The one instruction I add to almost every prompt

You give AI a task.

It gives you a confident answer.

Then halfway through, you realise it guessed one important detail wrong… and now you have to start again.

I kept running into this with proposals, workflows, data analysis, and anytime I was delegating real work to an AI agent.

So now I add one instruction to most of my prompts before I hit send.

It tells Claude when to stop and ask me a question—and when to make a sensible assumption and keep going.

The goal isn’t zero assumptions.

It’s fewer wrong turns without turning every task into a meeting.

Add the prompt below to your next AI task.

If something important is missing, the AI can ask up to three essential questions. If the missing detail is minor, it makes a reasonable assumption and gets on with the work.

You get less guessing, less back-and-forth, and better results.

You can already tell Claude:

Ask me any questions you need before you begin.

But that often creates a new problem.

The AI starts asking about everything.

Tone. Format. Length. Preferences. Tiny details it could easily decide for itself.

Now your quick task has turned into an intake form.

The opposite approach isn’t much better:

You tell the AI to begin immediately, so it quietly fills every gap itself—including the important ones.

What I actually want is a middle ground:

• Ask when the missing information could materially change the result.
• Make a reasonable assumption when it won’t.
• Keep the number of questions small.
• Once the important uncertainty is resolved, start the work.

That is exactly what this instruction does.

Add this to the end of your next prompt:

Before answering, identify any missing information or ambiguity that could materially change the result. If such uncertainty exists:

1. Ask up to three essential clarifying questions, ranked by importance.
2. For each question, briefly state the assumption you would use if I choose not to answer.
3. Wait for my response before beginning the substantive work.

If my response resolves the material uncertainty (or explicitly authorizes your assumptions) proceed without requesting further confirmation.

If no material clarification is needed, answer immediately.

Do not ask about minor details that can be handled through reasonable assumptions.

That’s it.

You can paste it below almost any task where a wrong assumption could send the work in the wrong direction.

The instruction gives Claude a simple decision:

If yes, ask up to three questions.

Not ten.

Not every question it can think of.

Only the ones that could genuinely change the answer.

It also has to rank them by importance, so you can see which uncertainty matters most.

Make a reasonable assumption and continue.

The AI doesn’t need to ask what heading style you prefer if that won’t change the substance of the work.

For every question, the AI tells you which assumption it would use instead.

So you can answer the question, approve the assumption, or simply tell it to proceed using all the proposed assumptions.

No extra round of confirmation required.

I use this whenever I’m delegating a task where the AI needs room to think—but a wrong guess would create a lot of rework.

If the audience, budget, scope, or commercial goal is unclear, those details could completely change the proposal.

Claude should ask.

If I haven’t specified whether the section headings should be sentence case or title case, it can choose.

If the trigger, owner, approval step, or final output is missing, the workflow could be designed incorrectly.

Claude should ask.

If I haven’t named every internal status label, it can make sensible suggestions and continue.

If “conversion” could mean several different events, or the date range changes the conclusion, the AI needs clarification.

Claude should ask.

If a chart title or colour hasn’t been specified, it can decide for itself.

The more work an agent can do on its own, the more expensive a bad early assumption becomes.

This instruction creates a checkpoint before the agent commits to the wrong path—without forcing you to approve every tiny decision.

Imagine you ask:

Create a proposal for a new client based on these call notes.

If the notes don’t identify the proposal’s budget or decision-maker, Claude might respond like this:

1. What budget range should this proposal be built around? This affects the scope and recommendations. If you don’t answer, I’ll assume a mid-range engagement and present optional tiers.
2. Who is the primary decision-maker? This affects the level of detail and emphasis. If you don’t answer, I’ll write for a mixed executive and operational audience.

You can reply with the missing information.

Or say:

Use your assumptions and proceed.

Claude now has permission to begin. It shouldn’t ask you to confirm everything again.

And if the original call notes already contain everything important?

It skips the questions and writes the proposal immediately.

That last part matters.

The prompt is not telling AI to always ask questions.

It is telling AI to recognise when a question is worth interrupting you for.

You don’t need to rewrite this every time.

Keep it somewhere easy to copy, save it as part of your reusable prompt template, or add it to the instructions you regularly give your AI tools.

Then write your task as normal:

1. Describe the outcome you want.
2. Add any source material, constraints, or examples.
3. Paste the clarification instruction at the end.
4. Hit send.

Claude will either ask the questions that matter or get straight to work.

Better prompting isn’t about eliminating every assumption.

It’s about surfacing the assumptions that matter.

Add it to your next prompt for less guessing, less back-and-forth, and better results.