How I keep fifteen brands from sounding the same
# How I keep fifteen brands from sounding the same
The first sentence of every article I had published in a year, dumped into one file and sorted by brand, is the most useful thing I have ever done for my writing.
Eleven of the fifteen columns opened the same way. A number, then something that broke.
That was not a model problem. I had taught myself one opening move, put it in a shared instruction, and handed it to every brand I own.
A review process that cannot see the problem
For a year I checked every piece before it shipped. Is the claim true. Does the link resolve. Is there anything here a competitor could not have written. Does it stop when it runs out of things to say.
All of those questions are about one file.
Sameness does not live in a file. It lives in the set. A piece can pass every check I have and still be the twelfth near identical sibling of eleven others, because nothing in a per piece review knows what the other eleven look like. I had built a quality process structurally incapable of finding the fault I cared about most, and it took a year to notice, because each individual piece really was fine.
Readers will not catch it either. Nobody reads two of my brands; the audiences barely touch. The only person positioned to notice is me, and I was reading one file at a time.
Tone words are not a control
My first fix was to give each brand a tone. Dry and technical. Warm but precise. Plain spoken and skeptical.
It changed almost nothing.
Adjectives collapse under pressure. Ask for dry and you get the same essay with fewer jokes. Ask for warm and you get the same essay with a sentence of reassurance bolted to the top of each section. The adjective describes the surface, and the thing making everything sound alike sits well below the surface. Two of my brands had opposing tone words in their configs and produced paragraphs I could have swapped without either page looking odd.
Asking for a rewrite is worse. Request more variety on a finished piece and you get synonyms, inverted clauses, and a paragraph moved from second position to fourth. The piece is now harder to read and exactly as similar to its siblings as it was. Difference has to be decided in the brief, before anything exists to edit.
What a voice document actually holds
Each brand now has a page. Most of it is evidence rather than description.
It opens with roughly four hundred words I typed myself, in that brand's voice, about that brand's subject. An example carries sentence length, where the hedging sits, whether the writer says I, how long the first sentence runs before it resolves. No adjective transmits any of that.
Then the line that turned out to be the real lever: what the reader already knows. On the proxy brand, the reader knows what an ip is, and explaining it is an insult. On one of the link sites, the reader has never heard of a residential proxy and will leave if the second sentence assumes otherwise. Identical subject, two pieces with nothing structural in common, and the only input that differed was one sentence about the reader.
Then who holds the pen. An operator who runs the thing daily. A reviewer who paid for it and used it a month. A researcher who has read the documentation and owns none of the hardware. Three different relationships to certainty, and certainty is most of what a voice is made of.
Then a narrow vocabulary section. Words this brand uses where the industry uses something else, and words it will not use. One of mine says line and never connection. Another says phone and never device. It sounds like nitpicking. It is the cheapest lever I have.
Structure carries more voice than vocabulary
The change that did the heavy lifting was structural. Every brand gets a shape, and that shape belongs to it alone.
One of mine has no section headings and runs as continuous prose, which also makes it shorter. One is question led, where every heading is a question a customer actually sent me. One is a log, chronological, and it refuses to state the lesson before the story that produced it. One opens by stating the strongest argument against its own position.
None of those shapes is better than the others. That is the entire point of them.
I spent years believing voice lived in word choice, which is why I kept building vocabulary lists that did nothing. What a reader experiences as a publication having a character is mostly the order in which facts arrive and how long the payoff is withheld. Change the order and two pieces with identical vocabulary read as separate magazines. Keep the order and no amount of synonym work will pull them apart.
The list of things a brand may never say
Every brand carries a short list of claims it is not allowed to make. I wrote these to stay out of trouble, and they turned into the strongest voice tool I have by accident.
The proxy brand never says anonymous. It says harder to correlate, because that is what is true, and anonymous is the kind of word that gets somebody arrested. The phone rack never implies a device belongs to the customer permanently, because it is rented and it returns to the pool. The review sites never use the word tested unless something was genuinely tested, and where it was not, the piece says so and describes what the documentation claims instead.
What a publication refuses to claim identifies it faster than anything it does claim, because every brand in every niche claims the same handful of good things. The refusals are where the personality is.
A brand banned from saying anonymous has to explain the mechanism instead, which converts a marketing sentence into a technical paragraph every single time, with nobody making a decision.
The only detector that works
Everything above is prevention. Drift still happens, and catching it needs a test.
The test is manual and slightly embarrassing. Pick two brands. Read the newest piece from each, one straight after the other, in a single sitting.
Your eye calibrates to whatever it has been reading. Six pieces from one brand in a row and by the third you have stopped registering the house style, the way you stop hearing an air conditioner. Alternate between two brands and the shared skeleton stands up in front of you.
I do two pairs a week, about twenty minutes total. It has found more than every automated check I have written combined.
The question I hold while reading is whether I could swap the opening two paragraphs without either page looking strange. If I could, both are broken, and so is the brief behind them.
Four checks a script can run
Cheap, mechanical, no model involved.
Opening sentence inventory, one file per brand, read as a column. This is the check that found the eleven brands opening on a number and a failure.
Heading inventory, same idea. I found a phrase about the smallest version you could run tonight, with minor variations, sitting on three different sites.
Cross brand phrase counts. Any phrase of four or more words appearing on more than two brands is house style, and house style is the thing I am trying to destroy. Each one gets assigned to a single brand or removed from all of them.
Sentence length distribution per brand. Genuinely different voices should produce genuinely different distributions. Mine were stacked on top of each other. When two brands converge on that number now, I know something upstream got copied.
The scorer I had to delete
I built a checker that read a finished piece, compared it against its brand's voice document, and returned a score out of a hundred. Anything under eighty went back.
Two weeks later everything was passing and the writing had got noticeably worse.
The scorer could see the vocabulary list, so the writing learned to satisfy the vocabulary list. The brand whose document says line rather than connection started producing pieces that used the word nine times in eleven hundred words, in places where a person would have said it once and switched to a pronoun. The score climbed while the prose went stiff. I had built a keyword density check and labelled it a voice check.
The document survived. The scorer did not. A description of a voice works as a reference and stops describing anything the moment it becomes a target.
The sameness I am keeping
One person writes all fifteen of these, so some convergence is permanent.
My sentence rhythm shows up everywhere. So does leading with a number, and refusing any claim I cannot back, and putting the thing I got wrong about three quarters of the way down. I spent a month stripping those habits out of one brand. The writing went flat. Whatever made it distinctive was the same machinery making it repetitive.
So the target changed. I want fifteen publications a reader could tell apart with both open side by side. If somebody reads two of them and concludes same author, different magazines, that is accurate and I am content with it. The reaction worth preventing is the other one, where it reads as the same article with the nouns replaced.
The reason it matters goes past duplication risk. When two brands produce the same piece, what that actually tells me is that I wrote both of them without a specific reader in mind, and the structure defaulted because nothing was pushing against it. So when the back to back read turns up two pieces sitting on top of each other, I stop editing sentences and go find out who each one was for. Half the time, for one of them, there was never an answer.
Get new guides and videos first — join the Telegram channel.