Lesson 5 of 32 · Part One, the principles

The Secret of Silent Suggestion

Few-Shot Prompting

Instead of describing the output you want, show two or three finished examples and let the model work out the pattern. A worked example carries decisions that adjectives cannot - how long an answer should run, where the line falls between two categories, which words you would never use. It is the difference between asking for a short caption and handing over three captions you consider short.

Reach for it whenWhen you already have something you did well - old exam answers, your three best captions, a few sorted feedback forms - and you want the next fifty to come out the same way.

How to use The Secret of Silent Suggestion in Claude

Wrap each demonstration in <example> tags, with the input and the output labelled inside. This is the most useful single habit on Claude and it matters most in this chapter: a tagged example is unambiguously reference material, so Parveen's three old captions are never mistaken for the posts to caption, and the three sorted comments are never mistaken for the batch to sort. Examples first, the real item last, and a short block naming what the examples teach - the length, the labels, where the line between two categories falls.

  • Wrap each demonstration in <example>, input and output labelled.
  • Examples first, the item to process last.
  • Add a block naming what the examples teach.

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

Wrap each demonstration in <example> tags, with the input and the output labelled inside. This is the most useful single habit on Claude and it matters most in this chapter: a tagged example is unambiguously reference material, so Parveen's three old captions are never mistaken for the posts to caption, and the three sorted comments are never mistaken for the batch to sort. Examples first, the real item last, and a short block naming what the examples teach - the length, the labels, where the line between two categories falls.

From the book Keep your best examples in a Project file or instructions so every chat follows the same pattern, without pasting them again. Large batches, like long feedback lists, can go in as an uploaded document.
Worth knowingClaude reads examples attentively enough that a disagreement between two of them becomes a visible problem - it may ask which you meant, or quietly split the difference. Check that your examples agree with each other before blaming the output.
  • Wrap each demonstration in <example>, input and output labelled.
  • Examples first, the item to process last.
  • Add a block naming what the examples teach.
  • Check your examples do not contradict each other.
  • Ask which example it copied, to see the pattern it read.
Describing a Pattern vs. Showing It

Instead of

Give me science questions for practice.

Claude version

<example>
<question>Why is a copper vessel coated with tin? (2 marks)</question>
<answer>Tin prevents copper reacting with acidic food, which could form harmful compounds.</answer>
</example>

<example>
<question>Why do we see a rainbow after rain? (2 marks)</question>
<answer>Water droplets split sunlight into its colours by dispersion.</answer>
</example>

<what_the_examples_teach>
Two marks. A "why" question about an everyday observation.
The answer is one or two lines, names the process, and stops. No introduction, no extra fact.
</what_the_examples_teach>

<my_exam>
Class 10 board, Science. Topics for this set: chemical reactions and light.
</my_exam>

<rules>
Follow the examples, not your own idea of a good question.
If one of your answers runs longer than my examples, cut it rather than keep it.
Do not restate these instructions in your output.
</rules>

<task>
Write five more question-and-answer pairs in exactly the pattern of the examples above.
After the five, one line telling me whether any of them drifted from my pattern, and how.
</task>
Open Claude 1,066 characters

Then loop it

  1. The pattern held. Now three more in the same <example> shape but worth 3 marks, with answers one line longer. Tell me if those answers are now too long to be worth three marks rather than four.
  2. Now hide the answers and quiz me one at a time, marking each out of its marks. Then save these examples as a file in a Project called "Board revision", so every chapter follows the same pattern without my pasting them again.

Why it worksTwo sample Q-and-A pairs teach the exact format, length and marks pattern better than any description.

"Write a Caption" vs. Captions That Match Your Voice

Instead of

Write captions for my tailoring posts.

Claude version

<example>
<post>A pheran in wool with embroidery, winter collection</post>
<caption>Warm wool, soft embroidery - a pheran made for Srinagar winters. Orders open till 10 Dec.</caption>
</example>

<example>
<post>A kurta set remade from a customer's old saree</post>
<caption>Mother's old saree, new life as a kurta set. Bring yours, we'll remake it.</caption>
</example>

<example>
<post>A bridal suit with heavy handwork</post>
<caption>Bridal suit, 40 hours of handwork. Every stitch, a blessing.</caption>
</example>

<what_the_examples_teach>
Two short parts. The material or the labour named plainly. One line of feeling, one line of invitation or fact.
Under twenty words. No hashtags. No emojis. No "DM for price".
The voice of a tailor who made the thing, not of a brand selling it.
</what_the_examples_teach>

<new_posts>
A kids' pheran.
A lehenga alteration.
A festive dupatta.
</new_posts>

<rules>
Match the examples, including their length. If a caption of yours is longer than all three of mine, it is wrong.
Do not invent prices, dates or offers I have not given you.
</rules>

<task>
Write one caption for each of the three new posts.
Then one line telling me which of my three examples each new caption is closest to - I want to know which voice you copied.
</task>
Open Claude 1,282 characters

Then loop it

  1. The lehenga alteration caption is perfect. Keep that voice. Write four more alteration captions in exactly that register - an old saree, an old suit, a mother's lehenga, a loose kurta. Tell me if any of them has slipped into sounding like an advertisement.
  2. Now add two Kashmiri or Urdu words to each, gently - words my own customers use. Keep them in an Artifact so I can pick from them as I post. Then put my three original examples into a Project file called "My caption voice".

Why it worksHer own best captions silently set the rhythm, length and emotion, so new ones sound like her.

Vague Sorting vs. A Pattern to Follow

Instead of

Sort this feedback from the relief camp.

Claude version

<example>
<comment>No drinking water since yesterday</comment>
<category>Water</category>
<urgency>High</urgency>
</example>

<example>
<comment>Thank you for the blankets</comment>
<category>Praise</category>
<urgency>Low</urgency>
</example>

<example>
<comment>Children need notebooks to restart school</comment>
<category>Education</category>
<urgency>Medium</urgency>
</example>

<allowed_categories>
Water, Food, Health, Shelter, Education, Praise, Other. No others.
</allowed_categories>

<allowed_urgency>
High, Medium, Low. No others.
</allowed_urgency>

<what_the_examples_teach>
A need that is unmet right now is High. A real need that can wait some days is Medium. Thanks, and comments that are not needs, are Low.
</what_the_examples_teach>

<rules>
If a comment carries two needs, label the more urgent one and put the second in the Notes column.
If nothing fits, use Other. Do not stretch a category.
Where you are genuinely unsure between two urgencies, write the lower one and flag it in Notes - at a relief camp I would rather check than have something quietly marked High.
Do not summarise, do not comment, do not reorder my rows.
</rules>

<comments_to_sort>
[Paste all the comments here, one per line, or upload the file. Names have been removed.]
</comments_to_sort>

<task>
Sort every comment above into one category and one urgency, following the examples. Give me a single table - Comment | Category | Urgency | Notes - in the order I gave them, as an Artifact so I can correct labels in place.
</task>
Open Claude 1,527 characters

Then loop it

  1. Now, in the same Artifact, add a count per category and bring the High-urgency rows to the top. Do not change any label you have already given. Tell me which rows you were least sure about.
  2. Now a five-line summary for our coordinator to share in the team meeting. Then put these examples and the category list into a Project called "Camp feedback", so tomorrow's batch needs only the comments.

Why it worksThree labelled examples show the exact categories and format, so hundreds of comments get sorted consistently.

Questions

What is The Secret of Silent Suggestion in prompt engineering?

Instead of describing the output you want, show two or three finished examples and let the model work out the pattern. A worked example carries decisions that adjectives cannot - how long an answer should run, where the line falls between two categories, which words you would never use. It is the difference between asking for a short caption and handing over three captions you consider short.

When should I use this technique?

When you already have something you did well - old exam answers, your three best captions, a few sorted feedback forms - and you want the next fifty to come out the same way.

How do I apply it in Claude?

Wrap each demonstration in <example> tags, with the input and the output labelled inside. This is the most useful single habit on Claude and it matters most in this chapter: a tagged example is unambiguously reference material, so Parveen's three old captions are never mistaken for the posts to caption, and the three sorted comments are never mistaken for the batch to sort. Examples first, the real item last, and a short block naming what the examples teach - the length, the labels, where the line between two categories falls.

What are the limits of this technique in Claude?

Claude reads examples attentively enough that a disagreement between two of them becomes a visible problem - it may ask which you meant, or quietly split the difference. Check that your examples agree with each other before blaming the output.

How do I improve the answer after the first try?

After the first answer, check whether the AI truly copied your pattern - length, tone, format, labels - or drifted. If it drifted, point to the gap: "Your answers are longer than my examples." If your examples are too alike, the output will be too, so add a different example and re-run. When the results are right, save your examples as a reusable template; next time, paste them and only change the new items.

Loop it

After the first answer, check whether the AI truly copied your pattern - length, tone, format, labels - or drifted. If it drifted, point to the gap: "Your answers are longer than my examples." If your examples are too alike, the output will be too, so add a different example and re-run. When the results are right, save your examples as a reusable template; next time, paste them and only change the new items.

This lesson on the other tools

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