ChatGPT Prompts for Blog Posts: 10 Outline Templates

Most ChatGPT-prompt lists for bloggers optimize for the wrong number. Forty, fifty prompts - impressive scroll length, but every prompt is three lines long, and three-line prompts produce the generic posts your readers already skip.

I kept this list short so it works harder: ten prompts covering the actual jobs of blog production - finding ideas, outlining, drafting, writing openings, and refreshing old posts. Each comes with the reasoning behind its structure and notes on adapting it, because a prompt you understand is a prompt you can fix.

The competition is doing this too - 80% of marketers now use AI for content creation, per HubSpot’s State of Marketing 2026 - which is precisely why unedited prompt output can’t differentiate you. I designed the prompts below to carry your specifics, because that’s the part nobody can copy.

How these prompts are built

Every prompt below follows the same four-part anatomy - role, context, constraints, output format - explained in my prompt library overview. Two rules I hold for all of them:

Fill the bracketed placeholders with specifics. “B2B SaaS for construction-project managers” beats “my company.” The model can’t be specific about facts it doesn’t have - vague context in, generic post out.

Attach one example you’re proud of. Pasting a past post you consider good raises output quality more than any wording tweak, because it shows the model your bar instead of describing it.

Ideation prompts

1. The problem-mining prompt

You are a content strategist for [company - what it sells, to whom].
Our readers are [role] who struggle with [3–5 real problems, in their words].

List 15 blog post ideas that each address one specific struggle.
For each: a working title, the search intent behind it
(informational / commercial / navigational), and the one question
the post must answer to be worth reading.

Constraints: no listicles of generic tips, no "ultimate guides,"
nothing we could not support with specific detail.
Format: numbered list, one line of rationale per idea.

Why it works: asking for the struggle behind the idea forces topics your audience actually searches for, instead of topics that merely sound like your industry. The “no ultimate guides” constraint blocks the model’s strongest cliché habit.

Adapt it by replacing the struggles with verbatim complaints from sales calls, support tickets or community threads. Real customer language is the highest-value context you own - my buyer persona prompts show how to mine it systematically.

2. The competitor-gap prompt

Here are the H2 headings of the top 3 ranking posts for
"[target keyword]": [paste headings].

You are an editor deciding whether we can write something better.
Identify: (1) what all three cover - table stakes we must include,
(2) what none of them cover that a [audience role] would still need,
(3) which single gap is the strongest angle for a new post.
Format: three labeled sections, then a one-paragraph recommended angle.

Why it works: the model is far better at analyzing text you give it than at guessing what ranks. Feeding it real competitor headings turns “write me something original” - which it can’t judge - into a comparison task, which it can.

Adapt it by pasting headings from the actual search results for your keyword (two minutes of copying) rather than asking the model what competitors “probably” say. The same paste-the-evidence rule scales up to full competitor analysis - reviews, ads and pricing pages instead of headings.

Outlining prompts

3. The intent-first outline

You are an editor at [publication-style reference, e.g. "a trade
publication for e-commerce operators"].

Post: "[working title]" targeting "[keyword]".
Reader: [role], who searched this because [the situation they're in].
They should leave able to [specific outcome].

Write an outline: H2s phrased as the reader's actual questions,
2–3 bullet points per section stating what must be said (claims,
examples, numbers to find - not filler descriptions), and a note on
what the post deliberately will NOT cover.

Constraints: max 7 H2s. No "Introduction" or "Conclusion" headings.

Why it works: “H2s as the reader’s questions” produces sections people (and answer engines) can actually navigate, and the “will not cover” line is the cheapest scope-control trick in editing - it stops the outline from bloating into a fake ultimate guide.

4. The skeptic’s outline pass

Here is a draft outline: [paste outline from prompt 3].

You are a skeptical [audience role] with 10 minutes. Go through it
and mark each section: WOULD READ, WOULD SKIM, or WOULD SKIP -
with one line explaining why. Then propose the revised outline:
cut the skips, tighten the skims, and add anything you needed
that is missing.

Why it works: models are more honest as critics than as authors - asking the same model to attack its own outline surfaces weak sections it would happily have written anyway. This is the single highest-leverage two minutes in the workflow.

5. The comparison-post outline

Post: "[X] vs [Y] for [audience]".
You are a practitioner who has used both. Reader: [role] trying to
choose, whose real constraint is [budget / team size / timeline].

Outline the post around decision factors, not features: 4–6 factors
this reader actually decides on, and for each, what would make X or
Y the right answer. End with a "choose X if… / choose Y if…"
section. No winner-takes-all verdict.

Constraint: mark every factor where the honest answer is "we'd need
to test this" - do not fabricate a difference.

Why it works: structuring by decision factors instead of feature lists is what separates useful comparisons from spec-sheet rewrites - and the “mark what needs testing” constraint stops the model from inventing differences to seem thorough, which is its default failure mode in comparisons.

Drafting prompts

6. The section-by-section draft

We are writing "[title]". Outline: [paste approved outline].
Voice: [2–3 adjectives + one sentence of what to avoid].
Here is a past post in the voice we want: [paste excerpt].

Draft ONLY the section "[H2 name]". 150–300 words. Every claim
either comes from the context I gave you or is marked [VERIFY].
Do not restate the section heading or preview other sections.

Why it works: drafting one section at a time keeps the model inside its quality window - full-post drafts degrade badly after ~800 words as the model starts padding. The [VERIFY] convention is the difference between AI-assisted writing and publishing hallucinations: everything marked gets checked or cut in editing.

Adapt it by running it once per section, in order, pasting the growing draft back in as context every second or third section so transitions stay coherent. If you have a brand voice system prompt set up, drop the two voice lines and let it do that job.

7. The evidence-weaver

Here is a drafted section: [paste]. Here are the facts I want
worked in: [paste stats/quotes, each with its source].

Rewrite the section to incorporate them where they genuinely
support the point. Rules: never alter a figure, attribute each to
its source in the sentence ("according to [source]"), and if a fact
does not fit naturally, leave it out and tell me - do not force it.

Why it works: separating drafting from evidence-insertion means you control the facts and the model controls the wording - the correct division of labor. Letting the model supply its own statistics is how fabricated numbers end up published under your byline.

Opening prompts

8. The anti-throat-clearing intro

Post: "[title]". Reader: [role] who searched "[keyword]" because
[situation]. The post's core promise: [one sentence].

Write 3 alternative openings, max 60 words each:
(a) start from the reader's pain, (b) start from a surprising or
specific fact I provide: [fact + source], (c) start from a
mistake the reader is probably making.

Banned: "In today's fast-paced world", "In the age of AI", any
opening that defines the topic, any rhetorical question.

Why it works: the ban list matters more than the instructions - models open with throat-clearing by default, and naming the clichés is the only reliable way to block them. Three framings in one pass costs nothing and one of them usually survives editing.

9. The hook-from-the-body prompt

Here is my finished draft: [paste]. Find the single most
surprising, contrarian or concrete claim in it. Write an opening
paragraph (max 50 words) that leads with that claim, then one
sentence promising what the post delivers.

Why it works: by the time a draft is finished, the best hook is usually buried in section four. Making the model mine the draft for it beats asking it to invent a hook from nothing - the raw material is already yours.

Maintenance prompts

10. The refresh prompt

Here is a post published on [date]: [paste]. Here is what has
changed since: [new facts/products/prices, with sources - or
"unknown, flag candidates"].

Audit it: (1) claims likely outdated - list each with why,
(2) sections weaker than what now ranks, (3) a refresh plan
ordered by impact: rewrite / update / delete per section.
Do not rewrite anything yet.

Why it works: refreshing beats republishing for accumulated authority, but only if you know what’s stale. Separating the audit from the rewrite keeps you in the editor’s seat - you approve the plan before any text changes.

Chaining them: the full-post workflow

The prompts are numbered because they chain. For a new post, my sequence is 1 or 2 → 3 → 4 → 6 (per section) → 7 → 8 or 9: mine the idea, outline for intent, attack the outline, draft in sections, weave in your verified evidence, then write the opening last - openings written first promise a post that doesn’t exist yet, openings written last summarize one that does.

Two chaining habits worth stealing. First, don’t carry one chat across stages: start a fresh conversation at the outline stage and again at the drafting stage, pasting in only the approved artifact from the previous stage. Long chats accumulate the model’s own earlier suggestions as context, and it starts agreeing with itself - the outline critique in prompt 4 is measurably softer when run in the chat that produced the outline. Second, keep the human approval between stages: the chain is prompt → artifact → your yes → next prompt. Automating straight through from idea to draft is how you end up editing a 2,000-word expansion of a section you’d have cut at the outline stage.

For refreshes, the chain is shorter: 10 → your decisions → 6 (only for sections marked “rewrite”) → 7. The audit stays cheap because you only draft what the audit condemned.

When the output is good enough

The rule from the prompt library applies doubly to full posts: ship AI-assisted drafts only when you’d have accepted the same work from a competent freelancer, and edit exactly as hard. In practice that means every [VERIFY] resolved - my fact-checking prompts are the toolkit for that pass - every generic paragraph cut or grounded in something specific to you, and at least one thing in the post - a number, a template, an example - that a competitor couldn’t paste into their own blog.

If you’re producing at volume, the bottleneck moves upstream from drafting to briefing - that’s covered in the AI content brief workflow, and the wider context of where prompts fit lives in the complete guide to AI marketing.

FAQ

Which of the ten should I try first? Prompt 4, the skeptic’s outline pass - even on outlines you wrote yourself. It’s the lowest-effort prompt here, it needs no setup beyond pasting an outline, and it’s the fastest way to feel the difference between AI as a generator (mediocre until you feed it well) and AI as a critic (immediately useful). Teams that start with the critic prompts also build the right habit from day one: the model challenges, the human decides.

Do these prompts work in Claude or Gemini too? Yes. The four-part anatomy (role, context, constraints, format) is model-agnostic. Only tone-tuning differs between models. If you switch models, rerun prompt 4 - different models pad different sections.

How long should the finished posts be? As long as the question requires and no longer. Word-count targets produce padding. The outline’s “what this post will not cover” line is a better length control than any number.

Will Google penalize AI-assisted posts? Google’s guidance targets unhelpful content, not the tool that produced it. The risk is publishing unedited output at volume. Posts that survive the freelancer test, with verified facts and something only you could add, are competing on the same terms as anything human-written.

Related: how marketers in different roles put prompts like these to work - see AI for content marketers. To pick the keywords these posts target, see how to use ChatGPT for keyword research, and to turn a finished post into a week of social posts, the caption prompts by platform.