Six Habits of Effective Prompts: A Practical Checklist

In the last article we saw that specificity is the biggest lever you control when prompting an AI. This time we turn that idea into a checklist — six habits you can apply to almost any prompt, whether you're drafting an email, planning a trip, or writing code. None of them require special tools. They just take the guesswork out of what you're asking for.

The six habits at a glance

Think of these as the things a thoughtful colleague would naturally do when handing off a task. Each one removes a different kind of ambiguity:

Six habits that remove a different flavour of ambiguity from every prompt you write.

Habit 1: be clear and specific

The model can't read your mind, so describe the output you actually want. Three small additions do most of the work: give details ("a 200-word summary for a non-technical reader"), assign a role ("answer as an experienced travel agent"), and show the shape ("reply as a bulleted list"). If a first answer comes back too long, too shallow, or in the wrong format, say exactly that — naming the problem is itself a clear instruction.

Habit 2: provide references

When you want an answer about a specific document, contract, policy, or dataset, give it to the model rather than hoping it remembers. "Using the text below, summarise the cancellation policy" almost always beats "what's a typical cancellation policy?" References are the single best defence against a confident, made-up answer — and if you ask the model to quote or cite the source, you can check the answer at a glance.

Habit 3: break big tasks down

A sprawling request like "plan my entire product launch" invites a shallow, scattered answer. Break it into stages — positioning, then channels, then a week-by-week timeline — and tackle them in turn. Smaller asks are easier for the model to get right, and easier for you to steer. It's the same reason we decompose problems in any kind of project: the whole is disproportionately harder than the sum of its parts.

Habit 4: let the model think

If you ask for a final answer immediately, you often get a rushed one. Add a phrase like "think it through step by step before giving your recommendation," and the model lays out its reasoning first — which tends to make the final answer more accurate. For anything involving logic, comparison, or trade-offs, this one line is the cheapest quality upgrade available.

Habit 5: use the right tool

Language models are remarkable, but they're not calculators, search engines, or databases. For exact arithmetic, the latest prices, or a precise record lookup, use the dedicated tool and paste the result into your prompt as context. Let each tool do what it's best at, and use the AI to interpret, summarise, or explain the result.

Habit 6: check the results

The earlier and smaller your check, the easier problems are to catch. Verify facts, re-run the numbers, and skim for claims that sound too neat. This isn't distrust — it's the same instinct that makes you proofread an important email. A two-minute review turns a plausible draft into something you can confidently stand behind.

Putting it together: before and after

Here's an everyday request rewritten with the habits applied. Notice how the "after" version isn't longer for the sake of it — every addition removes a specific guess the model would otherwise have to make:

The highlighted phrases are the habits in action: audience, specifics, tone, intent, and a length constraint.

The takeaway

Great prompting isn't about memorising clever phrases — it's a handful of habits that remove ambiguity before the model ever responds: be specific, provide references, break work down, let it think, use the right tool, and check the output. Internalise these and the quality of everything you get from AI rises. Next in the series, we'll go beyond habits into named techniques — from worked examples to step-by-step reasoning — that unlock even more.