have an example of what you want, show it. "Write it like this one" with a sample is the single most effective instruction there is.
4. Say what to avoid
"Don't use bullet points." "Don't suggest buying anything." "Skip the introduction." Models have habits, and telling them which ones you don't want is quicker than editing them out afterwards.
5. Ask for one thing at a time
"Fix my essay, make it shorter, check the facts, and add a conclusion" gets four half-done jobs. Ask for the structure first, then the wording, then the facts. Each step is easier to check, and you can stop when it's good enough.
6. Push back instead of starting over
The first answer is a draft. "Shorter." "Less formal." "The second point is wrong, it's actually X." "Give me three different versions." Follow-ups keep the context you already built, which is usually faster than writing a perfect prompt from scratch.
Before and after
Before: write an email to my teacher about the deadline
After: Write a short, polite email to my chemistry teacher asking for two extra days on Friday's lab report. I was sick Tuesday and Wednesday. Keep it under 80 words, don't over-apologise, and end by offering to show the half I've finished.
Same request. The second one needs no editing.
What prompting can't fix
A better prompt gets a better-shaped answer, not a more truthful one. If the model doesn't know a fact, wording won't make it know it. For anything you'll rely on, read how to tell when an AI answer is wrong.
Doing this once instead of every time
If you keep typing the same preferences ("keep it short", "I'm a student", "answer in Persian"), put them in Sevoria's Settings under Personalization ("What should Sevoria know about you?"), and every chat starts with them. Sevoria's memory can also keep a few facts about you between chats, and you can see, edit, or erase each one.
FAQ
Do I need to be polite to the AI?
It doesn't need it, but clear and polite prompts tend to also be clear and specific ones, which is what actually matters.
Are long prompts always better?
No. Longer is better only when the extra words add information the model couldn't guess. Padding just buries the real request.
Should I use "act as an expert" prompts?
They help less than people think. Telling it the audience, the material, and the format does more than telling it to pretend to be someone.
Try a better prompt Writing with Sevoria
Was this useful?
Thanks, noted.