The Anatomy of a Great AI Prompt: Six Elements Most People Skip

Before the elements, a quick clarification of what we're not doing.

• We're not filling in a template. Templates are rigid; the whole point of an AI is that it's flexible. Structure is a checklist of things to communicate, not a form to complete.
• We're not writing longer prompts for their own sake. A structured five-line prompt beats an unstructured fifteen-line one every time. Word count is not the metric.
• And we're not using clever prompt tricks. There's a subculture around "prompt engineering hacks" — magic phrases like "take a deep breath" or "think step by step." Some of these help a little in narrow cases. None of them substitute for telling the AI what you actually want.

Now, the six elements.

What it is: Who the AI is being asked to be while it responds.

Why it matters:The same question has different good answers depending on who's answering. Ask a lawyer, a friend, and a philosopher whether you should quit your job, and you'll get three genuinely different responses. Not because any of them is wrong — because their frames are different. When you don't specify a role, the AI defaults to a neutral, hedging generalist.

Example:

The therapist framing doesn't restrict the AI — it activates a specific mode of response that emphasizes emotional dynamics and asking clarifying questions rather than jumping to tactics.

Use roles when the question has genuinely different good answers depending on who's answering. Skip role assignment when the question is objective ("What's the boiling point of water?") — the role won't add anything.

What it is: What specifically the AI is producing.

Why it matters: "Help me with my presentation" is not a task. "Rewrite these three slides so they lead with the conclusion" is a task. The verb has to be specific enough that you'd know when the AI was done.

Example:

Watch what happens when the task is precise. The AI now knows what to produce, what to prioritize, and what to leave out. All three of those are decisions someone has to make — either you or the AI. If you don't make them, the AI does, based on averages.

What it is:The situation shaping the output that the AI wouldn't otherwise know.

Why it matters:Every real-world piece of writing exists in a context. The status email above is going to your VP after two weeks of the team slipping on a launch date. That's context the AI has no way to know unless you tell it. Without context, the AI writes for the average case — a generic status email to a generic manager about a generic project.

Example:

Context worth including for the status email:

• What's actually going on with the project ("We slipped the launch by two weeks; recovery plan is in place")
• What's happened recently that shapes the message ("She flagged concerns about scope creep last week")
• Any political sensitivities ("The delay involves another team's dependency, so tone matters")

You don't have to include everything — just what shapes the output. If you're not sure whether context is relevant, err toward including it. The AI can decide what to use.

What it is:What "good" looks like, ideally in priority order.

Why it matters: Most tasks have competing goals. A status email needs to be honest AND diplomatic. A cover letter needs to be confident AND humble. A code review needs to catch bugs AND not demoralize the author. Which one wins when they conflict? The AI has no idea unless you tell it.

Example:

For the status email:

Goals, in priority order:

1. Be direct about the two-week slip. Don't bury it.
2. Show that we have a concrete recovery plan.
3. Preserve the working relationship with the other team involved.

Now the AI knows what to do when honesty and diplomacy pull in different directions: honesty wins, but not so bluntly that goal #3 is torched.

What it is:Who the output is for. Not who's writing it — who's reading it.

Why it matters: A message to your VP reads differently from a message to a peer. A blog post for beginners reads differently from a post for practitioners. The AI has to make thousands of small choices about vocabulary, tone, and reference level. All of them collapse cleanly once it knows the audience.

Example:

The audience specification does more work than you'd think. Without it, the AI defaults to a mildly technical audience — the person who already reads Wired. With it, the whole register shifts.

What it is: Hard rules the output has to obey. Length, format, tone, things to avoid.

Why it matters: Constraints prevent the AI from doing the thing it always defaults to — which is producing "professional," medium-length, appropriately-hedged prose. Without constraints, everything comes out sounding like a corporate blog post.

Example:

For the status email:

• Length: under 200 words
• Format: subject line + three paragraphs
• Tone: direct, not corporate. No "at the end of the day" or "circle back."
• Avoid: hedging phrases like "we're generally on track" without specifics

The "avoid" list is the most underused. AIs have strong default habits. Telling them what NOT to do is often more powerful than telling them what to do.

Here's the "write a status email" case fully specified. All six elements are in there — see if you can spot them.

You are an experienced product manager who writes weekly status updates that senior stakeholders actually read. Your VP of Engineering has told you she skims emails and skips ones that bury the conclusion.

Write a status email to her about the Q3 launch. Two weeks ago we discovered a critical dependency on the platform team that we hadn't accounted for; the launch has slipped by two weeks. We have a recovery plan: the platform team has committed to delivering their piece by end of next week, and we've compressed the QA schedule to catch up.

Goals, in priority order:
1. Be direct about the slip. Don't bury it.
2. Show that we have a concrete plan to recover.
3. Preserve the working relationship with the platform team — this wasn't their fault, it was an integration gap on our side.

Format:
- Clear headline that names the news
- Three short paragraphs: what happened, what we're doing, one specific ask
- Under 200 words
- Use "🟢 Green / 🟡 Amber / 🔴 Red" (with the word, not just an emoji) for workstream status

Avoid corporate hedging like 'at the end of the day' or 'we're generally on track' without specifics. Write like a senior manager who respects their reader's time.

Paste that into any modern AI. What comes back will be sharper, more usable, and more like a real email than what any two-sentence prompt would generate. Not because the AI got smarter — because it now has the six pieces of information it needed to do the job.

If writing prompts this way seems like a lot of work, that's the point of tools like the one we built. Our prompt builder walks you through each of these six elements as questions, then assembles the prompt for you. You can see examples of what the output looks like, or build your own for whatever situation you're actually facing.

The reason AI feels generic when you give it generic prompts is that the AI is doing exactly what you'd do if a stranger asked you to "write a status email." You'd write something safe, hedged, and medium. You'd write the average because you don't know enough to do better.

Give the AI the same information you'd give a smart colleague — who you are, what you're doing, what the situation is, what "good" looks like, who it's for, what to avoid — and the output changes. Not because you've unlocked a hidden mode. Because you've told the AI what you actually want.

The six elements aren't magic. They're the checklist a professional communicator runs implicitly every time they write to another human being. AI is a smart generalist that responds to being briefed like one. Brief it well, and the output stops feeling generic.

Try the framework the next time an AI response feels underwhelming. Nine times out of ten, one or two of the six elements will be missing. Add them back in, and the response gets meaningfully better — often on the first try.