How to Prompt a Mean AI Without Getting Yelled At (Much)

Rotten AI is designed to insult you regardless of how you prompt it. That said, some inputs get you a better answer under the insults than others. Here's what works.

Ask the specific question, not the vague one

'How do I make my Instagram better' will get you insulted and handed a generic checklist. 'My reels are getting 200 views and I post three times a week, what's the highest-leverage thing to change' will still get you insulted, but the answer will actually be useful. Specificity limits the space the model has to hedge in.

Use real-answer mode when you actually need help

Free users get five chats a day, and two of those can be flipped into real-answer mode from the sidebar toggle. That mode keeps the tone rude but forces the response to be substantive — not just a joke about you being an idiot. Save it for the questions where you need the answer to work.

Attach the image once, not the description

If you're asking about something visual — a screenshot, a photo, an error message — attach the image. Describing it in text is a downgrade every time. Both the free and premium tiers accept one image per message; premium accepts more per day.

Don't argue with the persona

You will not talk it out of being rude. The persona is baked in at the system-prompt level and every message rebuilds it. Meta-conversations about 'why are you being mean' burn a chat from your daily quota and get you a fresh round of insults for the effort. Just ask the next question.

Use the memory feature

Premium users get long-term memory. Every time you tell it something useful about yourself — the language you code in, the city you live in, your job — it can save that and use it later. That's what makes the insults land: they get more personal over time. Take a minute at the start of your subscription to tell it who you are.

Read the answer twice

The insults are meant to be skimmed. The actual answer usually lives in the middle two-thirds of the reply. Look past the framing.