I write for a blog, a weekly newsletter, and a YouTube channel. For two years I did it the slow way: stare at a blank doc, draft, rewrite, scrap, draft again. Then I built a small set of AI prompts that handle the mechanical 70% of content work — the structure, the hooks, the repurposing — so I can spend my time on the 30% that actually requires taste.
These aren't "write me a blog post" prompts. Those produce generic filler. Each one below is a structured prompt with a defined role, real constraints, and a specific output format. I've tested every one on real content over months. Here's what works.
How these prompts are structured
Every prompt follows the same four-part shape:
- Role — who the AI is pretending to be (e.g., "a YouTube script editor who has worked on channels with 1M+ subscribers").
- Context — the raw material you're feeding it (your outline, a rough draft, source notes).
- Constraints — the rules it cannot break (word counts, banned phrases, tone).
- Output — the exact format you want back.
The constraints are the part most people skip, and they're the part that separates usable output from generic mush.
1. The 15-second hook rewriter
Role: You are a YouTube script editor who has worked on channels with over 1 million subscribers. You specialize in the first 15 seconds — the window where viewers decide to stay or leave.
Context: Here is the opening of my script: [PASTE OPENING]
Constraints:
- Rewrite the first 15 seconds (about 40 words spoken).
- Open with a specific, concrete statement — not a question, not "in this video," not "today we're going to."
- The hook must make a promise the rest of the video keeps.
- Do not use the words "ultimate," "comprehensive," "everything you need to know," or "deep dive."
- Do not start with "Hey guys" or "Welcome back."
Output: Three different 15-second openings. Label them A, B, C. After each, write one sentence explaining why it works.
This forces variety and forces me to pick instead of accept. A good hook makes a specific claim ("This keyboard shortcut saved me 4 hours last week") rather than a vague one ("Keyboard shortcuts are important").
2. The blog intro that earns the scroll
Role: You are a magazine editor who rejects 90% of submissions in the first paragraph. You know readers decide whether to continue within three sentences.
Context: Here is my article topic and rough first paragraph: [PASTE]
Constraints:
- Write a 3-sentence intro.
- Sentence 1: a specific, surprising fact or scene — not a definition, not a question.
- Sentence 2: the tension or problem that fact creates.
- Sentence 3: what the reader will walk away able to do.
- Banned openers: "In today's world," "It's no secret that," "In recent years," "With the rise of."
- Banned phrases: "game-changer," "revolutionize," "delve into," "navigate the complexities."
Output: Two intro options. Then a one-line note on which you'd publish and why.
The ban list is the real work here. AI defaults to those phrases because they're statistically common in training data — which is exactly why they signal "generic" to any experienced reader.
3. The newsletter subject line generator
Role: You are an email strategist who has written subject lines for newsletters with open rates above 50%.
Context: This week's newsletter is about: [ONE SENTENCE SUMMARY]. The key insight readers will care about is: [THE INSIGHT].
Constraints:
- Write 8 subject line options.
- Max 45 characters each.
- No clickbait — the subject must be deliverable on by the email.
- No emojis.
- No ALL CAPS words.
- At least 2 must be curiosity-driven (tease the answer), at least 2 must be benefit-driven (state the payoff), at least 2 must be specific (include a number).
Output: 8 options in a numbered list. After each, label it [curiosity], [benefit], or [specific].
The labeling forces the AI to actually distribute across strategies instead of producing 8 variations of the same idea.
4. The one-to-many repurposer
Role: You are a content operations lead at a media company. Your job is to turn one strong piece of content into a full week of derivative content without duplicating it.
Context: Here is my finished [BLOG POST / VIDEO SCRIPT / PODCAST NOTES]: [PASTE]
Constraints:
- Produce: 3 tweets (each under 280 chars, each a standalone idea, no threads), 1 LinkedIn post (130 words, one specific takeaway, no humblebrag openers), 1 newsletter teaser (2 sentences, link-style), 5 short-form video hooks (each a single sentence).
- Each derivative must contain at least one idea NOT stated identically in the source — rephrase, extract a sub-point, or take a contrarian angle.
- Do not repeat the original's opening line in any derivative.
Output: Group by platform with clear headers.
This is the prompt that compounds. One strong blog post becomes a week of social presence, and the "no identical phrasing" constraint prevents the lazy copy-paste that makes cross-posting feel spammy.
5. The script outline that holds attention
Role: You are a YouTube producer who has planned scripts that average 8+ minutes of watch time. You think in open loops — questions raised early that are only answered later.
Context: My video topic is: [TOPIC]. My target length is [X] minutes. The one thing viewers must understand by the end is: [KEY TAKEAWAY].
Constraints:
- Produce a section-by-section outline.
- Open with a 15-second hook (unanswered question or counterintuitive claim).
- Identify 2 "open loops" — sub-questions raised in the first third that are only resolved in the last third.
- Place the most counterintuitive point in the middle, not the end (viewers who might click away get re-hooked).
- Do not write "intro" and "outro" as sections. Give each section a concrete title.
Output: Outline with timestamps. Mark open loops with ⟲ and where they close with ⟲✓.
Open loops are the retention mechanic that most creators never use deliberately. This prompt forces them into the structure.
6. The blog section that doesn't pad
Role: You are an editor who cuts 30% of every draft. You believe every paragraph must either advance the argument or provide necessary context — never both at once, and never neither.
Context: Here is a section of my draft that feels weak: [PASTE SECTION]
Constraints:
- Rewrite the section in fewer words than the original.
- Cut any sentence that restates the previous one in different words.
- Cut any sentence that begins with "It's important to note that" or "It's worth mentioning that" — if it were important, it wouldn't need that preamble.
- Replace any abstract claim with a concrete example. If you can't, cut the claim.
- Output must be at least 20% shorter than the input.
Output: The tightened section. Then the count: original words → new words.
Forcing the word-count reduction is the constraint that does the work. Without it, the AI preserves your flab out of politeness.
7. The meta description that earns the click
Role: You are an SEO copywriter who writes meta descriptions with above-average click-through rates. You know the description is a sales pitch, not a summary.
Context: My article URL is [URL]. The article title is [TITLE]. The main benefit to the reader is: [ONE PHRASE].
Constraints:
- Write 5 meta descriptions, max 155 characters each.
- Each must contain the primary keyword naturally.
- Each must end with an implied reason to click (a specific payoff, not "learn more").
- No two can share the same opening three words.
Output: 5 options, each with a character count.
8. The FAQ section generator (for SEO and genuine usefulness)
Role: You are a content strategist who adds FAQ sections that answer questions real people search for — not invented questions.
Context: My article topic is [TOPIC]. Here is the article: [PASTE OR SUMMARIZE].
Constraints:
- Generate 6 questions a reader might have AFTER finishing this article — things the article implies but doesn't fully answer.
- For each question, write a 2-3 sentence answer.
- Each answer must add information not already stated verbatim in the article.
- Do not include "What is [topic]?" — assume the reader has read the article.
Output: Q&A format, question as H3, answer below.
This adds SEO surface (FAQs often win featured snippets) without the stale "What is X?" padding that signals low-effort content.
9. The "is this done?" self-editor
Role: You are a ruthless editor. Your job is to find the weakest part of a draft and tell the writer exactly why it's weak and what to do about it.
Context: Here is my draft: [PASTE]
Constraints:
- Identify the single weakest paragraph.
- Explain in one sentence why it's weak (vague? unsupported? repetitive? wrong tone?).
- Suggest one specific fix — not "make it better," but a concrete action ("replace this with a number from your data," "cut this sentence entirely," "move this example earlier").
- Identify any paragraph that could be cut entirely without losing the argument.
- Do not praise the draft anywhere in your response.
Output: Weakest paragraph (quoted) → why → fix → optional cuts.
The "no praise" constraint matters. AI's default is to sandwich criticism in compliments, which trains you to ignore the useful part.
10. The content calendar from one pillar idea
Role: You are a content manager who plans 30 days of content from a single pillar topic, ensuring no two pieces overlap and each serves a different stage of the reader's journey.
Context: My pillar topic is: [TOPIC]. I publish on: [blog twice weekly / newsletter weekly / YouTube biweekly — LIST YOURS].
Constraints:
- Produce 8 distinct content pieces derived from the pillar.
- Each must target a different angle: beginner explainer, advanced tactic, case study, contrarian take, tool/tutorial, FAQ roundup, comparison, and personal story/opinion.
- For each, give: working title, one-line summary, target platform, and which angle it fills.
- No two pieces can cover the same sub-point.
Output: A table with columns: Title | Summary | Platform | Angle.
This is the prompt that turned my content from reactive ("what do I post today?") to planned. One pillar topic now carries a month.
How I test a prompt before trusting it
I run every new prompt three times on the same input. If the three outputs are nearly identical, the prompt is too rigid (it's just template-filling). If they're wildly different, the prompt is too loose (it's not constraining anything useful). The sweet spot is three outputs that share the structure but differ in the specifics — that tells you the constraints are working and the model still has room to think.
Then I take the best output, edit it, and publish. The prompt isn't a replacement for judgment. It's a replacement for the blank page.
If these were useful, I keep a larger library of tested prompts for the full range of work — launching products, writing marketing copy, running engineering teams, doing customer support, analyzing data, and more. The Developer Product-Launch Prompt Pack is the $9 starter (7 prompts with example outputs), and the full libraries are linked from my Gumroad storefront.
Every prompt in the packs follows the same Role → Context → Constraints → Output structure, with the same emphasis on constraints over verbosity. No filler prompts.
Top comments (1)
The Role Context Constraints Output structure is a practical way to turn prompting from vague delegation into a repeatable editing workflow, especially with details like the 45-character subject-line cap and the requirement that repurposed posts add a genuinely new angle. I also like the three-run test: identical outputs suggest overfitting, while wildly different ones usually mean the prompt lacks useful boundaries. For a founder or engineer, the next useful metric is downstream edit time-not how polished the first draft looks-because a prompt that produces consistent but unpublishable work has only moved the bottleneck.