Lesson 18 of 25

The Principle of Controlled Imagination

Creativity needs a spark and a fence. The spark is what you want made - a story, an app concept, three lines of ad copy. The fences are the constraints that make it specific: the style, the setting, the audience, the word limit, the thing it must not do. "Write a fantasy story" has a spark and no fences, so you get the average of every fantasy story. Add four fences and the output has nowhere generic left to go.

Reach for it whenWhen you want something made rather than looked up - a story, a concept, a piece of copy - and the first attempt came back competent and completely forgettable.

Write the spark as one line, then the fences as a labelled list - Style, Setting, Protagonist, Conflict - so each constraint is separately visible and separately adjustable. The real advantage here is iteration: produce three short variants against the same fences, pick what works, then move the piece into Canvas and edit it there rather than regenerating the whole thing from a tweaked prompt. If you work in the same voice repeatedly, the style fence belongs in Custom Instructions.

Worth knowingThere is a point where fences stop shaping and start strangling. If you specify the style, setting, protagonist, conflict, length, tense, point of view and ending, you have not written a prompt - you have written an outline and asked for transcription. If the output feels mechanical, remove a fence rather than adding one. Also: naming a living author's style gets you a thin pastiche, so describe the qualities you want instead.
  • Spark in one line, fences as a labelled list.
  • Three short variants first, then expand the one that works.
  • Move the chosen draft into Canvas to edit, not regenerate.
  • Too mechanical? Remove a fence, do not add one.
  • Describe style qualities rather than naming a living author.
The Unfenced Prairie vs. The Steeplechase Course (Creative Writing)

Instead of

Write a fantasy story.

Perplexity version

## SPARK

A short fantasy story, 1,200 to 1,500 words.

## FENCES

- **Style:** the lyrical, unhurried, affectionately observant register of R.K. Narayan. Small details of ordinary life carrying the weight. Gentle irony rather than grand drama. Do not attempt pastiche or period diction - what I want is his attention to the small and the human, in contemporary prose.
- **Setting:** 2025, in a hidden, magical version of Malgudi. The magic is ordinary to the people who live there, and never explained to the reader.
- **Protagonist:** a young woman who can speak with the old banyan trees. Her ability is a fact of her life, not a discovery she makes in this story.
- **Conflict:** a tech company is planning a data centre on the site of the most sacred banyan, without knowing what it is. She must stop them by combining old magic with modern persuasion - a public consultation, an environmental clearance, a WhatsApp group.
- **Ending:** unresolved but hopeful. No last-minute rescue and no villain's change of heart.

## WHAT NOT TO DO

- No tradition-versus-modernity sermon. The tech workers are not villains; most of them would be on her side if anyone had asked them.
- No magic system explained. No chosen one. No prophecy.
- Do not resolve it with the trees doing something supernatural and decisive. The constraint is that she wins or loses on the human ground, with the trees as a reason rather than a weapon.

## HOW TO WORK

First give me the opening 300 words three different ways - three different first scenes, same fences. Label them A, B and C and give each one line explaining what it is trying. I will pick one and you will write it out in full in Canvas, where I can edit line by line.

Then one line, honestly: which fence is fighting the story? If the hopeful-but-unresolved ending is pulling against the data-centre plot, which has an obvious resolution, say so. I will loosen it.
Open Perplexity 1,906 characters
Abstract Art vs. Commissioned Portrait (Business Ideas)

Instead of

Think of a new business.

Perplexity version

## SPARK

A concept for a mobile app.

## FENCES

- **Target user:** elderly residents of large apartment complexes in Indian metros - Hyderabad, Bengaluru. Many living alone or as a couple, with children in another city or abroad.
- **Problems to solve:** isolation, and difficulty managing daily errands.
- **Core functions:** a community bulletin board; a trusted way to order groceries and medicine; a simple way to organise social gatherings within the complex.
- **Interface constraint:** very large type, extremely few choices per screen, voice input throughout. Assume the user has never installed an app themselves and will not create a password.
- **Context constraint:** it has to work alongside the apartment complex's existing WhatsApp group and RWA, not instead of them. Any concept that requires everyone to abandon WhatsApp is dead on arrival and the WhatsApp group is the honest competitor here.

## WHAT I NEED

1. **The concept** in one paragraph - what it is and the one thing it does better than the status quo.
2. **The three core screens**, described in detail: what is on each, what a 72-year-old's thumb does, what happens when they tap the wrong thing.
3. **The trust mechanism.** "Trusted service for ordering medicine" is the hard part and the part most concepts hand-wave. Who verifies the vendor, who handles a wrong delivery, who is accountable when a prescription is filled incorrectly?
4. **Onboarding.** Who installs it. Realistically, this is the son or daughter in another city, which means the app has two users with different needs, and that should shape the design.

## THEN, THE HARD QUESTIONS

- **Why would this be used after week two?** Most community apps are installed and abandoned. Give me the honest answer, not a feature.
- **The case against building it.** If the correct answer is a better-organised WhatsApp group and a vetted list of three shops pinned to the noticeboard, say so and tell me what the app adds beyond that.
- **What is condescending here.** This design brief assumes a lot about what elderly people can and cannot do. Tell me which of my constraints is likely to be patronising rather than accessible.

No invented market sizes or adoption figures. If a number would help, mark it illustrative.
Open Perplexity 2,263 characters
A Random Noise vs. A Lyrical Song (Marketing Copy)

Instead of

Write some ad copy for our new headphones.

Perplexity version

## SPARK

Three variations of ad copy for a new noise-cancelling headphone brand called Silence.

## FENCES

- **Audience:** frequent business travellers on long-haul flights. People for whom the flight is working time or the only rest they will get that week.
- **Tone:** sophisticated, calm, focused on serenity and escape. Quiet writing - the copy should sound like the product.
- **Key benefit:** a private sanctuary of peace in the middle of travel's chaos.
- **Call to action:** "Rediscover your focus. Find your Silence." - exactly this, unchanged, at the end of each variation.
- **Length:** each variation under 280 characters including the call to action. Count them and show me the count.

## THREE DIFFERENT ANGLES, NOT THREE REWRITES

This is the instruction that matters. Give me:

1. One built on a **specific moment** - a single concrete instant on a flight when the noise stops.
2. One built on **what the silence is for** - the work done or the sleep taken, not the quiet itself.
3. One built on **contrast** - the cabin as it is against the cabin as it becomes.

If all three come back as variations of the same sentence, the prompt has failed and so has the output.

## BANNED

- The words journey, immerse, immersive, elevate, experience, crystal-clear, game-changer, premium.
- No technical specifications. No decibels, no driver sizes, no battery hours. This is mood copy.
- No crying baby. It is the first image everyone reaches for and it makes the brand sound petty.

## THEN

1. The character count for each.
2. One line per variation naming which angle it took, so I can see they are actually different.
3. **The brand name problem.** "Find your Silence" works in English. Tell me honestly how the name and that line read to someone whose first language is Telugu, Hindi or Tamil, and whether the possessive construction survives. Most of this audience flies out of Indian metros.
4. **The one you would run**, and why - and if you think all three are weaker than the plain statement of what the product does, say that instead of picking a favourite.
Open Perplexity 2,078 characters