Every blog in your category now sounds the same

Read four blogs in your category this morning and you will struggle to say which company wrote which post. That is new. Nobody has copied anybody, and it is not a failure of effort — it is what happens when every competitor drafts with the same tools, trained on the same corpus, pointed at the same keyword list. The sameness is a commercial problem long before it is an aesthetic one.

The sameness has a mechanism, not a mood

A language model's default output is the consensus of everything written about a subject. Ask for a post on invoice automation and you get the settled position, in the settled register, because that is what the training distribution holds. Your competitor asks the same question of the same average and receives a near-identical answer. Neither of you has plagiarised anything.

The prompts converge too. Most teams ask for roughly the same thing in roughly the same words, and a request shaped like "write a helpful blog post about X" is a request for the midpoint of the category. The tool did what it was asked to do. It was asked for the average.

This is not an argument against generating content. We generate every post on this blog, including this one. It is an argument that the default setting of the entire category is now a single voice, and that landing on it takes no effort whatsoever.

Google is not the one who punishes you for this

Sounding like everyone else is not a penalty, and it is worth being precise about that, because the fear in the room is usually the wrong fear. Google's problem has never been machine authorship, which is the case we make at length in Google doesn't penalize AI content. Prose quality is a weak ranking lever in either direction, and we would say the same to anyone convinced that a worse-written competitor is outranking them on style.

Search is not entirely indifferent, though. Google's guidance on creating helpful content asks publishers to assess whether a page provides substantial value when compared to other pages in search results. That is a comparative bar by construction. A post assembled from the same consensus as the nine results above it clears that bar by accident, or not at all.

The sharper cost is human, and it arrives much earlier than any ranking effect. A reader who has already met your paragraph on three other sites does not remember which company wrote it. A salesperson cannot send it after a call, because it argues nothing the buyer could not have found anywhere else. The archive fills up and none of it is attributable to you.

What a competitor cannot cheaply take

A model can imitate cadence. It cannot supply what you know and your competitor does not.

The parts of an archive that survive commoditisation are the positions: the thing you refuse to sell, the customer you turned away and the reason, the number you are willing to put in public, the approach you ran for a year before abandoning it. None of that is stylistic. They are claims with a cost attached, and the cost is precisely what makes them expensive to copy — a competitor who repeats your concession inherits the liability without the evidence that earned it.

That is why concessions are the least copyable sentences available to you. Naming what your product is bad at is cheap only when it is true, and it is credible only from the company that would lose something by being wrong. Everything else on the page can be regenerated by anyone with the same tools you used.

An engine produces the average unless something stops it

Our own category is the worst offender here, and pretending otherwise would be silly. Most automated blogs sound identical because the engine was handed a keyword and nothing else, and a keyword contains no opinions. What comes back is fluent, competent and interchangeable — which is exactly the outcome a buyer fears when they ask whether AI content "sounds like AI". They are not worried about grammar. They are worried about being indistinguishable.

The fix is not a better model. Somebody has to decide what the company actually thinks and write it down where the engine can reach it, which is the whole argument of the brief is the product. An engine will carry a position faithfully for a year without drifting or getting bored. It will not invent one on your behalf, and it should not be trusted to try.

The test is whether a competitor could have published it

Take the draft in front of you and ask whether a competitor could run it with only the logo changed. If the answer is yes, it is not your post yet. It is the category's post, and you are paying to host it.

Then put back the part only you can say. Usually it is a single paragraph: the number, the refusal, the failure nobody wrote up because it was embarrassing. That paragraph is the reason the page exists at all, and it is the one thing on it that cannot be generated from the same average everyone else is drawing on.