Why Your Prompts Aren't Working (And How to Fix Them)

You asked the AI to help write a proposal. It gave you something generic, vague, and useless. You closed the tab.

The problem wasn’t the AI. It was that you knew everything about the proposal - the client, the stakes, the tone you needed, what had already been tried - and gave it none of that. The AI had no idea who you are, what you’re working on, or what “good” looks like in your context. So it produced something that technically answered your question and was completely wrong for your situation.

This is why most prompts fail. Not because AI is bad at the task. Because the person writing the prompt already knows the context, so they forget to include it.

The context gap

Here’s the thing about working with an AI: it only knows what you tell it in this conversation. No background. No memory of the last thing you worked on together. No idea what industry you’re in, who your client is, or what tone fits your brand.

When you ask a colleague for help, you don’t need to explain everything from scratch. They already know the context. They’ve sat in the same meetings, seen the same emails, absorbed the same culture.

The AI hasn’t. Every conversation starts from zero. So if you write a short prompt and get a generic answer, that’s not the AI failing - that’s the AI doing the best it can with almost nothing to work with.

The fix is simple: give it the context it doesn’t have.

The four ingredients

A good prompt doesn’t need to be long. It needs to have the right parts. There are four:

Context. Who you are, what you’re working on, what matters in this situation. Not a biography - just the one or two things that would change how someone approached the task if they knew them.

Bad: “Help me write a follow-up email.” Better: “I’m following up with a CFO after a first meeting where they were interested but mentioned budget pressure.”

Role. What job you’re asking the AI to do. Not “pretend to be an expert” - just a clear statement of the task. Editing, drafting, summarising, brainstorming, pushing back, explaining.

Bad: “Look at this document.” Better: “Summarise the three main risks in this document in plain language, for someone who hasn’t read it.”

Format. How you want the output. Bullet list. Email. Table. Executive summary. Single sentence. If you don’t say, the AI will choose - and it often chooses wrong.

Bad: “Give me some ideas.” Better: “Give me five ideas as a numbered list, one sentence each.”

Constraint. What to avoid, what limits apply, who the audience is. This is where the real specificity lives. “Don’t include pricing” is more useful than “keep it professional.” “The reader is not technical” is more useful than “keep it simple.”

Not every prompt needs all four. A quick question might only need context. A complex output needs all of them. The gap between a prompt that frustrates and one that works is almost always one of these missing.

Debug your prompt

When the output is bad, don’t start by trying clever wording. Check which ingredient is missing.

What context did I assume? If a colleague would need background before helping, the AI needs it too. The client, the audience, the stakes, the previous attempt - include the part that changes the answer.

Who is this for? “Make it simple” is vague. “The reader is a CFO who has not seen the data model” is useful. Audience changes vocabulary, structure, and what gets left out.

What format did I leave implicit? If you want an email, ask for an email. If you want a table, ask for a table. If you want three options with tradeoffs, say that.

What should it avoid? Constraints do more than narrow the answer. They teach the AI what would make the output unusable: don’t mention pricing, don’t make legal claims, don’t sound excited, don’t invent metrics.

Did I show it an example? One good example often beats a long instruction. If you have a previous proposal, email, memo, or analysis that worked, paste it and ask the AI to match the structure.

Here’s the simplest debugging move:

This output is too generic. The missing context is [what you left out].
The audience is [who will read it].
The output should be [format].
Try again, and avoid [specific thing that made the first version unusable].

Most failed prompts don’t need polish. They need the information you forgot you were carrying in your head.

Before and after

Here’s what this looks like in practice.

Weak prompt:

Write a proposal for a new project.

The AI will produce something structured and completely unusable - because it doesn’t know what the project is, who the client is, what the budget is, or what “proposal” means in your context.

Strong prompt:

I need to write a one-page proposal for a six-week data audit for a mid-size logistics company. The audience is their Operations Director, who is sceptical about AI but has agreed to a scoping conversation. The proposal should focus on what we’ll deliver and the business risk of not acting, not on technical methodology. Tone: direct and commercial, not consultancy-speak. Format: three short sections with headers - What we’ll do, What you’ll have at the end, What it costs.

Same task. Completely different starting point. The second prompt doesn’t just tell the AI what to write - it tells it who the reader is, what they care about, and what “good” looks like.

Templates to start with

Copy these and fill in the brackets:

I need help with [specific task].

Context: [who you are, what you're working on, any relevant background]
Goal: [what a great output looks like]
Format: [email / bullet list / one paragraph / table / etc.]
Constraint: [length limit / tone / what to avoid / who the audience is]

For tasks where you need feedback or critique:

Here is [document / draft / plan].

I want you to [review it for X / push back on Y / identify gaps in Z].
Focus on [the specific thing you care about most].
Tell me what's wrong before you tell me what's right.

The iteration mindset

One prompt is rarely enough. That’s not a failure - it’s how this works.

If the first output isn’t right, the instinct is to start over with a slightly different prompt. That’s the wrong move. The better move is to say what was wrong and why.

This isn't quite right. The issue is [specific problem with the output].
What I actually need is [clarification of what was missing or wrong].
Try again with that in mind.

That single message - specific, diagnostic, directing - is worth more than ten fresh attempts at the opening prompt. The AI keeps the context of what’s already been tried and corrects from there.

Treat it like a back-and-forth with a capable colleague, not a form you’re submitting. The output improves when you engage with what came back, not when you ignore it and start again.

The one thing not to do

Don’t ask vague questions and expect specific answers.

“Help me with my marketing” gets you a generic response about target audiences and brand voice. “Review this email for a cold outreach to a Series A founder and tell me why they’d delete it in the first three seconds” gets you something useful.

Specificity in equals specificity out. The prompt is the brief. Write it like one.

---

The next step after this is setting up context that persists - so you’re not re-explaining yourself every session. Memory, Project Files, and Retrieval explains which kind of context belongs where.