The Prompt Framework That Fixes Vague ChatGPT Answers
When a ChatGPT answer comes back generic, the instinct is to rephrase the question, try a different verb, add "please be specific," maybe throw in "act as an expert." None of that usually helps, because the problem was rarely the wording. It was that the prompt didn't contain the information ChatGPT would have needed to give a non-generic answer, and no amount of rephrasing manufactures information that was never there.
If you're still building your basic prompting habits, the Complete Beginner's Guide to ChatGPT is a good starting point before this one, which assumes you already know how to hold a conversation with it and is focused specifically on diagnosing why an answer feels flat.
Context, Task, Format, Constraints
A useful way to check what's missing from a weak prompt is to run it through four questions. Not every prompt needs all four spelled out explicitly, but a generic answer almost always traces back to one of these being absent.
Context: what does ChatGPT need to know about the situation that isn't in the request itself? Who's the audience, what already exists, what's the background it can't infer from the task alone.
Task: is the actual ask specific enough to point at one output, or does it describe a category of output? "Write a proposal" describes a category. "Write the recommendation section of a proposal for switching a 12-person team from Trello to Asana" describes a task.
Format: how should the output look? Length, structure, tone, whether it's bullets or prose, whether it's a first draft or a final version.
Constraints: what should it avoid, stay under, or definitely include? A word limit, a fact that must be in there, a phrase to never use, a deadline the reader needs to see.
Context
The background ChatGPT can't infer: who it's for, what already exists, why this matters right now.
Task
The specific ask, not the category of ask. Name the actual thing you want produced.
Format
Length, structure, tone. How the answer should look, not just what it should say.
Constraints
What to avoid, stay under, or definitely include. The edges of the answer.
Four real examples, weak to strong
Example 1: asking for feedback on a résumé bullet.
Make this résumé bullet better: "Responsible for managing social media."
”This will get you a marginally more polished sentence with the exact same problem: no numbers, no context about what "managing" meant, no sense of scale. ChatGPT can polish the words you gave it, but it can't invent the achievement behind them.
I need to rewrite this résumé bullet for a marketing coordinator role: "Responsible for managing social media." Context: I ran Instagram and TikTok for a 40-person DTC skincare brand, grew Instagram followers from 8,000 to 22,000 over 14 months, and our best campaign drove a 3x increase in site traffic for a product launch week. Task: turn this into one strong bullet that leads with the result, not the responsibility. Format: one line, resume style, starts with a strong verb, no first person. Constraint: under 25 words.
”The second version isn't cleverer, it's just not missing the achievement. Every specific number in that context box becomes something the resulting bullet can actually claim.
Example 2: asking for a business idea.
This produces a list that could have been generated for anyone, because it was, functionally, generated for anyone: no budget, no skills, no time constraint, no interest area.
Context: I'm a full-time nurse with about 6-8 hours a week free, roughly $500 to start, and I'm good at organizing and explaining things clearly to anxious people. Task: suggest 4 side business ideas that fit specifically those constraints, not a generic list. Format: for each one, one sentence on what it is and one on why it fits my specific situation. Constraint: nothing that requires ongoing in-person availability, since my nursing shifts are unpredictable.
”Example 3: asking for help with a difficult conversation.
Context: I'm a project coordinator, I've been given two additional accounts in the last month on top of my existing workload, and I haven't said anything yet because I don't want to seem like I can't handle it. My manager is direct and appreciates people who bring solutions, not just problems. Task: help me plan what to say in a 1-on-1 this week to raise this without sounding like I'm complaining. Format: a short talking-point outline, not a script to read verbatim. Constraint: I want to propose something specific, like re-prioritizing one account, not just flag the problem and leave it to my manager to solve.
”Example 4: asking for a workout plan.
This produces a generic three-day full-body split built for an imaginary average person: an hour per session, a gym's worth of equipment assumed, and no awareness of any physical limits, because none were given.
Context: I'm 35, new to structured exercise, and I have an old knee injury from soccer that rules out deep squats and lunges. I have a set of dumbbells at home and about 30 minutes, three days a week. Task: build a 3-day dumbbell routine that avoids deep knee flexion entirely. Format: list each day's exercises with sets and reps, grouped by day. Constraint: total session time under 30 minutes, no equipment beyond dumbbells and a mat.
”The difference isn't tone, it's that the second version physically cannot come back with a lunge in it, because the constraint ruled it out before generation even started. The first version had no way to know your knee existed.
Tip
Notice that none of the strong versions above are longer because they're more polished. They're longer because they contain more actual information. A five-word prompt with the right context can outperform a fifty-word prompt full of adjectives and no facts.
Using this as a diagnostic, not a template
You don't need to write out "Context: / Task: / Format: / Constraints:" labels every time, though doing that explicitly is a reasonable habit for anything that matters. The more useful version of this framework is as a checklist to run through silently when an answer disappoints you: which of the four did I actually leave out?
Most of the time it's context. People are far more likely to state the task clearly than to state the background, probably because the background feels too obvious to mention. It rarely is. The details that feel too obvious to say out loud are usually the exact details that separate a generic answer from a useful one.
Common mistake
Responding to a generic answer by asking ChatGPT to "be more specific" or "add more detail," without adding any new information yourself. It can only add invented specifics at that point, which often reads as more confident and is no more accurate.
Run a disappointing prompt back through Context, Task, Format, Constraints before you touch the wording again. Nine times out of ten, the fix is adding one sentence you'd left out, not finding a better way to phrase the one you kept.