“Does AI-written content get penalised in search?” comes up constantly. The short answer: search engines rank on whether a page is helpful, not on who wrote it. What fails is not machine authorship but mass-produced thin content — which fails because it is thin. Blur that distinction and you end up at one of two wrong conclusions: refusing to use the tools at all, or pouring output onto the web with no gate in front of it.
The output is judged, not the author
A search engine cannot open a page and determine whether a person or a model wrote it. All it can assess is the result: does this actually answer the question, is there something here that is not available elsewhere, do people who click stay.
Those criteria predate AI entirely. Plenty of human-written articles never rank, and the reason is usually the same — they rearranged information that was already public, and the search engine already holds several copies of that answer. Machines did not create the situation. They made it possible to produce the same failure much faster and at much greater volume.
Why mass production actually fails
Generating one article per keyword automatically does fail reliably. But you need the cause precisely, or you cannot fix anything. There are three.
First, there is nothing in it. A rearrangement of public information contains nothing that exists only on that page. Second, the pages eat each other. Articles spun from near-identical keywords split one question between them, so none of them answers it properly. Third, nobody is accountable. If a wrong sentence is spotted and there is no one to verify and correct it, it is reasonable to assume none of the other sentences were verified either.
What actually matters
First-hand information. Things you measured or lived through: the figures either side of a change, the failure you actually met and its cause, the request you turned down and why. One paragraph of that separates a piece unmistakably from a rearrangement.
Accuracy. Do not publish sentences you cannot verify, and be strictest about numbers. AI produces contextually plausible figures with great fluency, and the sentence reads well. Keep unmeasured numbers out from the draft stage onwards.
Accountability. Someone has to be able to answer for the claims in a published piece. A byline looks like a formality, but it is really the answer to “if this sentence is wrong, who fixes it?”
A review gate: machine checks, human substance
We publish AI-assisted articles ourselves, so this is not somebody else’s problem to us. The structure we actually run has two layers.
Machine checks look only at form: are the categories set, are there enough headings, do the internal links actually resolve, are there draft leftovers. A person reviewing this is bored by the third article and starts missing things; a script never does.
Substantive review is the opposite. Is this sentence true, can we verify it, what here could only have been written by us. Whether the reviewer is a person or a model matters less than you would think. What matters is that work which fails review genuinely does not get published. If something has ever gone out despite failing, you do not have a gate — you have decoration.
So: using AI is not the problem; publishing without a gate is. Designing that gate is covered further in the AI archive, and the way we work is set out on our process page. If you would rather check the technical ground before stacking content on it, a diagnostic is part of our optimization program.