Algeria AgencyDigital marketing
Reworking text you already have: three passes, not one
You have forty pages written over five years by six people. How to rework them without rewriting everything, by looking for one thing at a time.
The article accompanying this one teaches how to write a page: order before style, the objection before the argument, the number instead of the adjective. It assumes a blank page. That situation is rare.
The ordinary situation is the reverse. There is already a site, a brochure, product sheets, presentation text, written over several years by several people, some of whom have left. The whole is uneven, contradicts itself in places, and nobody knows where to start.
This article is about that rework. It begins by setting aside the solution always proposed — rewrite everything — and puts in its place a method of three passes, each looking for one thing: what can go, the words that commit you, and what the text assumes without saying.
The principle is ordinary and it makes all the difference: you do not see what you are not looking for. A general read catches whatever the reader was sensitive to that morning. Three targeted passes catch three times as much, in the same total time.
You are not going to rewrite everything
It is the proposal that always arrives, and it is bad for three independent reasons. It costs several times what a rework costs. It takes months during which nothing is corrected. And it throws away text that works, because nobody took the time to separate what is fine from what is not.
The fourth reason is the most serious and nobody states it: a complete rewrite also erases what was true. Text written by people who knew the trade contains exact details — a real lead time, a technical constraint, an exception — that an outside writer cannot reconstruct and that vanish silently.
What is needed instead is a sort, and sorting requires a criterion. The only one that holds: does this text do work today that somebody expects of it? A product sheet thirty people consult a month works; a "our philosophy" page nobody opens does not, and the best text in the world there will change nothing.
So the rework proceeds in an order dictated by use rather than by the structure of the site. The most-read text first, the text answering a question next, and the rest — often half of it — is not reworked at all: it is removed or left as it is.
That last point is worth saying plainly because it frees you: most of a business’s text does not need to be good. It needs to be accurate and to do no harm. Concentrating effort on five pages and leaving the other thirty-five alone is a rational decision, not a surrender.
The inventory: where your text actually lives
Before reading anything back you have to know what exists, and a business’s text almost never lives only on its site. It is scattered across places nobody thinks of as text.
Five places cover nearly everything. The site, obviously. The product sheets, often in a spreadsheet or a sales tool. The documents sent to customers: quotations, proposals, terms. The template messages salespeople have been copying for years. And the descriptions on third-party platforms, directories and business listings, which nobody has reopened since creating them.
The last two hold the worst surprises, for the same reason: they were written once, quickly, by somebody no longer there, and nothing brings them back to attention. A template message copied six hundred times a year is the most-read text in the business and it has never been read back.
Build the list with three columns only: where it is, roughly how many people read it, and who wrote it if known. The second column can be approximate — "many", "a few", "nobody" is enough — and it is what orders all the work that follows.
That inventory takes half a day and is already useful before any reading back, because it usually reveals two or three heavily read texts nobody remembered, and one or two contradictions visible to the naked eye between two places.
Three passes, never one
The core of the method rests on an observation you can verify on yourself: you do not see what you are not looking for. A reader asked to improve a text finds what jars on them, and what jars depends on their mood, their training, and what they read the day before.
The correction is never to look for more than one thing at a time. Three passes, each with a single question: what can be removed? which words commit us? what does this text assume the reader already knows? Nothing else is corrected during each pass, including obvious errors, which are noted and dealt with at the end.
The discipline looks slow and it is faster. A targeted pass over one page takes five minutes because the question is simple; three passes take fifteen. A general read of the same page takes twenty minutes, produces debatable style changes, and misses two of the three problems.
It has a second advantage, decisive when several people work: each pass can be done by whoever is competent for it. The first by anybody attentive; the second by somebody who knows your real commitments; the third by somebody who talks to customers and knows what they do not know.
And it makes the work verifiable. "Pass two has been done on these twelve pages" is a sentence that means something; "these pages have been reviewed" means nothing, and yet it is what most reports say.
First pass: what can go
The first pass looks for one thing only: what can be deleted without the reader losing information. Nothing else. You do not rephrase, you do not correct, you do not move — you remove, and you see.
Four categories recur everywhere and are spotted without literary judgement. Introductory sentences announcing what the paragraph will say. Adjectives that change nothing when removed — "true", "real", "unique", "complete". Repetitions from one paragraph to the next. And commercial politeness that commits nobody.
On the text we rework, this first pass routinely removes between a quarter and a third of the volume without a single piece of information disappearing. That is our own experience rather than a market statistic — but the order of magnitude is stable, and it surprises whoever wrote the text every time.
The gain is not brevity for its own sake. It is that what remains becomes visible: in a text lightened by a third, the real weaknesses appear, and above all the places where there was nothing to say become obvious because no filler is left to hide them.
A practical rule for this pass: cut first, judge afterwards. Removing a sentence and rereading is instant and reversible; wondering beforehand whether it serves a purpose produces a discussion that drags on and almost always ends in keeping it.
Second pass: the words that commit you
The second pass looks for words that create an obligation, and it does not judge style at all. Certain terms commit the business legally or commercially, and they are often written by somebody who did not know they did.
The list is short and memorable. "Guaranteed" implies a written guarantee and a duration. "Certified" and "approved" imply a body that did it and can be named. "Leader" and "number one" imply a ranking. "Free" implies no conditions at all, which is rarely true. And "delivery within 24 hours" is a promise a customer can hold you to.
The treatment is not to delete these words out of caution. For each there are two outcomes: either the fact exists and you write it beside — the name of the body, the duration, the condition — which makes the claim stronger than before; or it does not and the word goes.
The same pass handles figures, and that is the part most often defective. A percentage or a duration written four years ago by somebody who saw it somewhere has neither source nor validity. The rule is simple: any figure nobody today can say the origin of is removed, not corrected.
This pass has to be done by somebody who knows the business’s real commitments, never by a writer alone. It is the only one of the three that cannot be delegated outside, because the question is not "is this word excessive" but "do we do that".
Third pass: what the text assumes
The third pass looks for the unsaid, and it is the hardest to do yourself. It puts one question to each paragraph: what does this text assume the reader already knows?
The difficulty is structural. Whoever wrote it knows the trade, and everything they know seems so obvious it is invisible. They write "we attend within 48 hours" without saying from when, because in their head it is from the call — information the reader does not have and which decides everything.
Four unsaid things recur constantly and are worth hunting by name. When a lead time starts. What the price includes and excludes. What the customer must supply or prepare. And what happens if it does not work.
The method that works is to have the text read by somebody who does not know the trade, with one precise instruction: mark every place where they would have asked a question. Not comment, not improve — mark. The marked places are exactly the unsaid, and that person finds them in ten minutes where the author never will.
Failing an outsider, whoever answers the telephone does almost as well, because they spend their days filling those gaps aloud. Just ask them: on this page, what are people still going to ask me?
The title and the first sentence
After the three passes there is substantive work left, and it is concentrated on very little text: the title and first sentence of each reworked page. That is where the remaining time goes, and there is no better use for it.
The reason is mechanical. Those two elements are read by everybody, the rest by a minority. Improving a page’s third paragraph serves the few readers who reach it; improving the title decides how many people read anything at all.
For the title, one criterion is enough and it is checkable: does it contain what the page delivers, in words the reader would use? "Our solutions" delivers nothing. "Air conditioner repair in Algiers, attendance within 48 h" delivers three pieces of information and turns up in a search.
For the first sentence the criterion is simpler still: does it already answer the question that brought the reader? Most opening sentences introduce the business, when the reader wants to know whether you do what they are looking for. Invert that and the page immediately works better, with nothing else changed.
Fixing Arabic that was translated once
Many businesses had their text translated once, long ago, and never went back to it. The result is almost always recognisable, and it is not a question of language level.
The typical fault is Arabic that is correct but traced: the French sentence structure preserved, the turns of phrase translated word for word, and a written vocabulary a reader understands but nobody uses to search. The text is accurate and it is not natural, which is felt within three lines.
A more concrete problem for this market is added to it: the terms people actually type are not the ones a professional translator chooses. The commercial vocabulary used in Algeria differs from the standard one, and a page written in the second is correct and unfindable.
So the rework is not retranslation but having it read by somebody who writes Arabic for this audience, with two instructions: make the sentence natural, and replace technical terms with the ones a customer would use. It is shorter work than a translation and it changes more.
One practical point finally: never mix the two languages within a block of text, even when a technical term has no common equivalent. Write the locally used term and give the other in brackets if needed — the reverse produces a page neither reader reads comfortably.
What needs rewriting rather than correcting
The method has a limit and it needs recognising, otherwise hours go into improving a text that cannot be saved. Three signs indicate a rewrite.
The first is a text with no subject. It talks about the business in general, it would serve any competitor, and removing the name would change nothing. No reading pass creates a subject; you have to decide what the page says and start from there.
The second is a text describing an offer that no longer exists. Correcting the sentences one by one produces an incoherent assembly, because the structure itself assumed the old offer. Rewriting three hundred words is faster than patching.
The third is a text whose every paragraph is good and whose order is wrong — where the answer arrives fifth and the first four are context. That too is a reorganisation rather than a correction, and it is exactly what the neighbouring article calls ordering before writing.
In all three cases the rewrite covers a page, not the site. That is the difference from the proposal set aside in section 1: you rewrite what has to be rewritten, one page at a time, for a reason you can state.
Text written by a machine
A growing share of the text to be reworked was produced by a tool, and it is recognisable by precise marks rather than by a general impression. Spotting them saves time because such text is handled differently.
Three marks are enough. An abnormal regularity: every paragraph the same length, every list three items. A vocabulary of scale — "landscape", "universe", "ecosystem", "at the heart of". And above all the absence of any verifiable fact: no proper noun, no figure, no date, no exception.
The treatment is particular. The first pass removes far more here than elsewhere, often half, because such text is made of filler. But the third pass is the one that matters: the problem with this text is not its style, it is that it contains nothing the reader could not have guessed.
So the rework consists of adding what the machine could not know: your real lead times, your prices, the name of the city, the particular constraint of your trade, the case you refuse to handle. Generic text plus five true facts becomes useful text; the same text written better stays empty.
That also holds as a rule going forward, and it is unambiguous: these tools are useful for shaping and shortening, and they cannot produce the material, because the material is what only you know.
Stopping it happening again
A successful rework degrades within eighteen months if nothing changes in how text is produced. Three decisions are enough to hold it, and they are light.
The first is naming who approves, per text rather than in principle. Not "management approves communications", but "this person approves product sheets, that one approves documents sent to customers". A text with no named approver returns to its former state.
The second is the one-page guide the neighbouring article describes, reduced here to what a reading pass needs: the forbidden words and why, the register, the language to reply in, and the four unsaid things always to fill. One page, not ten, displayed where people write.
The third is a trigger, because without one nothing happens: half a day a year, on a fixed date, to pass over the five most-read texts. It is short because the substantive work has been done; it is what stops the accumulation starting again.
Add if you can a rule that costs nothing and prevents half the future problems: every new customer-facing text is read by a second person before publication, with the third pass’s question alone — what are people still going to ask me?
What we do, and what we refuse
What we do is bounded. We build the inventory of texts and their real use, we run the three passes in order, we rework the titles and opening sentences of the selected pages, and we leave behind the one-page guide and the list of approvers.
We refuse to quote for a complete rewrite of the site when asked. It costs more, takes longer, corrects nothing for months, and erases exact details written by people who knew the trade and that no outside writer can reconstruct. We propose the rework instead, and page-by-page rewriting where it is justified.
We also refuse to remove a claim without asking you what fact sits behind it. "Certified" may be accurate and merely missing the name of the body; deleting the word in silence is the easy move, and it costs you a true argument. The second pass is done with you, never for you.
And we do not promise that reworked text will convert better. What we commit to is checkable and sufficient: what you publish will be shorter, will contain no claim you cannot support, and will answer the four questions your customers ask anyway. There is also a first move you can make without us: open the template message your salespeople send every day, and read it.
Frequently asked questions
Do we have to rewrite everything?
Almost never. A complete rewrite costs several times more, corrects nothing for months, and erases exact details written by people who knew the trade — a real lead time, a technical constraint — that an outside writer cannot reconstruct. Rework the most-read texts instead, and rewrite page by page only where there is a reason you can state.
Why three passes rather than one good read?
Because you do not see what you are not looking for: a general read catches whatever the reader was sensitive to that morning. Three targeted passes — what can go, the words that commit you, what the text assumes — take the same total time and catch three times as much. They also let each pass go to whoever is competent for it.
Which words should we watch for?
The ones that create an obligation: guaranteed, certified, approved, leader, free, and any promise of a delivery time. For each there are only two outcomes — the fact exists and you write it beside, which strengthens the claim, or it does not and the word goes. Same treatment for figures: any figure nobody can trace is removed, not corrected.
How do we spot what a text assumes?
Have it read by somebody who does not know the trade, with one instruction: mark every place they would have asked a question. Not comment — mark. Hunt by name for four unsaid things: when a lead time starts, what the price includes, what the customer must supply, and what happens if it does not work.
How do we recognise text written by a machine?
By three marks: abnormal regularity — paragraphs of equal length, lists always of three items; a vocabulary of scale; and above all no verifiable fact, no proper noun, no figure, no exception. The problem is not the style but the emptiness: add five facts the machine could not have known and the text becomes useful.
Our Arabic was translated long ago. Should we retranslate?
Rarely. The usual fault is not language level but tracing: French structure preserved and standard vocabulary nobody types when searching. Have it read by somebody who writes Arabic for this audience, with two instructions — make the sentence natural, and use the terms an Algerian customer would use. Shorter than a translation, and more useful.
Where we come in
The middle column — how many people actually read each text — orders all the work. It is also the only one nobody fills in honestly.
- We fill it from your send and visit figures, never from an estimate.
- We start with your most-read text: the one sales sends out each morning.
- We look for one thing at a time, in three passes spread apart.
Out of forty pages, if fewer than ten have a real reader, leave the other thirty alone: they cost nothing while they are not lying.
Read next
Arabic, French, darija: which language to answer in, and to whom
The language of a message is not always the language of the expected answer. The rule is decided per channel, written on one page, and held to.Copywriting: writing is not styling, it is ordering
Almost nothing that decides a commercial text is a matter of style. It is the order, the vocabulary, and what you are not entitled to claim.Four pages that have to stay true
An owned page goes on answering after the answer changed. How to write the four that matter, and the hour a quarter that keeps them true.
Let us talk about your project
A free audit, no commitment: we look at your online presence and tell you what is holding it back.