You have a blog post to write. The idea is already in a chat message to a colleague, or in a note you typed on the train, and you know two things about it. The first is that an assistant could turn it into 1,200 tidy words in about twenty seconds. The second is that those would be the same 1,200 words everyone else published on the topic last month, with your logo on top.
Blog posts are where generic AI writing is most visible, because the reader arrived from a search with a question and has eleven other tabs open. They can tell within two sentences whether the post was written by someone who did the thing or by something that read about it. The model's defaults produce the second kind: a scene-setting introduction, label headers, invented examples, equal-length sections, and a conclusion that restates the introduction.
Vibe writing for a blog post means directing with your angle and your evidence, letting the assistant propose structure and a first draft fast, judging against what the reader came for, and owning the post for its whole life, including the day a number in it goes stale. This guide walks that loop for the format, with the Blog Post mega prompt doing the set-up and the annotated example showing a full run.
What changes for this format
The method is the same everywhere. These are the places where this format bends it.
Search intent is the reader's question, and your angle is the reason to read yours
The model knows the generic shape of the answer to almost any query, and it will produce that shape on request. That shape is what a reader skips past to find the post. Your angle is the part the generic answer gets wrong, or leaves out, or that you learned by doing it: the figure that surprised you, the approach you abandoned, the trade-off nobody mentions. Brief the assistant with both. The search question tells it what the reader needs to find. The angle tells it what you have that nobody else does. If you cannot state an angle in one sentence, the honest move is to wait until you can.
Headers that make claims
AI drafts produce headers that are labels: Benefits of X, Key Considerations, Getting Started. A reader skimming labels learns nothing about what you think. Ask for headers that state what the section argues, so that reading the headers alone gives the argument: Free trials bought us paperwork, not loyalty. Run the test on every draft. If a header could sit above a section arguing the opposite, it is a label, and the section underneath it is probably generic too.
Examples from your own work, or none
Ask for an example and the model invents one: a marketing team at a mid-sized software company, a freelance designer named Sarah. These are the most damaging sentences in an AI-assisted post because they are fiction presented as experience. Your raw material block should hold the real thing: your numbers, the screenshot you took, the message a customer actually sent (with permission), the attempt that failed. Where you have no example, instruct the assistant to leave a [VERIFY] tag rather than fill the gap, and then either find one or cut the claim.
The listicle trap
The model loves parallel structure: seven ways, five tips, each section the same length with the same internal shape. Parallel sections read as thorough and are usually padding, because real knowledge is lumpy. In a post written by someone who did the thing, one section is three times longer than the others because that is where the thinking was. Allow uneven sections in the brief and ask for them in the outline. If the post really is a list, make the items unequal and cut any that exist only to reach a round number.
Update dates and the post's lifetime
A blog post lives for years, and the model has no sense of time. It writes as-of-this-year flourishes, cites tools and prices that will change, and presents last quarter's figure as permanent. Keep a short list, in your source note, of every claim in the post that has an expiry: a price, a version number, a statistic, a feature. When you update the post, change the updated date honestly and only when something substantive changed. A fresh date on an unchanged post is a small lie that readers notice.
The introduction and conclusion you will delete
Nearly every AI blog draft opens with scene-setting (in an increasingly competitive landscape) and a roadmap (in this post, we will explore) and closes with a restatement and a call to action. Plan to cut both. Start at the claim or the number. End when the argument ends. If you want a closing line, write it yourself, and if you want a call to action, decide what the one action is and ask for exactly that sentence rather than a paragraph.
The process
- 01You
Write the angle in one sentence
Before the chat, write two lines. The first is the search question in the reader's words: how do I decide whether to offer a free trial. The second is your angle, as a claim: our free month bought paperwork and almost no loyalty, and here are the numbers. If the second line is a restatement of the first, you have a topic and no post yet. Go find the thing you learned, or write about something else.
- 02You
Gather the evidence you actually have
Pull the numbers from the real spreadsheet. Take the screenshot. Find the message, the commit, the before and after. Note what you tried that did not work and why. This is the raw material block, and it is the single biggest lever against a generic post. Half a page of real, specific material produces a draft that needs editing, but a one-line topic produces a draft that needs replacing.
- 03AI proposes
Outline from the reader's question
Use the Rough Idea to Outline prompt with the search question, the angle, and the raw material. Ask for headers that are claims, sections of uneven length, and a note on which section carries the angle. Reject the first outline if the headers are labels or the sections are parallel, and say why in one line.
Send this The reader's search question is: [SEARCH QUESTION]. My angle is: [ANGLE AS ONE CLAIM]. Raw material: [NOTES, NUMBERS, EXAMPLES]. Propose one outline. Every header must be a claim the reader could disagree with, not a label. Sections may be uneven; mark the one that carries my angle. No introduction section, no conclusion section. Facts only from my material; mark gaps [VERIFY]. Then wait.
- 04AI proposes
Draft once, with the Blog Post mega prompt
Brief the session with the Blog Post mega prompt or the universal brief, paste the accepted outline, and ask for one draft. Not three. The first draft's job is to show you where the material is thin and where the voice is wrong, and one draft shows that better than three variations of the same guesses. Expect to keep the structure and about half the sentences.
- 05You
Judge against the reader, section by section
Read each section asking one question: did the reader who typed that query get something here they could not get from the generic answer? Where the answer is no, the fix is usually more raw material rather than a better prompt. Use bounded edits for the sections that are close, and keep your own master copy outside the chat so a regenerated section cannot quietly change the ones you accepted.
Send this Change only the section under the header [HEADER]. The problem: it is the same length and shape as the others, but it is the section where my angle lives and it deserves twice the room. Expand it using only these notes: [ADDITIONAL RAW MATERIAL]. Do not touch any other section. Show the full new section, then wait.
- 06You
Final human edit, title last
Run the Final Human Edit pass: delete the introduction and conclusion the model wrote, start at the first claim, read aloud, rewrite by hand any sentence that sounds like a different writer. Check every link. Write the title last, when you know what the post actually says. The model's title suggestions are useful mainly as a list of what to avoid calling it. If your site discloses AI assistance, add the note now, in the words of your own policy.
- 07You
Set the update plan
Before publishing, list the claims that will expire and roughly when. A price, a version, a statistic from a dated report. Put the list with your source note. When one expires, update it, change the updated date, and leave a line saying what changed. A post that is honest about its age earns more trust than one that pretends to be new.
Toolkit
The prompts that fit this format, and when to reach for each.
- Mega promptBlog Post mega promptThe session set-up for this format. It carries the search question, the angle, the claim-header rule, and the raw material in one brief.
- Stage promptRough Idea to OutlineBefore drafting, once the angle is written. Ask for claim headers and uneven sections, and reject a parallel outline.
- Stage promptWrite First DraftFor the single draft after the outline is accepted. One draft, no options, so you can see where the material is thin.
- Stage promptFinal Human EditThe last pass before publishing. It cuts the intro and conclusion, reads aloud, and catches the sentences that sound like anyone.
Before and after
A paragraph the way an assistant tends to return it, and the same paragraph after a writer took it back.
In today's competitive landscape, many businesses offer free trials to attract new customers. While this strategy can be effective, it also comes with significant drawbacks. Free trials can attract customers who are not serious about your product, leading to higher churn rates and wasted resources. Additionally, they can devalue your offering in the eyes of potential clients. In this post, we will explore why we decided to eliminate our free first month and the key lessons we learned along the way.
We stopped offering a free first month in March. In the eleven months before that, 61 clients signed up on the free month and 14 were still with us a year later. That is a 23 percent keep rate, and the 47 who left cost us roughly 90 hours of onboarding, most of it spent chasing documents from people who had not yet decided to be clients. The ones who stayed, when we asked them, could not remember that the first month had been free. What the free month bought us, it turns out, was paperwork.
The opener and the roadmap are gone. The post now starts at the decision and the month it was made. Numbers replaced adjectives: 61, 14, 23 percent, and 90 hours came from the writer's own books, where the AI draft had only higher churn and wasted resources. And one claim was cut rather than improved: the line about devaluing the offering sounded plausible, but the writer had no evidence for it, so it went. The last sentence states the angle as a claim the headers below can then carry.
Pitfalls
The post opens with the state of the industry and a roadmap.
Delete the first paragraph and start at the first claim or number. Add no introduction section and no roadmap to the brief so it stops coming back.
The headers are labels and the sections under them could be about anyone.
Rewrite each header as a claim by hand, then ask for a bounded change to each section so it argues the header. Where no claim is possible, the section is padding, so cut it.
There is an example in the draft involving a company you have never heard of.
It is invented. Replace it with your own material or a [VERIFY] tag, and ask the assistant to list every example in the draft with its source in your notes.
Seven sections, all the same length, all the same shape.
Ask which section carries the angle and expand that one with more raw material. Cut the sections that exist to reach a number. Allow uneven in the brief.
The post says in 2026 and cites a price that will be wrong by spring.
Strip the time flourishes, keep an expiry list in your source note, and set the updated date only when you change something real.
It ends with in conclusion and three calls to action.
End where the argument ends. If you need one action from the reader, decide what it is and write the sentence yourself.
Questions
Will search engines penalize a blog post written with AI?
The major engines say in their own guidelines that they rank content on whether it is helpful and original, whatever tool produced it. In practice, a post made from the model's defaults is neither helpful nor original, because it is the generic answer the reader already skipped. The angle and the evidence are what make a post rank and what make it worth reading, and both have to come from you.
How long should it take to write a blog post this way?
For a 1,000-word post with your material ready, expect an hour to ninety minutes: ten minutes briefing, a few minutes for the outline and draft, and most of the rest judging and editing. Gathering the raw material is the part that varies, and it is also the part that determines whether the post is any good. If it takes you twenty minutes end to end, read the post again, because something generic probably slipped through.
Should I disclose that I used AI on my blog?
Decide a policy for your site and apply it consistently. A short note that explains your process once, linked from each post, is enough for most personal and company blogs, and it builds more trust than it costs. Any publication or platform with its own rule overrides yours.
Can I use the assistant to write the title?
Use it to list what the post could be called, after the post is finished, and treat the list as a map of formulas to avoid. Model-suggested titles tend toward the shape the reader has seen a hundred times. Write the title yourself, as a claim, using the same test you applied to the headers.