Content writing

Learning copywriting

Copywriting means writing text that asks the reader for one specific action: buy, sign up, get in touch, download. Whether it is good is not settled by taste; it is settled by whether people did the thing.

  • Lesson 2 of 8
  • Beginner
  • Free, no signup

The four steps a piece of sales copy has to move a reader through

The AIDA frame. It belongs to no course and no brand, and it was written down in advertising literature long before the internet.

  1. 1

    Attention

    The headline job. If it fails here, the rest of the text effectively does not exist.

  2. 2

    Interest

    The reader sees that this is about their situation, not about you.

  3. 3

    Desire

    They see what changes for them afterwards. This is the place for a benefit, not a feature list.

  4. 4

    Action

    They know exactly what to do and what it takes. One action, not three.

one specific action

The narrowing shows the order of the steps only; there is no numeric ratio in it. Nor is the frame a formula: copy that fills these four bands mechanically does not get read.

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

How copywriting differs from writing

Copywriting is writing text that asks the reader for one specific action. The definition is simple and it has a hard consequence: the measure of quality comes from outside the text, not from inside it. A beautiful sentence after which nobody does anything is a bad sentence in this craft, however fond of it you are.

So the question changes. In a teaching piece we ask whether the reader learned something. In copy we ask whether the reader understood what to do next and why now. Every small decision in the text, from sentence length to paragraph order, is an answer to that second question.

There is another difference that gets less airtime: in copy, cutting takes longer than writing. The first draft is always longer than what is needed, and the second pass is almost always the shortening pass. If your text got longer after an hour of work, you probably moved in the wrong direction.

A funnel of paper sheets with a red headline bar at the top, green lines and one blue button at the bottom

How a headline actually gets written

The headline is the only part everyone reads, and it has exactly one job: get the reader to the next line. Not to sell, not to summarise the whole page. That is all.

Three things burn a headline. First, being generic: "a comprehensive solution for your business" is about nobody, so nobody claims it for themselves. Second, cleverness: a headline that has to be read twice to be understood does not get read once. Third, promising something the rest of the page does not deliver; that one earns the click and burns the trust, and burnt trust is not recovered by the next headline.

The practical method is plain and needs no taste. Instead of hunting for one good headline, write seven, each making a different promise rather than seven wordings of one promise. Then ask each one a question: after reading this, what exactly can the reader do? Any headline whose answer is "know more" gets deleted. Usually two of the seven survive, and choosing between two is not hard.

Feature or benefit: which one goes in

A feature is something the product has. A benefit is what changes for the buyer because of that feature. "A six stage filter" is a feature; "the water tastes different and the kettle stops scaling" is a benefit.

The common rule says write benefits, not features, and it stops halfway. A benefit with no feature behind it is an unsupported claim, and a buyer who is comparing goes looking for exactly that feature so they can put two options side by side. The order that works is: benefit first, the feature immediately behind it, and any number you have right there, so the claim can be checked.

One thing we keep noticing in technical sales: the more expert the buyer, the heavier the feature side gets. For someone who sets up servers themselves, writing "fast" says close to nothing and writing a model and a number says something. So the audience sets that balance, not a general rule repeated in every course.

Two ways of writing the same product

Both columns describe the same water filter; only the order differs.

Features only

  • Six stage filter
  • Removes water hardness
  • Three litre tank
  • Valid warranty

Benefit, with the feature behind it

  • The water tastes different: six stage filter
  • The kettle stops scaling: hardness removed
  • A day for a family of four: three litre tank
  • If it breaks we come to you: warranty terms on this page

The second order has a condition too: a benefit you cannot deliver every time is a lie with better rhythm. If you have no number, do not write one.

A real rewrite: before and after

The example is a laptop repair shop, because it is simple and nobody in that trade is making grand claims. The original, the kind of text you see everywhere:

"Provision of specialist laptop repair services drawing on an experienced team and quality parts, together with a valid warranty. Your satisfaction is our pride."

And the rewrite:

"Laptop will not turn on? Diagnosis is free, and by six the same day we tell you what is wrong and what it costs. If it cannot be repaired, you pay nothing."

Four things changed and none of them is about being prettier. First, the opening line names the reader situation instead of introducing the company; the person reading this has a laptop that will not start right now. Second, adjectives gave way to a checkable commitment: "experienced" and "quality" cannot be disputed, "by six the same day" can. Third, a clear financial condition arrived, which moves the risk off the buyer. Fourth, "your satisfaction is our pride" is gone, because it is about the writer rather than the reader.

And now the condition that teaching material usually leaves out: the second version is only better if every sentence in it is true. If you cannot answer by six every time, do not write it. A specific promise that goes unmet does more damage than a vague sentence, because the buyer knows exactly where you lied.

Why plainness pays off more in Persian

Our position is plain: clarity sells more than cleverness, and in Persian the gap is wider than in English. The reason is not cultural, it is a writing habit. Formal Persian has been written for years with heavy verbs and bureaucratic constructions, and that register carried over into advertising copy. A Persian reader recognises the tone and has learned, without thinking about it, that there is usually nothing behind it.

The practical consequence is short. Use plain verbs. Then take out the adjectives that measure nothing, such as "professional", "high quality", "leading" and "the best", and see what the sentence lost. Usually nothing, which is the proof that they were saying nothing to begin with.

For us this is not personal taste, it is a written rule. The site publishing this lesson keeps a content rules file that explicitly bans twenty two Persian phrases, and no text goes out without being checked against that list. The editorial policy is the public version of the same rules. The reason for the strictness is simple: those phrases are not merely useless, they signal to the reader that the writer had nothing to say and is filling space.

Before you publish the copy

Final check on a piece of copyBefore publishing

Must be there

  • The reader situation is named in the first sentence
  • One specific promise whose truth can be checked
  • Benefit first, feature immediately behind it
  • One action, with everything needed to take it

Must not be there

  • An adjective that measures nothing
  • A promise you cannot keep every time
  • Two or three calls to action on one page
  • Praising yourself instead of talking about the reader

This list does not make bad copy good, it only stops bad copy from going out. The real work is still choosing the promise.

The fast path, with AI

What beginners do is ask a model for "ten catchy headlines" and get back ten wordings of one promise. The professional move is the reverse: force the model to vary along the promise axis rather than the wording axis, then choose between promises yourself. The difference looks small and it changes the output entirely.

  1. Before anything else, write in one sentence what this page promises. If you cannot, the problem is not the text and no model will solve it.
  2. This one needs a strong model, not a cheap one. Judgement is required here and light models fall back on adjectives. The current ranking is listed in this site AI section.
  3. Run the recipe below with the current page text. It has three constraints that matter: a different promise per headline, the assumed reader beside each one, and an explicit ban on empty adjectives.
  4. Delete the headlines whose answer to "what do they do next" is "know more". Usually half of them go right there.
  5. From what is left, cross out any headline whose promise you cannot keep every time. The model cannot do this pass, because it does not know what you can actually deliver.

Copy-ready recipe

This is the current text of the {product or service} page:

{current text}

Reader: {the reader in one sentence}
What they should do: {the action}

1) First write what this text promises. If it makes no clear promise, say that.
2) Write seven headlines. Each one makes a different promise, not seven wordings of one promise.
3) Beside each headline, write what situation you assumed the reader is in.
4) For each headline, write what exactly the reader can do after reading it.
5) Do not use the words "professional", "best", "high quality", "leading", "comprehensive" or "solution".
6) Do not write any number that is not in the original text. If a number is needed, leave a blank.

Before you trust the output: The model does not know what you can actually deliver. Every headline that comes out of this list is a promise, and it does not get published until you have confirmed you can keep it every time. That is the last step of this recipe and it is not optional: the model produces a sentence that sits well, and you decide whether it is true.

AI in this kind of work

In copywriting, AI is a good sparring partner and a bad judge. Sparring partner means it hands you ten angles in a few minutes that you would not have reached in an hour. Judging means deciding which angle is true and which promise you can keep; that needs knowledge of your business, and the model has none. If you take one thing from this section, take that division of labour.

Tools that actually help

  • Claude It holds negative constraints well across a long output, and the recipe above is built on exactly that: a banned word list and a ban on invented numbers. Iran is on neither of Anthropic two supported country lists; we read that on Anthropic own page.
  • Gemini It understands Persian itself, so you can ask which sentence carries the bureaucratic tone and where a heavy verb sits, without translating the text 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.
  • ChatGPT The most common choice and acceptable for producing different angles. We make no claim here about its availability or payment: OpenAI supported countries page answers our server with a 403, and we do not write down what we could not read ourselves.

Where it backfires

Two risks, and the first is specific to this job. A model will happily invent a warranty, a delivery time, a discount or a number you never offered, because statistically that is where such a sentence goes. Copy published with that error is no longer marketing, it is a false statement about your terms of sale, so every number and every promise in the output has to be checked against what you actually do. The second risk is scale: producing this same copy for dozens of near identical landing pages is the pattern Google spam policies target, and Google own guidance says its focus is unhelpful content produced at scale rather than the use of AI.

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

Where this advice stops

Copywriting does not make up for what the product does not do. If your price is double the market and you have no reason for it, no headline solves that; better copy only reveals it faster. Two other boundaries. First, this lesson is about the text and not the page: layout, images, load speed and trust signals carry a large share of the result and belong elsewhere. Second, these same techniques work on a false promise, and at that point it is not copywriting any more. The line is simple: if you cannot keep a promise every time, do not write the sentence.

From our own work

An example from this site, of exactly the kind this lesson is about. The guarantee and refund page has two columns, and the second is a list of what we do not guarantee: domain, hosting and plugin costs, up to a specific numeric position in Google. From an advertising point of view that column looks like a loss, and in practice it is the only thing making the first column believable. A promise that excludes nothing is not a promise.

Real follow-up questions

How many headlines should I write to get a good one?

The number does not matter, the variety does. Five headlines making five different promises beat twenty repeating one promise in different words. If after a few you notice they all say the same thing, the problem is not the writing; you have not yet decided what you are promising.

Do frameworks like AIDA still work?

As a checklist yes, as a template no. It is useful after writing, to see whether all four steps are present or you skipped one. If you sit down and fill it band by band from the start, you get copy whose structure is visible from a distance, and that visible structure is what pushes the reader away.

How should I write product descriptions for an online shop?

With the same order as this lesson, but far shorter. One sentence saying who and what this is good for, then the specifications as an actual list, and at the end the answer to the question a buyer asks right before the buy button: size, compatibility, delivery time. A description that copied the technical specs from the manufacturer site helps neither the buyer nor the page.