Writing a blog article
Writing a blog article is three decisions and one execution: which question, from which angle, for whom. The execution takes half a day and the decisions half an hour, and it is that half hour which settles whether anyone reads it.
- Lesson 3 of 8
- Beginner
- Free, no signup
Where an article travels, from end to end
The first two stations happen on paper, and that is where most articles fail, not in the writing.
-
1
The idea
The question a reader types, in their own words.
-
2
The angle
What you know that the top results do not say. If there is none, do not write it.
-
3
The outline
The headings, and what question each section answers.
-
4
The draft
Fast and bad. This is not the place to judge sentences.
-
5
The edit
Cut, check, open every link. Three separate passes.
-
6
Publish
And then reading it on a phone, at the real width of the site.
The path is not one way. The editing station usually sends you back to the outline, and going back is not a sign of bad writing.
Last checked: Facts and tool names in this lesson are re-checked against their sources on this date.
Where topics come from, and what an angle is
An angle is what you know about a topic that the first page of results does not say. An idea is different: the idea is the question a reader types. An article that gets read has both, and most articles have only the first.
Ideas are not scarce. Customer messages, the question that comes up on the phone three times a week, the thing you explained to somebody last week: all of them are ideas. The first lesson of this path has a method for collecting them and it is not repeated here.
The angle is harder, because it comes from experience rather than from searching. Before you open the editor, finish this sentence: "what I know about this question and the top ten results do not say is that ...". If the sentence will not finish, do not write the article. A ninth version of an existing text, however cleanly written, has nothing for someone who has seen the eight before it.
There is a simple way to find an angle. Read the top three or four results for the same question and look for the place where all of them go quiet. That place is usually one of three things: the real cost, the failure that tends to happen, or the condition under which the advice stops working. That is your angle, because somebody who has not done the work cannot write about it.
There is a simple test for an angle too: say it to somebody who does not know your line of work. If they say "well, obviously", you have no angle. If they ask "meaning what" or "why", you have one.
Put the answer first, not last
A reader arriving from Google has one question and no patience. So the first two sentences of the article answer that question: not an introduction to the topic, not an explanation of why the topic matters.
This is not the same as writing short. The answer comes first and then the article earns the rest: why the answer is right, where it differs, where it does not work at all. A reader who got the answer and stayed is worth more than one who had to hunt down to the sixth paragraph, and half of those left before it.
The common Persian habit is the reverse. The first paragraph warms up, the second says how important the subject is, and the answer sits halfway down the page. That habit came into blogging from formal writing, and in blogging it is expensive, because nobody there is obliged to read to the end.
The order that works, after the answer: depth, meaning why and how; then the exception, meaning where this is not true; then the reader next step. And if you end the article with a summary section that repeats the sentences above it, delete that section. Someone who read that far does not need a summary.
A short example. An article about what a website costs could open like this: "Having a website is essential for every business, and one of the questions that always comes up is the cost of building one." Or like this: "A simple company site and an online shop do not cost the same, and most of that gap does not come from the number of pages; it comes from work that never gets named on the invoice." The second sentence may not be complete, but after reading it the reader knows what they are going to get. The first only confirmed that they have a question, which they already knew.
Write the outline before the draft
An outline is the list of headings plus one sentence under each saying which question that section answers. That is all of it. If the sentence under a heading will not get written, the section does not exist yet, and it will not come out right in the draft either; it turns into a few paragraphs of generality.
An outline does two things that are invisible while you write. First, it stops repetition: when you see two sections answering the same question, you delete one there and then, not after a thousand words. Second, it keeps the order changeable, because moving two lines is easier than moving two written paragraphs, and a writer who spent four hours on a section will not delete it even knowing they should.
The draft comes after the outline and it is supposed to be bad. Write fast and do not judge the sentences. Writing that slows down almost always has one cause: you are writing and editing at the same time. Those two jobs do not happen together, and separating them is the largest saving of time available in this work.
One practical rule: if you get stuck in the middle of a section, the problem is probably not the sentence, it is the outline. Go back and see which question that section was meant to answer. Usually it turns out that either the question was not clear or you do not have the answer.
How to choose the section headings
A section heading is written for two people: someone scanning the article to see whether it is any use to them, and someone landing in the middle of the page straight from search results. Neither of them has any use for "introduction" and "final words".
The practical rule is simple: a heading is the task or the question a reader types. "How much does an online shop cost to build" is a heading; "a look at the costs" is a label. An abstract noun does not tell the reader what they will find under it, which is why the scanner passes over it.
Something that gets said less: vary the grammatical shape of the headings. If all six start with "how to", the article outline reads like a software manual. Let one be a question, one a task, one a statement.
And one point that carries weight in modern templates: headings are not only headings. Many sites build a jump index, anchor links and sometimes a figure out of those same h2 elements, and every one of them shows exactly the words you wrote. A heading that says nothing says nothing in all of those places too.
The number of headings is set by the question as well, not by a template. If you added headings to reach a count, those sections usually say what the previous one said, and the person scanning sees exactly that repetition.
Two sets of headings for the same article
Both columns are the heading list of one article about what a website costs.
Headings nobody types
- Introduction
- A look at the contributing factors
- The importance of choosing well
- Final words
Headings that do get typed
- What a simple company site costs
- What pushes the price up
- Ready made theme or custom build
- What to ask before you order
The good column has a rule of its own: a heading that asks a question the section does not answer is worse than an abstract one.
When is a post finished
A post is finished when it has answered the question and also answered the next question that the answer creates. Not when it reaches a thousand words.
Word count is a poor measure because the right length is set by the question, not by the writer. The self assessment questions Google publishes for content are all about the reader and none of them is about length. So if you are stretching a post to hit a number, you are damaging the very thing you wrote it for.
The editing pass has three jobs and they do not get done at the same time. First, cutting: every sentence that says nothing new goes, and the second pass is almost always the shortening pass. Second, checking: every number, every name and every claim has to connect to something you saw yourself. Third, the links: open every one. A link to a page that does not exist is the worst kind of carelessness, because it is the only error the reader sees with their own eyes.
There is a common trap too: reading the text in the same editor it was written in. Read it at the real width of the site and on a phone. A seven line paragraph is tolerable on a monitor and a wall on a phone.
Two habits work better than any technique. First, read the text out loud; wherever you run out of breath, the sentence is too long. Second, if you have the time, leave the article overnight. A text that was perfect last night usually has two spare paragraphs in the morning.
And the opposite boundary, which matters as much as the first: if the question needs depth and you wrote four hundred words, the post is not finished. It is just short.
Is the article ready to publish
Must be there
- The answer in the first two sentences
- One thing that only comes from experience
- The place where this advice does not work
- Every link opened and alive
Must not be there
- A warm up paragraph before the answer
- A heading that is only a noun
- A number whose source you have not seen
- A summary repeating what is above it
This list does not tell you the article is good. It only stops an unfinished one from going out.
Where this lesson stops and SEO starts
So far we have talked about the article itself. Getting that article seen is a different job with a path of its own. Search intent decides what shape of page answers this question at all, and whether an article is the right shape. Content SEO is about authority and topic coverage, and on page SEO is about the title, the links and the structure of the page.
The order matters too, and it is usually done backwards. Decide first what the article says and who it is for, then go to those three. An article built around a keyword from the start usually has nothing to say, and the reader works that out sooner than Google does.
And a boundary that gets said less: some questions do not want an article at all. If the honest answer to a question is two sentences, a thousand word article for that question is a bad page however well written it is. That belongs in a short answer on an FAQ page, or as one section of a larger article.
The fast path, with AI
What most people do is hand the topic to a model and get back a whole article, and the result is text with nothing of its own in it. The professional move is to change the order: outline first, then divide the work, and let the model itself propose the division. That is, you ask it to say which sections it must not write.
- Write the angle yourself in one sentence and the reader situation in another. The model knows neither, and if you do not write them it will invent them.
- The outline needs a strong model, because that pass is judgement. Drafting the mechanical sections needs only a cheap, fast one. The current ranking for each is listed in this site AI section.
- Run the recipe below. Its real point is rule four: the model must mark every heading E or M, experience or mechanical, and for the E ones only ask you questions.
- Write the experience sections yourself, using the answers to those questions. That is the only part nobody else can write, and it is what separates the article from the rest.
- At the end, read the whole text yourself once and remove every sentence you could not have said yourself. This pass is not optional.
Copy-ready recipe
The question the reader types: {question}
My angle: {what I know that the top results do not say}
The reader: {who they are and what situation they are in right now}
1) Write the outline: between four and seven headings.
2) Every heading must be a task or a question this reader types, not an abstract noun. Do not write "introduction" or "conclusion".
3) Under each heading, write one sentence saying exactly which question that section answers.
4) Mark every heading E or M. E means writing it needs first hand experience; M means it is mechanical and comes from general knowledge.
5) Write nothing for the E headings. Instead ask me two questions whose answers you would need.
6) Write no number, date, price or brand name unless I wrote it above. If one is needed, leave a blank.
7) At the end, write where this article is of no use to this reader.Before you trust the output: The model does not know what you did or what you saw. Any sentence it wrote itself in the experience sections is manufactured experience, and that is exactly what both readers and search engines penalise. So two things are settled before publishing: you wrote the marked sections yourself, and you saw every number, name and link that is left in the text.
AI in this kind of work
In article writing, AI is good at structure and has nothing at all in the way of experience. That one sentence settles the division of labour: the outline, the ordering, the summarising of sources you gathered yourself and the editing pass can all go to it; the thing that separates your article from eight others cannot. A blog article is where that difference shows more than in any other text, because what makes an article worth reading is exactly the part the model cannot supply.
Tools that actually help
- Claude It holds a negative constraint across a long output, and the recipe above counts on exactly that: not writing the marked sections and not writing a number we did not give it. Iran is on neither of Anthropic two supported country lists; we read that on Anthropic own page.
- Gemini It understands Persian itself, so the editing pass can run directly on the Persian text: which sentence makes the same point twice, where the tone turned bureaucratic. Google own page says the app works in over 230 countries and territories and more than 70 languages, and Iran is not on that list.
- NotebookLM It is useful for an article that leans on sources, because it works only on the files and links you gave it, which closes off a whole class of errors. Google own help says it runs in the same regions where the Gemini app runs, and Iran is not on that list.
Where it backfires
Two risks, and both are more expensive in a blog article than elsewhere. First, a model will happily invent a number, a date or the name of a study, and an article is where a reader picks that number up and repeats it somewhere else, so your mistake travels. Second, Google line on AI content is widely misquoted: Google has not banned the use of AI, it has targeted scaled production of unhelpful content, and its own spam guidance says exactly that. So an article written with a model that has something to say is fine, and an article that is only filler is not, even if a person wrote it.
Sources: Google Search Central: Google Search and AI-generated content Google Search Central: spam policies for Google web search Anthropic: supported countries and regions Google: where Gemini Apps are available
Where this advice stops
This lesson is about the article itself and not about getting it seen; that job lives in the SEO path, and without it even a good article may find no readers. Two more boundaries. First, a blog fixes nothing for a business with nothing to say; writing regularly about nothing is only regular nothing. Second, not every question wants an article: where the honest answer is two sentences, a short page or one section of another article is the right option, and a separate article only adds one more weak page.
From our own work
On this site, section headings are not only headings, and you can see that in the theme code. For every blog article the theme builds two things at render time out of those same h2 elements: the jump index beside the article, and a figure whose node labels are literally the heading text. Even the type of figure is chosen by the headings; an article with a three column comparison table gets a comparison shape, one with three to six short headings gets a constellation, and the rest get a numbered rail. There is no hand written data behind any of it, so if you write a heading that says nothing, exactly those words land on the map of the article and you see for yourself that the section has no point. It is the most practical feedback on headings we have had.
Real follow-up questions
How long should a blog article be?
As long as it takes to answer the question and cover the next one. Any number you are handed came from averaging competing pages and says nothing about your question. There are two practical boundaries: a text stretched to reach a number loses the reader, and a text that leaves a deep question after four hundred words is not finished.
Should I write a second article on a topic I already covered?
The default answer is no. Two pages answering one question compete with each other over it instead of reinforcing each other, and usually neither replaces the earlier page. If you have something new, improve the existing article; a second one is right only when its question is genuinely a different question. The mechanics are in the article on keyword cannibalisation.
Is there a problem with writing the article using AI?
It depends which piece you hand over. Google own guidance targets the scaled production of unhelpful content, not the writing tool, so an article that has something to say is fine even with a model helping. What is not fine is handing over the sections that need first hand experience, because there the model guesses instead of knowing, and its guess looks like fact.