Covered, but no longer the same company
Meta description: When every AI-written page reads the same, verifiable firsthand detail is the remaining signal that survives at the answer layer.
Miklos Roth
The content review process in these companies works as designed. Each page is checked against the style guide: tone, terminology, approved claims, legal sign-off. Each page passes. And when I read the library as a whole — or when a model does — the pages answer as if they came from three different firms: different emphases, different depths, different implicit theories of what the company knows. The style guide held. Something beneath the style guide did not, and the company has no name for what was lost, because its instruments all operate at the level the style guide covers.
[AWAITING MIKLOS — scene: pages that pass review and still answer as if they came from three firms]
What does review currently check?
Review checks the page against the guide: is the tone right, are the terms approved, are the claims substantiated, is the logo used correctly. These are page-level properties, and they are genuinely enforced. What I notice is the shape of what review cannot check, given its instruments: whether this page, added to the other four hundred, makes the company more or less legible as the specific company it is. Consistency at the page level and consistency at the signal level are different properties, and the review process has eyes for one of them alone.
What inconsistency survives that check?
What I mean by a real signal: specific, verifiable, useful detail that comes from direct experience — an operational observation, an unusual customer problem, a firsthand account of a process limitation, a small but revealing distinction in how a service is delivered. These details can appear minor to a traditional content production team. From my perspective, they are often where genuine differentiation begins. Originality should not be fabricated through invented stories or invented client results; it should reflect the business itself.
The surviving inconsistency is substantive, not stylistic. Pages agree on tone and disagree on what matters. One cluster treats the company’s method as proprietary depth; another treats the same territory as generic industry practice. One writer implies the company serves enterprises; another writes for small teams. Each page, read alone, is defensible. Read as a set — which is how a model reads them — they describe a company with no stable shape. The inconsistency is not in any page. It is between them, which is why page-level review structurally cannot see it.
Why does scale make reviewers late?
Read more: Miklos Roth’s profile.
The false explanation is staffing: more reviewers, a second pass, a senior editor. I have watched review headcount grow and the drift continue, because the drift enters at the brief, not the draft. Briefs are written by different people, at different times, from different sources, with different implicit understandings of what the company is. By the time a reviewer sees a page, the drift is baked in — and the reviewer’s job is the guide, not the whole library. Scale does not break the review. It multiplies the number of independent drifts the review was never designed to reconcile.
What pattern marks signal-level break?
The pattern I look for is in the library’s aggregate behavior, not in any page. Ask a model what the company knows — not what it sells, what it knows — and read the answer. In companies with intact signal, the answer is specific and stable across phrasings. In companies with signal-level break, the answer is generic, hedged, or shifts with the phrasing of the question. The library contains everything. The company, as the model assembles it, stands for nothing in particular.
Drift enters through generic production: producing another acceptable-looking page is easier than developing an informed point of view. Over more than twenty years in SEO, I have watched tactics become templates. Once enough businesses imitate a successful format, the format itself stops being a meaningful competitive advantage. The defense is editorial: keep the technical foundations and distribution work that make pages discoverable, but do not confuse discoverability with a reason to be chosen.
[AWAITING MIKLOS — scene: second scene]
What question is a page for, if not coverage?
Coverage thinking asks whether the library contains a page for each topic. The question I find missing is about the page’s job in the aggregate — what this page, added to the set, makes the company more clearly the master of. Libraries built on coverage grow toward encyclopedia shape: broad, neutral, interchangeable. The question that would change the trajectory is about the set, not the topic list, and in these companies it has no natural moment to be asked.
The question that should sit above the publishing calendar: of everything published this quarter, what could only have come from this business — and what would a competitor’s identical version look like?
PR excerpt: [PR — awaiting Miklos material; cut 80–120 words from his paragraphs once available.]
