Prompting, the six-move method
How to write instructions that produce useful results first time. The six-move method, the techniques that really work in 2026 and ten ready-made prompts for work.
Mattia PanattaUpdated 15 October 202615 minBeginner
In short
- Context beats magic words. A good prompt says who the model should be, what it must do, for whom, with which constraints and in what format.
- The most underrated technique: let the AI interview you before it writes.
- Don’t chase the perfect prompt: iterate. The second and third requests are where the value is.
- To avoid sounding “AI-written”, give examples of your style and a list of things to avoid.
Why the prompt matters (and why it matters less than before)
2026 models understand vague requests far better than two years ago. “Write an email to a client” already produces something decent. But decent is the problem: an average, generic text, the same anyone else would get. The difference between an average result and a genuinely useful one almost always lies in information only you have: the client, the context, your style, the real goal.
As we saw in the guide on how an LLM works, the model works on what’s in the context. The prompt is how you put it there.
The six-move method
Six elements: use all of them for important things, some for quick ones.
- Role. Who the model should be. “You are an employment consultant writing for small-business owners.” The role steers vocabulary, detail and viewpoint.
- Goal. What it must produce, with a clear verb. “Write”, “compare”, “summarise in”, “propose three alternatives”.
- Context. What the model can’t know: who the reader is, what happened before, why you need it, what you’ve tried.
- Constraints. Length, tone, things to avoid, sources to use, what not to invent.
- Format. How you want the answer: table, list, email with subject line, three draft versions, code.
- Examples. One example of a result you like beats ten adjectives. Optional, but it’s the move that improves style most.
Try building one. The meter shows how complete it is:
Five techniques that really work
1. Get interviewed
Instead of struggling to write all the context, ask the AI to draw it out of you. Many experts call it the most effective technique, and it’s the simplest.
I want to prepare [what]. Before writing anything, ask me the questions you need to do it well, one at a time, at most 7. When you have enough information, summarise what you understood and ask me to confirm.
2. Separate instructions from data
When pasting text to process, delimit it clearly: triple quotes, dashes, or labels like “TEXT TO ANALYSE:”. The model won’t confuse your document with your instructions, and it’s also a first defence if the text contains hidden instructions.
3. Ask for alternatives, not “the” answer
“Propose three versions: one formal, one direct, one warmer.” Choosing between options is easier and faster than fixing a single answer, and helps you figure out what you actually want.
4. Make it critique before delivering
Reread your answer as if you were [the recipient / a demanding reviewer]. List the 3 most important weaknesses, then rewrite the answer fixing them.
5. Iterate one thing at a time
The first answer is a draft. The best corrections are specific: “30% shorter”, “drop the first paragraph”, “tone is too salesy”, “add a numeric example”. Avoid a generic “improve it”.
How not to sound AI-written
Generated texts have recognisable tics: emphatic openings, triplets of adjectives, phrases like “in today’s ever-evolving landscape”, conclusions that recap everything. People spot them and trust collapses.
What works:
- Give a sample of your style: paste two emails you wrote.
- Give a blacklist: “Do not use: crucial, delve, landscape, unlock, game-changer, not only… but also, exclamation marks.”
- Ask for short, concrete sentences, with examples and numbers instead of adjectives.
- Reread and cut: no generated text should go out without your pass.
Create your style profile
Paste 3-5 texts you wrote and like into a chat and ask: “Analyse my writing style and write a 10-point style guide another person could follow to write like me.” Save the result in custom instructions or a project: every future text will start from your voice.
Ten ready-made prompts for work
Replace the parts in square brackets.
I need to write to [recipient and relationship] to [goal: chase a payment, say no, announce a delay]. Context: [what happened]. Tone: firm but polite. Max 150 words. Propose two versions and a subject line for each.
Summarise the document below for [who has to decide]. Structure: 1) what it is in 2 lines; 2) the 5 points that matter, with page numbers; 3) risks or unclear points; 4) what I need to decide and by when. Do not add information that is not in the text.
DOCUMENT: [paste or attach]
Tomorrow I’m meeting [who] about [topic]. My goal is [outcome]. Prepare: the 5 questions they’re most likely to ask with a short answer for each, the 3 hardest objections and how to respond, and a 30-second opening.
Explain [concept] as if I were smart but not an expert in the field, with a concrete example. Then ask me 3 questions to check I understood, and correct me.
In Excel I have [column description with a sample row]. I need a formula that [result]. Give me the formula, explain each part and tell me one case where it might fail.
Act as an editor on the text below. Don’t rewrite it: list in a table the sentences to improve, the issue (clarity, length, tone, errors) and your suggestion. Keep my style.
I must choose between [option A], [option B] and [option C] for [goal]. The criteria that matter to me are [criteria, in order of importance]. Make a comparison table, score each criterion 1-5 with reasons, and tell me which you’d pick and what would change your mind.
Give me 15 ideas for [goal]. The first 5 obvious, the next 5 unusual, the last 5 bold. For each: one line of description and the first concrete step to try it this week.
Translate the text below into [language] for [audience]. Don’t translate word for word: adapt expressions and examples to the cultural context. Flag in square brackets where you had to choose between interpretations.
I want to learn to do [task] myself. Don’t do it for me: guide me step by step, let me try each step and correct me. Start by asking what I already know.
The last prompt may be the most important. Using AI to avoid all effort is convenient today and costly tomorrow: people who study learning (from Cal Newport to many researchers) warn that skills you don’t exercise atrophy. Use AI to remove repetitive work, not the effort of understanding.
Common mistakes
- Asking too much in one message: ten different requests produce ten mediocre answers. Split them.
- Not saying who it’s for: a text for a client, a colleague or a judge are three different texts.
- Accepting the first answer: quality arrives at the second or third iteration.
- Assuming it knows: if information is in your head and not in the prompt, the model invents or ignores it.
- Using the same chat for everything: context gets dirty (see the context and memory guide).
Sources
Informational content, current as of the date shown. Prices, models and rules change often: always check official sources. Written with the support of AI tools and reviewed by the author.
