Blog Posts Prompts

AI is great at first drafts and terrible at insight. Treat it like a fast junior writer: it can produce a 1,200-word draft in two minutes, but it will not know your customers, your data, or your point of view. Your job is to bring those three things; the model's job is to assemble them into prose.

The structure that consistently produces publishable long-form starts with an outline prompt (the model proposes), an outline edit (you correct it), a section-by-section draft (the model writes 400 words at a time), and a single human pass at the end. Skipping the outline edit is the most common cause of generic AI articles.

1. Decide the angle before you open the chat — the contrarian take, the new data, the lesson learned.
2. Ask the model for three competing outlines, not one.
3. Edit the chosen outline yourself — kill bland subheads, add one specific anecdote per section.
4. Draft section by section, never the whole post at once.
5. Final pass: trim 15% of the words and replace every adjective with a number.

• Asking for the whole article in one prompt — you get a 1,500-word average.
• Skipping the voice reference — the model defaults to LinkedIn-thought-leader cadence.
• No source material — the model invents stats and quotes that will not survive fact-checking.
• Generic H2s — "Why this matters", "Key takeaways", "Conclusion". Replace with claims.
• Adjective stacking — "powerful, intuitive, scalable solution". Replace each with a measurement.
• No internal links — every published post should reference 2-4 of your own posts naturally.

1. The competing-outlines prompt

You are a senior content strategist for [audience: e.g. early-stage SaaS founders].
Topic: "[working title]"
Angle: [your contrarian or specific take in one sentence]
Search intent: [what the reader is trying to do or decide]
Target length: ~1,500 words

Produce THREE different outlines:
- Outline A: chronological / story-driven
- Outline B: framework / numbered system
- Outline C: contrarian / "what everyone gets wrong"

For each: H2s with 1-sentence rationale, plus the angle of the introduction.
Do not write any body copy.

2. The section-draft prompt

Using the outline below, draft only the section titled "[H2]".
~400 words. One specific example. One concrete number or quote.
No transition sentences ("Now let's talk about…"). End on a sentence that earns the next section.

Voice reference:
"""
[paste 200-400 words from a past post in your voice]
"""

Source material to draw from (use only these facts):
"""
[paste interview notes, internal docs, data]
"""

3. The SEO refinement pass

You are an SEO editor. The target query is "[primary keyword]".
Secondary entities to cover at least once: [list 5-8].
Search intent: [informational | comparison | how-to | navigational]

Review the draft below and produce:
1. A revised H1 (under 60 chars, includes the primary keyword naturally)
2. A meta description (under 155 chars, includes the primary keyword)
3. 5 H2 rewrites that better match how people search
4. 3 internal-link opportunities with anchor text
5. One paragraph to add that covers a missing entity

Do not rewrite the body. Output only the changes above.

1. Voice extraction: paste 3 past posts and ask Claude to produce a "voice card" (sentence length, favourite words, structural habits). Reuse it as a prompt header for every future article.
2. Adversarial review: after the draft, ask a second model "What would a skeptical industry expert push back on?" Address the strongest objection inside the post.
3. Quote sourcing: extract genuine quotes from interview transcripts instead of letting the model invent them. Real voices are the single biggest credibility lever.
4. Numbered claims: replace every soft assertion with a number. "Many teams" → "7 of the 11 teams we interviewed".