The Pattern Behind Generic Output
Ask "write me a marketing email" and you'll get a marketing email — generic, technically correct, and usable for almost no one specifically. The model isn't failing; it's doing exactly what a vague prompt asks for, which is to produce something plausible for an unspecified audience, tone, and goal. Generic input produces generic output, every time, regardless of which model you're using.
The Four Missing Inputs
Most prompts that disappoint are missing at least one of four things:
- Audience — who is this actually for, specifically
- Constraint — length, format, tone, what to avoid
- Context — relevant background the model can't otherwise guess
- Example or reference point — what "good" looks like in this case
Adding even two of these usually transforms the output dramatically.
Before and After
Generic: "Write a LinkedIn post about remote work."
Specific: "Write a LinkedIn post for mid-career software engineers who are skeptical that remote work hurts career growth. Keep it under 150 words, conversational tone, no corporate buzzwords like 'synergy' or 'leverage.' Open with a contrarian statement, not a question."
The second prompt doesn't just produce a better-written post — it produces a post that could only have been written for that specific situation, because the model now has actual constraints to satisfy instead of guessing at all of them.
Give It Context It Can't Invent
A model has no idea what your company does, what tone your brand uses, or what's already been tried unless you tell it. Pasting in 2-3 sentences of relevant background — "this is for a B2B SaaS company that sells to HR teams, our existing content is fairly formal" — does more to improve output quality than almost any clever phrasing trick.
Use Examples as a Steering Tool
If you have even one example of writing you like — a previous email, a competitor's post, a paragraph from somewhere else — include a short excerpt and say "match this tone and structure." This is more reliable than describing a tone in the abstract ("make it punchy and confident") because the model is matching a concrete pattern instead of interpreting an adjective.
Iterate in the Same Conversation, Not From Scratch
When the first response is close but not right, the fastest fix is rarely a brand-new prompt. Instead, respond directly: "Good structure, but make the opening line less formal and cut the third paragraph." The model retains the context of what it already produced, so corrections compound rather than starting the guessing process over.
A Simple Pre-Send Check
Before sending any prompt, ask: could this exact wording produce a useful response for a completely different person or company? If yes, it's still too generic. Add the specific detail that would make the answer wrong for anyone else, and that's usually the detail that makes it right for you.
Nohaya's PromptAi collection includes prompt templates already built around this principle — each one includes the context and constraints baked in, so you can see the pattern and adapt it to your own situation.