Lesson 1 of 32 · Part One, the principles
The Principle of the Definite Chief Aim
A prompt gets you what you actually asked for, so the useful skill is naming the exact thing you want: the decision you are making, by when, with what numbers, and what you will do with the answer. Vague asks get average answers, because the model has to guess, and it guesses towards the middle of everything it has ever read.
How to use The Principle of the Definite Chief Aim in Claude
Separate the parts of your aim with tags - <the_decision>, <my_land>, <constraints>, <if_unsure> - so your facts are not read as instructions and your instruction is not read as background. When you are pasting something long, such as a CV or a syllabus, put it at the top and leave the instruction for the very bottom; Claude follows the last thing it reads most reliably. The <if_unsure> block earns its keep in this chapter, because a definite aim usually turns on a number you will act on, and telling Claude to write [confirm] rather than guess is the difference between a gap you can fill and a figure you quote to a dealer.
- Tag the parts: <the_decision>, <constraints>, <if_unsure>.
- Long material first, your instruction last.
- Ask for its reasoning before its verdict, so you can argue.
From GenAI for Everyone: Prompt Engineering Handbook, First Edition 2025, by Raja Rajeshwari Seshadri and Shreeram Iyer.
Separate the parts of your aim with tags - <the_decision>, <my_land>, <constraints>, <if_unsure> - so your facts are not read as instructions and your instruction is not read as background. When you are pasting something long, such as a CV or a syllabus, put it at the top and leave the instruction for the very bottom; Claude follows the last thing it reads most reliably. The <if_unsure> block earns its keep in this chapter, because a definite aim usually turns on a number you will act on, and telling Claude to write [confirm] rather than guess is the difference between a gap you can fill and a figure you quote to a dealer.
From the book Start a Project for an ongoing goal, such as this season's farm plans, and write your aim in the project instructions so every chat stays focused on it. If an answer drifts, edit your earlier message to sharpen the aim and try again.
- Tag the parts: <the_decision>, <constraints>, <if_unsure>.
- Long material first, your instruction last.
- Ask for its reasoning before its verdict, so you can argue.
- Tell it to write [confirm] rather than guess a figure.
- Keep a standing aim in a Project's instructions.
Instead of
Tell me about drip irrigation.
Claude version
<my_land> Village near Yavatmal, Maharashtra. Six acres under cotton, black soil. One borewell. It runs low by February every year, which is when the crop still wants water. No canal. No second source. I am considering drip on 2 acres only, as a trial, not on all six. </my_land> <the_decision> Go ahead with drip on those 2 acres this season, or wait another year. I must decide by next month, before sowing arrangements are made. </the_decision> <what_i_already_know> I know what drip irrigation is. Do not explain it to me. I need the money and the paperwork. </what_i_already_know> <if_unsure> Cost and subsidy figures change by district and by year. Where you are not confident, write [confirm at agriculture office] instead of a number. A gap I can fill is better than a figure I act on and find wrong. </if_unsure> <task> Work through this in order: 1. Reason aloud for four or five lines about which way my borewell situation points, before you give any verdict. I want to argue with your thinking, not only with your answer. 2. Then a yes-or-no checklist for the decision - the conditions under which drip is worth it on 2 acres, each phrased so I can answer yes or no myself. 3. Then cost per acre, the usual change in water use and yield, and the subsidy schemes to ask about by name. 4. End with the five questions to put to a local dealer before I pay anything. </task>
Then loop it
- Your reading of the borewell was right. Keep the checklist and the dealer questions untouched. Rewrite the subsidy section in simple Marathi so I can show my wife, and tell me if any of it loses its meaning in translation.
- Settled. Put the five dealer questions into an Artifact as a one-page printable list I can carry, and rewrite my whole prompt, tags and all, with blanks for [crop], [acres] and [the decision], for a Project called "Farm decisions".
Why it worksIt names the exact decision, the deadline and the land, so the AI aims at one target instead of writing a textbook chapter.
Instead of
How do I get a job?
Claude version
<my_cv> [Paste your full CV here, and the text of your internship certificate. The long material goes at the top; my instruction is at the very bottom.] </my_cv> <about_me> B.Com graduate, Lucknow. Six months' internship at a CA firm, where I did GST filing and Tally entries. No other work experience. </about_me> <this_weeks_goal> Get shortlisted for an accounts-assistant role at a mid-sized company. Not a career plan, not a list of job sites. One result: a shortlist call. </this_weeks_goal> <constraints> 120 words or fewer. GST filing and Tally must appear by name. No "hardworking", "passionate", "dynamic" or "seeking an opportunity". Plain English that a busy accounts manager reads in ten seconds. </constraints> <if_unsure> If something in my CV is too thin to claim, say so instead of inflating it. I would rather drop a line than be caught out in the interview. </if_unsure> <task> First, one line naming the strongest thing in my CV for this particular role, and one line naming the weakest. I want to know what you are building on. Then the 120-word summary. Then three things to add to a fresher's CV for this exact role and three to remove, each with its reason in one line. </task>
Then loop it
- Good diagnosis. Keep the shape of the summary. It is too formal - make it warmer, still professional, still under 120 words. Tell me if warming it costs me something a recruiter would have valued.
- Now a two-line version for the job portal's "About me" box. Then put the finished summary into an Artifact so I can adjust it as I apply, and save this prompt in a Project called "Job applications".
Why it worksReplacing "a job" with one role and one result tells the AI precisely what success looks like.
Instead of
Give me some wedding menu ideas.
Claude version
<my_business> Wedding catering in Kochi, Kerala. My own kitchen, few hands on the morning of a function. </my_business> <this_job> Wedding lunch. 300 guests, mostly Kerala Christian families. About 40 vegetarians. Budget: ₹650 per plate, vegetarian plates included in that figure. The quote must reach the family by Friday. </this_job> <what_matters_to_me> Staying inside ₹650 without the plate looking thin. Dishes I can part-cook the day before, because labour on the morning is my real constraint. Food these families will recognise as right for the occasion. </what_matters_to_me> <format> Two options. For each: Starter, Mains, Rice, Dessert, Drink, with the vegetarian substitute named beside each non-vegetarian dish, and a marker against every dish that can be cooked a day ahead. Then a rough per-plate cost split for each option. </format> <if_unsure> Where you are guessing at a Kochi rate, write [check with my supplier] rather than giving me a figure I might quote to a family. </if_unsure> <task> Before the menus, two lines on the hardest trade-off at ₹650 for this particular gathering - what has to give. Then the two options. Then tell me which one you would quote, and why. </task>
Then loop it
- Option two is closer. Keep its starter and mains exactly. Swap the dessert for something lighter, still under ₹650. Tell me if the lighter dessert will look mean beside the rest of the plate - I would rather hear it from you than from the family.
- That works. Now draft the WhatsApp quote message to the family in an Artifact - warm, respectful, under 150 words - so I can keep editing the same copy as we go back and forth.
Why it worksGuest count, budget, deadline and audience turn a loose wish into a goal the AI can actually hit.
Questions
What is The Principle of the Definite Chief Aim in prompt engineering?
A prompt gets you what you actually asked for, so the useful skill is naming the exact thing you want: the decision you are making, by when, with what numbers, and what you will do with the answer. Vague asks get average answers, because the model has to guess, and it guesses towards the middle of everything it has ever read.
When should I use this technique?
When you are facing one real decision with a date on it - whether to buy, which role to apply for, what to quote - and you need something you can act on rather than a page of background.
How do I apply it in Claude?
Separate the parts of your aim with tags - <the_decision>, <my_land>, <constraints>, <if_unsure> - so your facts are not read as instructions and your instruction is not read as background. When you are pasting something long, such as a CV or a syllabus, put it at the top and leave the instruction for the very bottom; Claude follows the last thing it reads most reliably. The <if_unsure> block earns its keep in this chapter, because a definite aim usually turns on a number you will act on, and telling Claude to write [confirm] rather than guess is the difference between a gap you can fill and a figure you quote to a dealer.
What are the limits of this technique in Claude?
Tags are labels and nothing more. A <constraints> block that says "make it good" is no better than no tag at all - the aim has to be definite inside the tag, or you have only tidied up a vague wish.
How do I improve the answer after the first try?
With a definite aim, the first answer is easy to judge: hold it against your goal. Did it answer your exact decision, or drift into general facts? Did it respect your numbers, place and deadline? If it wandered, restate the aim in one line: "Remember, I only need to decide X by Y." If it hit the target, narrow further - shorter, simpler, in your language - then save the winning prompt so you can reuse it next season or for the next job.
Loop it
With a definite aim, the first answer is easy to judge: hold it against your goal. Did it answer your exact decision, or drift into general facts? Did it respect your numbers, place and deadline? If it wandered, restate the aim in one line: "Remember, I only need to decide X by Y." If it hit the target, narrow further - shorter, simpler, in your language - then save the winning prompt so you can reuse it next season or for the next job.
This lesson on the other tools
The technique is the same; the way you write it is not. ChatGPT, Gemini.Perplexity.