Chapter 01 · Section V · 8 min read
1.5 Being specific: the difference one extra sentence makes
Specificity is the single biggest lever in prompting. Where vague prompts fail, what to add, and how much is too much.
If you fix only one thing about your prompts this week, fix specificity. Vague prompts don’t fail because the model is bad; they fail because the model has no principled way to choose between the fifty reasonable interpretations of what you asked.
Where vagueness sneaks in
The five most common vague words in prompts:
- “Better” — better how? Shorter? Friendlier? More formal?
- “Detailed” — detailed in what dimension? Length? Vocabulary? Coverage of edge cases?
- “Professional” — professional in what industry, at what seniority, in which country?
- “Simple” — simple for a Class-10 student, a busy PM, or an engineer?
- “A few” — three? Five? Ten?
Every one of these words is a coin flip disguised as an instruction. Replace them.
Four dimensions to be specific about
When your output is off, one of these is almost always the reason:
Audience. Who is this for? “Explain HTTPS to a Class-8 student in Chitwan who has never used the term before” is a different prompt from “Explain HTTPS to a backend engineer switching from HTTP.” Both are legitimate; the model needs to know which one you meant.
Length. How long? “150 words,” “one paragraph,” “no more than five bullets.” “Short” is not a length.
Format. What shape? “Return a JSON array of strings,” “Return a Markdown table with columns A, B, C,” “Return only the answer with no preamble.”
Register. How formal? “Colloquial Kathmandu-English, contractions welcome,” “Formal report style, no first person,” “Newspaper headline voice.” Register is where most rewrites feel off even when the content is right.
The one-sentence upgrade
A prompt like:
Summarise the article below.…becomes dramatically better with one added sentence:
Summarise the article below in exactly five bullet points for a busy PM who has never heard of the topic. No jargon.That single sentence pins down length (five bullets), audience (busy PM, unfamiliar), and register (no jargon). Same task, much narrower target.
The other direction: too much
Specificity is not the same as verbosity. A three-paragraph preamble about the model’s role, the company’s values, the writing style guide, and the phase of the moon usually helps less than one crisp sentence per dimension. When a prompt is failing, add precision, not volume.
A checklist to run before you send
- Have I named the audience?
- Have I given a length (words, bullets, sentences)?
- Have I specified the format (JSON, Markdown, prose)?
- Have I named the register (formal, casual, headline)?
- Have I said what to skip (no preamble, no apologies, no code fences)?
If you can answer all five in one line, you have a specific prompt.
Check your understanding
Quick check
—