Lesson 17 of 32 · Part One, the principles

The Language of Simplicity

A hesitant request earns a hesitant answer. "I was sort of hoping you could maybe give me an idea of the main points" invites a wander; "Summarise this in three bullet points" commands a result. Keep your fine language for the thing you want written - the instructions themselves should be short sentences, strong verbs, exact counts and exact formats. This is not the mark of a simple mind. It is the mark of someone who has decided what they want.

Reach for it whenWhen you know precisely what you need - six points, under twelve words each, in Tamil - and the answers keep arriving at the wrong length, in the wrong shape, or full of words your reader does not use.

How to use The Language of Simplicity in Claude

Here the tags do the work that your sentences no longer have to. Keep the lines inside them bare - one instruction to a line, no joining words, no explanation of why - and put the instruction itself at the bottom, which is where it is followed most faithfully. Two additions earn their place. Ask it to restate your rules as a numbered list before it answers; if it has miscounted your six points or missed a word limit, you see that in two seconds rather than in the finished sheet. And say explicitly that the output is the thing and nothing else, because unprompted it will wrap plain work in explanation. Standing rules - short sentences, no idioms, answer in Tamil - belong in a Project's instructions or a Style.

  • One instruction per line inside the tags. No joining words.
  • The instruction goes last - it is followed most faithfully there.
  • Ask it to restate your rules as a numbered list first.

From GenAI for Everyone: Prompt Engineering Handbook, First Edition 2025, by Raja Rajeshwari Seshadri and Shreeram Iyer.

Here the tags do the work that your sentences no longer have to. Keep the lines inside them bare - one instruction to a line, no joining words, no explanation of why - and put the instruction itself at the bottom, which is where it is followed most faithfully. Two additions earn their place. Ask it to restate your rules as a numbered list before it answers; if it has miscounted your six points or missed a word limit, you see that in two seconds rather than in the finished sheet. And say explicitly that the output is the thing and nothing else, because unprompted it will wrap plain work in explanation. Standing rules - short sentences, no idioms, answer in Tamil - belong in a Project's instructions or a Style.

From the book Put your plain rules - short sentences, no idioms, answer in Tamil - in a Project's instructions, so every chat in that Project follows them. You can also create a Style for simple, direct writing.
Worth knowingIts natural register is the opposite of this chapter: measured, qualified, a little literary. Give it a plain instruction and it will often produce something more polished than you asked for - a clinical step turned into a graceful sentence, a six-word point grown to fourteen. Count the words yourself on the first draft. It also likes to explain what it has done; the restate-then-output discipline is what keeps that out. And where you have forbidden a technical word, check rather than trust: it will sometimes use the word, notice, and apologise beneath the draft instead of rewriting the line.
  • One instruction per line inside the tags. No joining words.
  • The instruction goes last - it is followed most faithfully there.
  • Ask it to restate your rules as a numbered list first.
  • Say the output is the thing and nothing else.
  • Count the words on the first draft; it writes longer than asked.
The Rambling Plea vs. the Plain Command

Instead of

Ye school wala kaagaz hai, kuch samajh nahi aa raha, dekh lo na.

Claude version

<photo>
A notice from my son's school. Attached.
</photo>

<about_me>
I read very little. I am asking by voice.
</about_me>

<rules>
Answer in Hindi.
Three short sentences.
Simple words.
No English words.
Give the date as a date. Not "next week".
If the notice does not say something, say so. Do not guess.
Nothing before the answer. Nothing after it.
</rules>

<three_questions>
What must I do.
By which date.
How much money to pay, if any.
</three_questions>

<task>
Read the notice. Then answer the three questions. Follow the rules.
</task>
Open Claude 544 characters

Then loop it

  1. Date phir se likho. Akele, badi line mein.
  2. Ab teacher ke liye ek chhota jawab likho. Hindi mein. Do line. Main Saturday ko school aaungi.
  3. Rules yaad rakho. Agli baar bhi aise hi jawab dena - teen chhoti line, aasaan shabd.

Why it worksShort sentences with one action each - read, tell, three sentences, which date - leave nothing to guess, which matters most when you speak instead of type.

The Vague Wish vs. the Exact Count

Instead of

Can you maybe write something nice for customers about phone repairs and stuff?

Claude version

<shop>
Mobile repair shop, Coimbatore. This slip is handed to customers at the counter.
</shop>

<rules>
Exactly 6 points.
One sentence per point.
Under 12 words per point.
Every point starts with a verb.
No idioms. No jokes.
No greeting. No closing line.
English first. Then the same 6 points in Tamil.
Output only the two lists.
</rules>

<the_six_points>
Back up photos and contacts.
Remove the SIM and the memory card.
Never share passwords or OTPs with anyone, including us.
Unlock the phone yourself when we test it.
Check and note any old damage.
Bring the token slip when you collect the phone.
</the_six_points>

<task>
First, restate my rules as a numbered list, so I can see you have them. Count the word limit and the number of points back to me.

Then write the English list. Then the Tamil list.

Then, below both, give me the word count of every point. I will check it myself.
</task>
Open Claude 899 characters

Then loop it

  1. Point 4 is 14 words. Cut it to 8. Count it again.
  2. Now make it an Artifact laid out for a half-A4 sheet in large print, with [shop name] and [phone number] at the top.
  3. Put these rules in a Project called "Shop slips" so the next slip follows them without my repeating anything.

Why it worksExact counts, a word limit and strong starting verbs turn a vague hope into a command the AI can carry out precisely.

Loose Wording vs. Clear Rules

Instead of

Make these exercise notes kind of easier for my patient.

Claude version

<my_notes>
[Paste your exercise notes here. Remove the patient's name first.]
</my_notes>

<who_i_am>
I am the physiotherapist. The exercises, the counts and the order are mine, not yours.
</who_i_am>

<who_reads_this>
A patient of 70. Reads slowly. Will follow it alone at home.
</who_reads_this>

<rules>
One action per sentence.
Short sentences.
No medical terms.
No idioms.
Keep every number exactly as written.
Add no exercise. Remove no exercise. Change no order.
Last line, word for word: Stop and call the clinic if you feel pain.
Output only the steps and that line.
</rules>

<task>
Restate my rules as a numbered list first.

Then rewrite my notes as plain steps under those rules.

Then, separately from the sheet, two short lists: every clinical word you replaced and the plain words you chose, and any word you were unsure about, which I will decide myself.
</task>
Open Claude 879 characters

Then loop it

  1. Step 2 still says "supine". Replace it with plain words, and tell me which other clinical word nearly survived.
  2. Now the same sheet in Malayalam, as an Artifact. Numbers as digits. All rules standing.
  3. Read the Malayalam back to me in plain English, step by step, so I can confirm no count or instruction moved.

Why it worksPlain rules such as "one action per sentence" and "keep every number exactly" stop the AI from rewording loosely or changing her plan.

Questions

What is The Language of Simplicity in prompt engineering?

A hesitant request earns a hesitant answer. "I was sort of hoping you could maybe give me an idea of the main points" invites a wander; "Summarise this in three bullet points" commands a result. Keep your fine language for the thing you want written - the instructions themselves should be short sentences, strong verbs, exact counts and exact formats. This is not the mark of a simple mind. It is the mark of someone who has decided what they want.

When should I use this technique?

When you know precisely what you need - six points, under twelve words each, in Tamil - and the answers keep arriving at the wrong length, in the wrong shape, or full of words your reader does not use.

How do I apply it in Claude?

Here the tags do the work that your sentences no longer have to. Keep the lines inside them bare - one instruction to a line, no joining words, no explanation of why - and put the instruction itself at the bottom, which is where it is followed most faithfully. Two additions earn their place. Ask it to restate your rules as a numbered list before it answers; if it has miscounted your six points or missed a word limit, you see that in two seconds rather than in the finished sheet. And say explicitly that the output is the thing and nothing else, because unprompted it will wrap plain work in explanation. Standing rules - short sentences, no idioms, answer in Tamil - belong in a Project's instructions or a Style.

What are the limits of this technique in Claude?

Its natural register is the opposite of this chapter: measured, qualified, a little literary. Give it a plain instruction and it will often produce something more polished than you asked for - a clinical step turned into a graceful sentence, a six-word point grown to fourteen. Count the words yourself on the first draft. It also likes to explain what it has done; the restate-then-output discipline is what keeps that out. And where you have forbidden a technical word, check rather than trust: it will sometimes use the word, notice, and apologise beneath the draft instead of rewriting the line.

How do I improve the answer after the first try?

After the first answer, check it against your own instructions. Did it follow every number, every limit and every format you set? If not, do not argue or explain at length. Repeat the rule in one short sentence: "Use exactly five points." If the reply itself is hard to read, say: "Rewrite it using words a Class 5 student knows." Simple instructions in, simple answers out. When a prompt works, save it exactly as it is, so you can reuse it with new content.

Loop it

After the first answer, check it against your own instructions. Did it follow every number, every limit and every format you set? If not, do not argue or explain at length. Repeat the rule in one short sentence: "Use exactly five points." If the reply itself is hard to read, say: "Rewrite it using words a Class 5 student knows." Simple instructions in, simple answers out. When a prompt works, save it exactly as it is, so you can reuse it with new content.

This lesson on the other tools

The technique is the same; the way you write it is not. ChatGPT, Gemini.Perplexity.