Lesson 24 of 25

The Treasure of Specificity

A vague request gets an average answer, because the model has to pick a middle of the road you did not describe. Every concrete detail you add - a word count, a named place, a numbered structure, who is reading it - removes a guess it would otherwise make for you. Specificity is not about writing a longer prompt; it is about replacing the decisions you left open with the ones you actually want.

Reach for it whenWhen the first answer came back technically correct and completely generic - the kind of output you could have got for any company, any city, any reader.

Use Markdown structure to make specificity scannable: headings for the sections of the brief, numbered lists where the order of the output matters. The useful platform trick is to put the permanent specifics - who you are, what your business does, who your audience is - into Custom Instructions once, so each prompt carries only what is specific to the task. For long outputs, specify in Canvas and refine section by section rather than regenerating the whole thing to fix one paragraph.

Worth knowingThere is a point where more detail stops helping. Pile on thirty constraints and some get dropped silently - usually the ones in the middle of a long paragraph. Rank what matters: put the three constraints that must hold in their own numbered list and separate them from the preferences. And note that the specificity which counts is the kind that removes a choice; adjectives like "engaging" and "high quality" feel specific and remove nothing. Also, this platform's image generation does not follow a long paragraph of visual detail reliably - lead with the subject.
  • Put permanent specifics in Custom Instructions, not every prompt.
  • Separate must-hold constraints from preferences, numbered.
  • "Engaging" and "high quality" are not specifics.
  • Thirty constraints means some get dropped silently.
  • Long outputs: specify in Canvas, refine section by section.
Example 1: The Vague Wish vs. The Specific Blueprint (Content Creation)

Instead of

Write a blog post about travel in India.

Perplexity version

Write a 1200-word travel blog post titled "A 3-Day Cultural Deep Dive into Hyderabad for First-Time Visitors".

## WHO IS READING THIS

Someone who has never been to Hyderabad, is planning a long weekend, and is choosing between this and three other cities. They want to know what makes this one different. Tone: enthusiastic and informative, never a brochure.

## STRUCTURE - DAY BY DAY

1. **Day 1 - the Old City.** Charminar, Laad Bazaar, Chowmahalla Palace.
2. **Day 2 - the Nizams' era.** Salar Jung Museum, Golconda Fort including the evening light and sound show.
3. **Day 3 - modern Hyderabad.** Hi-Tech City, and a biryani meal somewhere well known - Paradise or an equivalent.

For each day: a short opening paragraph on its character, then each stop with roughly how long to spend, what to actually look at when you are there, and one specific detail a first-time visitor would otherwise walk past.

## MUST HOLD - these three, before anything else

1. Each day must work as a route, with the stops in an order that makes geographic sense. A day-by-day plan that has the reader crossing the city twice is not a plan.
2. Practical tips, specific: the metro and where it does and does not reach, when autos and cabs are the better option, and what time of year to come - say which months are uncomfortably hot rather than "visit in the cooler season".
3. Concrete over evocative throughout. "Charminar at 7am, before the shops open" is useful. "Charminar at sunset, bathed in golden light" is in every other post about this city.

## PREFERENCES

- Mention two or three things to eat by name beyond biryani.
- A line on what to wear for the Old City and the fort.
- End with what you would skip if you only had two days.

## FLAG FOR ME

At the end, list anything you have stated that I should verify before publishing: opening hours, ticket prices, whether the light show still runs, whether a restaurant is still open. You are working from training data and these change. Do not quietly omit them to look confident, and do not state them as current - list them as things to check.
Open Perplexity 2,091 characters
Example 2: The Blurry Request vs. The Sharp Command (Business Analysis)

Instead of

Analyze my sales data.

Perplexity version

Act as a senior data analyst. My sales data for the last quarter is below in CSV format.

```
[Paste CSV data here]
```

## TASKS - in this order

1. **Total revenue for the quarter.** State the figure and the number of rows it came from, so I can see nothing was dropped.
2. **Top 5 products by units sold**, as a Markdown table: product, units, revenue, share of total revenue. Note where the top seller by units is not the top by revenue - that gap is usually the most useful thing in the data.
3. **Week-by-week sales trend.** A table, then one paragraph: increasing, decreasing or flat, and whether the movement is large enough to be a trend or just noise at this sample size.
4. **One-paragraph recommendation** on which product category to focus marketing on next quarter, with the specific numbers from tasks 1 to 3 that support it.

## MUST HOLD

1. Every figure traceable to the data. No industry benchmarks, no comparisons to "typical" performance, nothing from outside the file.
2. If a column is ambiguous, missing or inconsistent, stop and ask before computing. An analysis built on a guessed column definition is worse than no analysis, and I will not know you guessed.
3. Show the arithmetic for the totals. One line is enough - I need to be able to check it.

## AND TELL ME

- What this data cannot answer. One quarter with no prior period means I cannot see seasonality, and a recommendation that ignores that is overconfident.
- Which fields I should start capturing to make next quarter's analysis better.

If the dataset is large, work in Canvas so I can see the tables build and query individual rows rather than scrolling a wall of output.
Open Perplexity 1,663 characters
Example 3: The Rough Sketch vs. The Detailed Commission (Image Generation)

Instead of

Create an image of a woman working.

Perplexity version

Generate a vibrant, photorealistic image.

**Subject:** a young Indian woman in her late twenties, a software developer, sitting at a desk and typing on a laptop. She is smiling confidently - at her work, not at the camera.

**Setting:** a modern, brightly lit office in Hyderabad's Hi-Tech City. Large windows behind her with a city view. Colleagues visible but out of focus in the background.

**Her appearance:** stylish professional Indian attire - a smart kurti with trousers.

**Camera:** mid-shot from slightly to one side, as if someone glanced over from the next desk. Shallow depth of field - she is sharp, the background soft. Natural daylight from the windows as the main light.

---

**How to use this prompt.** Order matters more than length here: the subject first, then the setting, then the camera. A single paragraph with twelve visual details in it will have the later ones dropped, so each element gets its own line - and if something must be in the picture, put it in the first two lines, not the last.

**Then iterate rather than rewrite.** Change one thing per attempt and ask for it as an edit of the image you have, not as a fresh generation: "same image, but she is looking at a second monitor rather than the laptop". A full rewrite gives you a different woman in a different office and you lose whatever was working.

**Expect these to need fixing.** Hands on a keyboard, and any text on screens or signage - both come out wrong routinely. Frame to avoid them, or plan on several attempts.

**And the specificity worth more than any visual detail:** say what the image is for. A blog header, a conference slide, a LinkedIn post - each wants a different crop and a different amount of empty space for text to sit in. If you need text over it, ask for clear space on one side and say which.
Open Perplexity 1,816 characters