Eight articles that answer the question beat sixty that circle it
Most AI blog programs fail the same way. They produce volume, the volume competes with itself, and the site ends up thinner than it started. We use the models to draft faster, not to publish more.
In short: We build each article around a question your customers really ask, put your own measurements and experience inside it, and edit it until it reads like a person who knows the work wrote it. Fewer pages, each one meant to be the last page somebody needs to open.
We start from questions, not a keyword list
A keyword list tells you which words people type. It does not tell you what they wanted, what they had already tried, or what would end the search for them. So the brief comes from somewhere else. We read your support inbox, sit in on a sales call or two, pull the queries in Search Console that never got a page of their own, and look for the question a prospect asks twice because the first answer was not good enough. That question becomes the brief, and the brief is what the draft gets generated against. The model writes toward an answer rather than toward a word count.
This changes what the finished page looks like. It usually opens with the answer instead of a paragraph of throat-clearing. It covers the awkward part your competitors skipped, because the awkward part is why the person is still searching. And it stops when the question is answered, which is sometimes at eight hundred words and sometimes at three thousand.
Where briefs come from
Support tickets, sales objections, your Search Console queries, and the forums where your buyers argue with each other about the thing you sell.
What a brief contains
The question, who is asking it, what they already tried, which evidence settles it, and an explicit list of claims the page is not allowed to make.
Where the model helps
First drafts. Restructuring a section that arrived in the wrong order. Six alternate openings in a minute. Compressing three hundred words that were saying sixty.
Where a person takes over
Every factual claim. The numbers. Anything an expert reader would push back on. The judgment about what to cut, which is most of the work.
Why we argue against sixty posts a month
Sixty thin pages a month are not sixty chances to rank. They are sixty pages competing with each other over the same small set of questions, splitting whatever authority the site has earned, and teaching a crawler that most of your URLs are not worth coming back to. The shape of it is familiar. Traffic rises for a quarter. Then it flattens. Then the article that used to do well is outranked by a weaker one you published yourself, and nobody can tell which page is supposed to be the real answer.
We would rather produce eight pages a quarter that are the best available answer to eight questions. Be clear about the trade. It is slower, the first couple of months look quiet, and there is no chart to show at the end of month one. It also means turning down topics that somebody else has already answered well, unless you have something to add that they do not have. If a volume target is the point, we are the wrong agency, and there are plenty of others who will hit it.
Your numbers are the part no model can supply
Ask a model about typical lead times in your industry and it will write a confident, fluent paragraph. It has no idea what yours are. That gap is the entire opportunity. The material worth reading lives inside your business and nowhere else: the failure rate you measured across a couple of hundred installs, the material everyone recommends that you stopped using and why, the four things that go wrong on site and what each one costs to fix. None of that can be copied by a competitor or invented by a model.
It is also the part that gets quoted. When an assistant answers a question by drawing on sources, specific attributable claims travel and generic advice does not. That is the practical connection between writing well and being cited, and it is what our generative engine optimization work and content structuring are built around. So we ask for an hour with the person who actually does the work, and we ask for access to the figures you are comfortable publishing. Without those, we can only write what is already out there.
The tells that give machine-made writing away
You feel it before you can name it. Then you start noticing the sentences are all roughly the same length, eighteen to twenty-two words, one after another, with nothing short to land a point on. The opening paragraph defines the topic instead of saying anything about it. Adjectives stand in for measurements, because the model has no access to the measurement: significantly faster, considerably more efficient, dramatically improved. Every section is weighted the same as every other section, so nothing is emphasized and nothing is skipped. And the final paragraph restates the first one, because that is the shape a conclusion takes when there is nothing left to add.
None of this gets fixed by running the draft through a humanizer. Cutting a sentence short because the point has already landed, swapping an adjective for the number it was covering for, deleting a conclusion that adds nothing, giving one section three times the space because it is three times as important — that is editing. It is exactly the same work that makes any piece of writing good, done by somebody who can tell which parts are worth keeping. The reason our drafts do not read as machine-made is not a detection trick. It is that they get edited by a human who knows the subject.
What we won’t do
Publish a statistic we cannot source
If a figure cannot be traced to your own records or to a named public source we can link to, it comes out of the draft. That holds even when the number would have made the argument land harder. Models produce plausible statistics on request, which is precisely the problem.
Write fake reviews, testimonials or case studies
We do not write praise in a customer’s voice, and we do not write up projects that did not happen. If you want case studies, we will interview real clients and write what they actually say. If none will go on record, the page says nothing instead.
Take a volume contract we think will hurt the site
If the brief is thirty posts a month on topics you have no particular authority over, we will say no rather than take the money. We would be selling you the dilution problem we are supposed to be fixing.
What question does your site still not answer properly?
Tell us the one your sales team keeps answering by email. We will tell you whether it is worth a page and what evidence it would need.
Will readers be able to tell the articles were drafted with AI?
If they can, we did the job badly. A model writes the first draft; a person restructures it, adds your evidence, cuts the padding and rewrites whatever carries weight. What ships reads like it came from someone who knows your business, because the parts that matter did.
How many articles do you publish a month?
Usually two or three, sometimes fewer. We plan in quarters rather than months, and we would rather spend a week on the page that answers your most valuable question than fill a calendar. If you need thirty posts a month, we are the wrong agency for that work.
What do you need from us to write well?
About an hour with someone who does the work – an engineer, a senior technician, a founder – plus access to the numbers you are comfortable publishing. Search Console access helps. Without any of that, we can only write what is already published elsewhere.
Can you improve articles we already have?
Often that is the better spend. A page sitting in positions eight to fifteen with a real question behind it usually needs evidence and structure rather than replacement. We audit what you already have before proposing anything new.
