zmime.com
Category · Text

Text prompts

Prompts for the writing you actually ship: articles, landing pages, emails, summaries and rewrites.

  • Long-form
  • Copywriting
  • Summarising
  • Editing
  • Tone control
01 — Overview

Why these prompts look the way they do

Text is where most people start with prompting, and it is also where the sloppiest prompts survive longest. A vague request still returns something readable, so the feedback loop is weak. The fix is to make every prompt state four things: who is writing, who is reading, what the piece must accomplish, and what "done" looks like.

The templates below are written for that structure. Replace the bracketed slots, keep the constraints, and you will spend far less time rewriting output. Each one has been kept model-neutral, so the same prompt works whether you are in ChatGPT, Gemini or Claude.

02 — Anatomy

The anatomy of a reliable text prompt

  • Role Who is speaking, and with what expertise. "You are a B2B content editor for a payroll SaaS."
  • Audience Who reads it and what they already know. Beginners need definitions, practitioners need specifics.
  • Task One verb, one deliverable. Draft, rewrite, summarise, outline — not all four at once.
  • Context Source material, product facts, links, transcripts. Paste it in rather than hoping the model recalls it.
  • Constraints Length, reading level, format, banned words, structure. This is where quality is won.
  • Output shape Markdown headings, a table, JSON, or a numbered list. Name it explicitly.
03 — The prompts

5 copy-ready text prompts

Replace everything in [brackets]. Keep the constraint lines — they do more work than the descriptive ones.

Nothing matches that word. Clear the field to see every prompt.

Prompt 01

Long-form article draft from an outline

Turns an approved outline into a first draft that stays on structure.

  • blog
  • long-form
Template · plain text
You are an experienced [industry] writer producing a draft for [publication or blog name].

AUDIENCE: [who they are, what they already know, what decision they are making]
GOAL: the reader should be able to [specific outcome] after reading.
SEARCH INTENT: [informational / comparison / how-to]

OUTLINE TO FOLLOW EXACTLY:
[paste H2s and H3s]

SOURCE MATERIAL (do not invent facts beyond this):
[paste notes, product docs, data, quotes]

CONSTRAINTS
- 1,200-1,500 words, Markdown headings that match my outline.
- Grade 9 reading level. Short paragraphs, max 3 sentences.
- No introductions that begin with "In today's world" or "In the fast-paced".
- Every claim traces to the source material. If something is missing, insert [NEEDS SOURCE].
- One concrete example per H2.
- End with a 3-bullet summary, no call to action.

Write the draft.
Prompt 02

Landing page copy with a proof-first structure

Homepage or feature page copy that leads with outcomes instead of adjectives.

  • copywriting
  • conversion
Template · plain text
Act as a conversion copywriter. Write landing page copy for [product].

WHAT IT DOES: [one sentence, plain language]
WHO IT IS FOR: [segment], who currently [status quo / workaround]
PAIN: [top 3 pains, most acute first]
PROOF: [metrics, customers, integrations, guarantees]
DIFFERENTIATOR: [why this instead of the obvious alternative]
TONE: [e.g. direct, warm, no hype]

DELIVER IN THIS ORDER
1. Five H1 options, each under 60 characters, benefit-led.
2. One subhead per H1 that names the mechanism.
3. Three benefit blocks: heading (max 6 words), 2 sentences, one proof point.
4. Objection handling: 4 objections with a one-paragraph answer each.
5. Two CTA options — one low commitment, one high intent.

RULES
- No superlatives ("best", "revolutionary", "seamless").
- Verbs over adjectives. Numbers over claims.
- Do not mention features without stating the outcome they cause.
Prompt 03

Faithful summary with a strict no-inference rule

Meeting notes, reports and research papers, without invented conclusions.

  • summary
  • research
Template · plain text
Summarise the document below for a [role] who has 2 minutes.

OUTPUT
- TL;DR: 3 bullets, max 20 words each.
- Key decisions: bullet list with the decision and who owns it.
- Open questions: only questions actually raised in the text.
- Numbers mentioned: table of figure, what it measures, and where it appears.

RULES
- Use only what is in the text. Do not add context, causes or recommendations.
- If something is ambiguous, write "unclear in source" instead of guessing.
- Preserve the author's terminology, do not paraphrase named concepts.
- Quote at most 15 consecutive words at a time.

DOCUMENT:
"""
[paste text]
"""
Prompt 04

Rewrite for tone without losing meaning

Editing an existing draft to hit a specific voice and reading level.

  • editing
  • tone
Template · plain text
Rewrite the passage below.

KEEP: every factual claim, all numbers, the argument order, and any named entities.
CHANGE: tone to [target tone], reading level to [grade], sentence length to average [n] words.

STYLE RULES
- Active voice. Cut hedging ("maybe", "sort of", "it could be argued").
- Remove filler openers and restatements.
- Keep technical terms, but define each one the first time in a half-sentence.
- Target [n]% shorter than the original.

DELIVER
1. The rewritten passage.
2. A short change log: what you cut and why (max 5 bullets).

PASSAGE:
"""
[paste text]
"""
Prompt 05

Cold email sequence grounded in one insight

Outbound or follow-up email that reads like a human wrote it.

  • email
  • outreach
Template · plain text
Write a 3-email sequence from [your role] at [company] to [target role] at [target company].

RESEARCH INSIGHT (the only personalisation you may use):
[paste the specific, verifiable observation]

OFFER: [what you do, in one sentence a stranger understands]
PROOF: [one comparable customer or metric]
ASK: [single low-friction next step]

CONSTRAINTS
- Email 1: max 90 words. Email 2: max 60. Email 3: max 40.
- Subject lines under 45 characters, lowercase, no clickbait.
- One question per email, at the end.
- No "hope this finds you well", no flattery, no "circling back".
- Each email must make sense if the previous one was never read.

Return as a table: email, subject, body, send day.
04 — Craft notes

What separates a good prompt from a wasted run

Do this 05

  • Paste the source material instead of asking the model to recall facts.
  • Give one task per prompt, then iterate in follow-up turns.
  • Specify length in words and structure in headings.
  • Add a "if information is missing, write [NEEDS SOURCE]" rule to expose gaps.
  • Show one example of the voice you want — a paragraph beats an adjective.

Not this 05

  • Stacking "write, then optimise for SEO, then translate" into one request.
  • Asking for "engaging" or "professional" copy with no reference point.
  • Letting the model choose the format when you already know it.
  • Accepting statistics without a source you provided.
  • Requesting a word count without also capping paragraph length.
05 — Questions

Asked often, answered plainly

Why does my text output still sound generic?

Generic output is usually a context problem, not a model problem. If the prompt contains no product facts, no audience specifics and no voice sample, the model fills the gap with the statistical average of the internet. Paste two paragraphs of your own writing and label them as the voice reference.

Should I use a system prompt or put everything in one message?

Put durable rules (role, voice, banned phrases, output format) in the system instruction and the changing material in the user message. That way you can reuse the setup across dozens of drafts. The OpenAI API guide covers how roles are weighted.

How many examples should I include?

One or two well-chosen examples usually captures format and voice. More than three tends to make output repetitive and mimic the examples too literally, especially for creative writing.

06 — Sources

Read the primary documentation

Vendors change flags, limits and defaults often. Where this site disagrees with official docs, the docs win.