Content writing

What content writing is and where to start

Content writing means making text, images, audio or video that moves one specific reader one specific step forward. Its real work is not writing but deciding: what gets said, to whom, and why now.

Lesson 1 of 8
Beginner
Free, no signup

Five formats doing one job

A format is where and in what state the reader consumes the thing. The goal at the centre stays fixed; the formats move around it.

The reader goal the format is chosen after this

Articles and guides
Social media
Video and scripts
Email
Product and sales copy

No business needs all five, and someone starting all five at once usually keeps up none of them. Pick the two where your audience actually is.

Last checked: 3 October 2026 Facts and tool names in this lesson are re-checked against their sources on this date.

What exactly is the job called content writing

Content writing means making something that moves one specific reader one specific step forward. The definition is short and the word "specific" carries all of its weight, twice. If it is not clear who it is for and which step it moves them along, what you made is not content, it is filling a gap.

In practice the job has three parts and only the second one looks like writing. First the decision: what gets said and why now. Then the making: writing, recording, designing. Last the delivery: where this thing is published and how it reaches the person it was made for. People who do this professionally spend most of their time on the first part and the third. Beginners spend nearly all of theirs on the second, then wonder why nobody read their good piece.

One common assumption is worth dropping here: content writing is not a synonym for writing articles. An FAQ page answering ten real customer questions, a two minute video showing how a product is installed, and an email telling a buyer what to do after the purchase are all content, and all three usually do more work than a twelfth article.

Four media discs, a page, a photo, a wave and a film strip, orbiting a single white step

What content actually does for a business

Three things, and none of them is a direct sale. First, it shows someone who does not know you yet that you can do the work; the thing usually called trust, which in practice means the reader got their question answered and now guesses you have the answer to the next one too. Second, it takes repeated questions off the sales side. Third, it builds an asset: a page written this year that is still working next year, unlike an advert that stops the moment the budget does.

Now the honest version. Content either accumulates quietly or burns fast, and there is no third mode. Accumulating content answers a question that gets asked every month; it starts slowly, nothing happens for a long stretch, and then it quietly becomes a fixed share of what comes in. Timely content is attached to something that matters right now: a piece of news, a price change, an occasion. Its peak is sharp and two weeks later there is nothing left of it.

Neither is better than the other and the common mistake is not choosing. Someone writing only timely pieces starts from zero every month. Someone writing only accumulating pieces cannot sit through the early months and gives up. The mix that works for most small businesses is plain: put the weight on accumulating content, and write timely pieces only when there is genuinely news.

Two kinds of content that get confused

Both are legitimate and built for two different jobs. The trouble starts when you write one and expect the result of the other.

Timely content

Attached to news, a change or an occasion
Peaks within days, then it is over
Written fast and stale just as fast
Its home is social media and the newsletter

Accumulating content

Attached to a question asked every month
Shows nothing for a long while, then holds steady
Written slowly and needs updating now and then
Its home is a site page and a guide

Neither wins, and accumulating content is not free either: it needs maintenance and it goes stale if it is never updated. What ruins the work is not choosing.

Pick the format from the audience, not from the trend

Every so often a format becomes fashionable and everyone runs at it. The question almost nobody asks is this: where and in what state does my reader consume this? Someone hunting mid task for the cause of a technical error does not want an eight minute video, they want text they can search inside. Someone scrolling Instagram in bed at night is not going to read a two thousand word article.

So the right order is: first the thing the reader wants done, then the place they do it in, then the format. With that order, a lot of decisions people argue about for hours answer themselves.

There is one exception worth stating plainly: sometimes the format is set not by the audience but by what you can actually sustain. If you are not comfortable on camera, a bad video is worse than good text. That constraint is real, and hiding it only means six months spent on something that never gets published.

Where exactly to start

Not from a topic list. Start from two columns, one of which usually never gets written down at all.

Column one: the things you actually know, meaning the things you learned by doing the work rather than by reading about it. Column two: the things customers actually ask, in the words they use themselves. The source for column two is already in your hands and nobody goes looking for it: WhatsApp messages, questions asked on the phone, support tickets, and whatever a customer asks right before buying. Where the two columns meet is a list no keyword tool can hand you, because no tool knows what you happen to know first hand.

Take the first ten pieces from that overlap and write them in an order that answers the most repeated questions first. Do not write to finish the list; write to find out which ones get read. After ten you will discover a third column yourself: the things you assumed mattered and did not.

Where a topic worth writing sits

The overlap of two lists, and neither of them on its own.

What you actually know

What you learned by doing the work
A mistake you made and then fixed
A number you measured yourself

A topic worth writing

A repeated question you can answer first hand
Something a competitor said vaguely and you can say precisely

What customers actually ask

Questions repeating in messages and calls
What gets asked right before a purchase
A complaint that repeats after the sale

Either list alone misleads: one produces a piece nobody reads, the other a piece a plain search already finds. Only the overlap is something only you can write.

What actually sinks content

Three things, and none of them is the quality of the writing.

First, irregularity. An average piece published every week beats an excellent one that turns up once a quarter, because you get the practice and the reader learns that something is here. Second, having no single point: a piece trying to say everything says nothing. Third and worst, writing about something nobody asks. That last one usually happens when the topic came off a keyword list and nobody asked what the person searching that phrase is actually after.

And one thing that does not sink content although everyone worries about it: length. A short piece that answers beats a long one circling itself. Length should come from the question, not from a number on a checklist. If the answer is three paragraphs, write the three paragraphs and stop.

The fast path, with AI

There is one task that takes hours and almost nobody has the patience for it: reading a hundred real customer messages and counting which question repeats, and in whose words. That is exactly where a good language model takes the work off you, on one condition: do not let it add anything of its own.

Gather the raw material: WhatsApp messages, phone questions you wrote down, tickets, comments. Strip names, numbers and anything that identifies a person. It takes five minutes and cannot be undone later.
Pick a cheap, fast model. This pass is mechanical and needs no judgement, so a light model of the Flash class is enough, and the current pick is listed in this site AI section. Keep the expensive model for decisions, not for counting.
Run the recipe below. Its main constraint is that the model may use only words present in the input, and must put anything it added in square brackets, so you can see at a glance where it guessed.
Put the output next to your "what I actually know" column. Every repeated question you can answer first hand is a topic. The rest of the list is a list and is not a topic yet.
If a question repeats but you have no first hand answer, either say plainly that you do not know or leave it alone. Both beat publishing a rewrite of what already exists.

Copy-ready recipe

These are real customer messages from a {type of business}:

{paste the messages}

Rule: use only words that appear in this text. Put anything you had to add, that is not in the text, in square brackets.

1) Group the questions and write how many times each group repeats.
2) For each group, quote three phrases verbatim in the customers own words.
3) Say whether each group is asked before the purchase or after it.
4) Separate any group that appears only once and put it at the end.
5) Do not suggest any topics. Give me only these four things.

Before you trust the output: The model count is approximate and it drops things on long inputs; if a number from this output is going to be published anywhere, count it yourself. More importantly, this output tells you what customers asked, not what is worth writing. That decision stays with you, because the model does not know which question you can answer better than anyone else.

AI in this kind of work

At this stage AI is useful where few people reach for it and harmful where everyone does. The useful end is the top of the job: you have a pile of real customer messages and questions that would take hours to read and sort. The harmful end is the middle, the decision about what is worth saying. The model does not know your market, and instead of choosing it hands you the average of what everyone has already written.

Tools that actually help

Claude A fit for reading a pile of raw messages and grouping it without adding anything, and it holds the "use only words from the input" constraint well. Iran is on neither of Anthropic two supported country lists; we read that on Anthropic own page.
Gemini It understands Persian itself, so Persian customer messages need no translation first. 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 You supply the sources yourself and the answers stay tied to them, which is exactly what this job wants. Google own help says it works in the same regions the Gemini app works in, and Iran is not on the supported countries list.

Where it backfires

Two risks, and the first is the very thing most people do first. If you ask a model for a topic list without your own raw material, what you get is the average of the pages that already exist; publishing that in bulk is the pattern Google spam policies describe as scaled content abuse, and Google own guidance says its focus is unhelpful content produced at scale rather than the use of AI. The second risk concerns that raw material: customer messages are personal data, with names, numbers and complaints in them, and whether your conversation is used for training depends on the plan and the settings. Anthropic states its rule for consumer products on its own page. Strip the identifiers before you paste.

Sources: Google Search Central: spam policies for Google web search Google Search Central: Google Search and AI-generated content Anthropic: supported countries and regions Google: where Gemini Apps are available Anthropic: is my data used for model training

Where this advice stops

This lesson is a map, not the skill. Two things are deliberately left out. First, ranking: choosing terms, page structure and competing in Google results belong to the SEO path and are only pointed at here. Second, per platform rules: Instagram, Telegram and LinkedIn each behave their own way and change month to month, so any number written about them here would be wrong within months. And one boundary larger than both: content does not fix a product or a price. If what you sell solves nothing for anyone, content only tells more people faster that it solves nothing.

From our own work

The page you are reading is an example of the claim above. Every lesson in this section is a data file with two mandatory fields, and the section validator rejects a lesson missing either one: the date the facts were last checked, printed just above, and a field named "where this advice stops working". The reason is what the second section said; content meant to accumulate needs maintenance, and we chose to make that maintenance a technical condition rather than a promise.

Real follow-up questions

Do I need to be a writer to produce content?

No, but you do need to be able to explain one thing clearly. Most of the writing that works in technical fields comes from people who know the work and write plainly, not from writers who do not know the subject. What is genuinely required is practice: everyone first ten pieces are bad and there is no way past them except writing them.

How many pieces should I publish a week?

Whatever number you can keep up for months, and not one more. For a single person with another job, one a week is realistic, and anything above it usually reaches zero within two months. Regularity beats volume, because it builds your practice and it teaches the reader to expect something.

Should I rewrite a competitor content?

It works right up until somebody looks properly. A rewrite means publishing the average of what everyone has said, without giving any reason to read your version. If you are short of time, instead of rewriting a page, write one small thing you experienced yourself; it is shorter and it lasts longer.