AI prompts for blog posts: 12 that produce drafts worth editing

The usual prompt is one line: "Write a blog post about X." It hands the model a topic and leaves every other decision to it. Anthropic's documentation advises the opposite, being explicit rather than "relying on the model to infer this from vague prompts." The twelve prompts below split a post into the stages where a prompt can do real work, and each comes with what to paste in and what to check in the answer. The wording is ours, and none of it depends on a particular chat tool.

What makes a blog prompt work

Give the model the context it lacks. OpenAI's guide builds a prompt from four parts, identity, instructions, examples and context, where context is information the model needs "like private/proprietary data outside its training data." For a blog post that means the reader, your angle, your notes and the sources you trust.

State the constraints. Anthropic's guide says to "be specific about the desired output format and constraints," and offers a test worth borrowing: show the prompt to a colleague with minimal context on the task, and "If they'd be confused, Claude will be too."

Show an example. Anthropic calls examples "one of the most reliable ways to steer Claude's output format, tone, and structure," so paste a paragraph you wrote in the voice you want. OpenAI's advice is to "show a diverse range of possible inputs with the desired outputs."

Say what to do, not only what to avoid. Anthropic recommends telling the model what to do instead of what not to do, so ask for "plain sentences, one idea each" rather than pasting a list of banned words.

Trust no fact. Anthropic's documentation says even the most advanced models "can sometimes generate text that is factually incorrect," and recommends explicitly allowing the model to say it does not know. Several prompts below build that in.

Prompts before the draft

1. List the reader's questions

I am writing a blog post for {reader} about {topic}. List the questions this reader is likely to have, from first to last, grouped by stage. Mark any question you are guessing at.

Paste in a one-sentence description of the reader. Check the answer against real demand before planning around it: People also ask, autocomplete and the queries in your Search Console. A list of plausible questions is a hypothesis, not search data.

2. Read the intent of the results page

Below are the titles and headings of the top five results for "{keyword}". What format do they share, what do they all cover, and what does none of them cover? Answer from this text only.
<results>
{titles and H2s, copied from the live results}
</results>

The tags mark where pasted material starts and ends, which both OpenAI's and Anthropic's guides recommend. Open two of the pages yourself before trusting the summary. The gap the model names is a candidate angle; whether it fits the query's search intent is your call.

3. Outline from a brief

Using the brief below, propose an outline: H2 headings written as the questions a reader asks, in the order they need them, each with one sentence on what the section must deliver. No introduction or conclusion headings.
<brief>
{query, reader, angle, points to cover, sources}
</brief>

This prompt is only as good as the content brief pasted into it. Check that each H2 answers one question and that your angle shows up in the outline, not only in the brief.

Prompts for the draft

4. Draft one section under constraints

Write the section "{H2}" in the voice of the example paragraph. Use only the facts in my notes. Write plain sentences, one idea each, in under {n} words. Where you need a fact my notes do not contain, write [SOURCE NEEDED] instead.
<example>{a paragraph you wrote}</example>
<notes>{your notes and sources}</notes>

One section at a time is easier to check than a whole post. Look for every [SOURCE NEEDED] marker and for any number, name or date that is not in your notes: that is where invention happens.

5. Ask what needs evidence

List each claim in this draft that a careful reader would want a source for. For each, say what kind of source would support it and what to search for. Do not name specific studies, URLs or titles.

Find and open the sources yourself. The last line keeps the model from supplying references from memory, which you would have to verify anyway.

6. Argue against the draft

You are a skeptical expert in {field}. What are the three strongest objections to this draft, and where should it address them?

Keep the objections you agree with and answer them in the post. Honest trade-offs are one of the things a generic draft lacks.

Prompts for the edit

7. Mark the tells

Quote every sentence in this draft that uses a stock phrase, a list of three adjectives, a "not just X, but Y" construction, inflated significance or a closing summary. Suggest a plainer version of each. Change nothing else.

Accept the edits one by one. The full list of patterns, with examples, is on signs of AI writing.

8. Cut the padding

Shorten this section by about a third without losing any fact, example or step. Then list what you removed.

Read the removal list. If a fact or a caveat is on it, put it back.

Prompts for publishing

9. Title and meta description

Write five title options for this post. Each must promise the specific thing the post delivers and contain "{keyword}" near the start. Then write a meta description in two plain sentences.

Pick the title the post fully delivers on. Google's guidance on AI-generated content says the review before publishing covers titles, meta descriptions, structured data and image alt text as well as the body.

10. FAQ from People also ask

Below are the People also ask questions for "{keyword}". Answer each in two or three sentences using only what this post says. If the post does not answer one, say so instead of answering.

The unanswered questions are the useful output: add a section for the ones that belong, drop the rest.

11. Internal links

Here are my published posts with their URLs. Which three to five should this draft link to, from which sentence, and with what anchor text? Use only URLs from the list.
<posts>{titles and URLs}</posts>

Check that every suggested URL is on your list and that each anchor describes the page it points to. Internal linking covers how many links a post needs and where they do the most good.

12. Final fact-check list

List every number, name, date, quote and factual claim in this post, one per line, with the sentence it appears in. Then, using only the sources below, give the supporting quote for each, or write "no support found".
<sources>{text of your sources}</sources>

This follows Anthropic's advice to have the model find a supporting quote for each claim and retract any claim it cannot support. It only works against the sources you paste in, so every "no support found" line is yours to fix or cut. The full writing process these prompts slot into is on how to write a blog post, and the same principles at the scale of a whole blog are covered on SEO content writing.

How Structura handles this

Structura replaces the prompt chain with settings you make once per campaign. You write the objective and define the author persona: its expertise, audience, style, and what it should always or never do. On paid plans, you also approve the keyword bank and the list of sources posts may cite. Structura then reads the pages ranking for each keyword before drafting, links each post to posts you have already published, and fetches every outbound link before the post ships. Those settings cover most of what prompts 1 to 5 and 11 supply by hand.

Campaign posts have no per-post prompt. For a one-off post, Generate post takes a topic description and, on paid plans, research files you attach, such as reports or interview notes. New campaigns hold finished posts for your review by default, and that review is where prompts 7, 8 and 12 still apply: checking facts against reality is not something a link check does. The WordPress AI writer page shows a run from keyword to finished draft.

FAQ