Somebody finally measured it properly, which is more than most of the industry has managed.
A content strategist at Previsible produced the same article three ways. Written entirely by a machine. Written by a machine and tidied by a person. Written by a person with the machine only in the wings. Then she watched all three in Search Console for the better part of a year.
The pure machine pages, published in April 2025, did fine for a moment. By January 2026 they had all but vanished. The hybrid pages needed a full human rewrite before they moved at all. The human-written pages performed.
One person's experiment is not a law of nature and I would not present it as one. But it matches what everybody quietly observed and nobody had bothered to instrument.
Where the machine is genuinely excellent
Let me get this on the record before the rest, because the useful conclusion is not "do not use it."
Structuring something you have already thought about. Turning your notes into paragraphs. Cutting eight hundred words to three hundred. Producing twenty variations of a headline. Making your voice note into prose. All real, all worth having, all things I use.
What actually got graded
The interesting part is not that machine-written pages faded. It is why.
These systems produce the most statistically likely text on a subject. That is, definitionally, the average of what already exists. A page that is a competent summary of everything already published adds nothing to the internet, and increasingly it is measured that way.
Effort used to be a decent proxy for value. It is not any more, because effort became free. What is left is novelty: information that was not on the internet before you put it there.
Which is brutal for the standard agent blog, because the standard agent blog is almost entirely a restatement of publicly available facts.
The tell your readers hear first
Before any of that shows up in a ranking, a human being reads two paragraphs and decides nobody was home. They do not complain. They leave.
The measurement follows the behaviour. It does not cause it.
The playbook
Stop producing the encyclopaedia. "What is earnest money" is answered everywhere including in the assistant people asked before they got to you. You will not win it and winning it would not produce a client.
Publish what only you have. What is actually happening on the streets you work. Which listings sat and why. What a price reduction did. Your own numbers. That information does not exist anywhere until you publish it, which is the entire definition of novelty.
Write less, on purpose. One genuinely original piece a month beats twelve summaries, costs less, and does not require a second shift of fixing.
Use the machine on your own material. The failure mode is asking it to know things. It is very good at shaping what you already know. Keep it downstream of your judgement rather than upstream of it.
Any road up
The uncomfortable conclusion is that producing content got free and producing anything worth reading got harder, and those two facts arrived together.
You are standing in a market every day, watching things nobody else is writing down. That is the one asset in this that cannot be generated, and almost nobody is using it.