What AI Search Actually Rewards
Google does not care whether a human or a machine wrote your content. Its published guidance says so directly — it rewards quality content regardless of how it was produced. What it cares about is whether your page adds something a reader can't already get somewhere else. That's the whole test, and it's where most AI-written content fails.
I run a $2M junk removal business from 1,200 miles away in under an hour a day. What a job costs me and what I charge is not something any model can generate.
The test isn't who wrote it
Google published guidance on AI-generated content and it's been consistent since: the focus is on the quality of the content, not the method used to produce it. Using AI isn't against the rules. Producing content whose main purpose is manipulating rankings is — and that's been true since long before AI existed.
So the question was never "will I get caught using AI."
The question is whether the page you published says anything the internet didn't already have. And that's exactly where AI-written content tends to die — not because a robot wrote it, but because a robot can only give you back what's already out there. That is the definition of adding nothing.
Information gain: what it is and what it isn't
Google holds a patent describing this idea. It's called "Contextual estimation of link information gain," filed in 2018 and granted in June 2022. The patent describes scoring a document by the additional information it contains beyond what the reader has already seen in documents they viewed before.
That's a real, citable document. You can pull it up on Google Patents yourself.
Two honest caveats, because this is where the SEO world overreaches.
First, a patent is a patent. Companies patent enormous numbers of things they never ship. Nobody outside Google can tell you whether this specific system is running in search today, and anybody who says otherwise is filling in a blank.
Second, read what it actually describes. It ranks documents by how much they add relative to what a reader has already seen — not a simple "originality score" stamped on your homepage. It says nothing about AI Overviews, and nothing about which pages an assistant picks to cite.
A patent tells you which direction a company was thinking, not what it runs. The practical takeaway survives all that caveating intact: if your page is the same five points as every other page, there's no reason to surface it. Add something that isn't online yet.
The 0-to-127 score people keep getting wrong
You've probably seen someone claim Google scores your content 0 to 127. That number is real. It's in the 2024 documentation leak, which Google confirmed was authentic. But almost everyone repeating it gets it wrong in two ways.
First, they call it information gain. It isn't. Information gain is the patent above, and it wasn't in the leak at all. The 0-to-127 comes from a separate field called OriginalContentScore. Two different things, glued together by people repeating each other.
Second, 0 to 127 isn't the whole picture. The field's own description says it applies to pages with little content on them. Analyses of the leak put the broader range at 0 to 512.
And here's the part nobody says out loud: no one outside Google has ever seen a real number from that field. Not one. So when somebody tells you most content scores a four and yours needs to be in the 90s, they made that up. There is no public data. It doesn't exist.
That's the whole problem with how this stuff circulates. Real document, real field name, and then somebody invents the interesting part because the real part isn't interesting enough on its own.
The actual takeaway doesn't need a number. Google has a field for measuring whether content is original. Being original is the play. You don't need a score to know whether you're saying something nobody else can say.
The watermark panic
On August 2, 2026, transparency rules under Article 50 of the EU AI Act took effect. Providers of generative AI have to mark their outputs in a machine-readable way so they're detectable as artificially generated.
Anthropic confirmed shortly after that new Claude models embed an invisible watermark into generated text, applied globally rather than only in Europe. Google has used its SynthID system for some text. OpenAI has applied provenance signals to images and audio and has said it intends to extend that to text. So this is uneven across companies right now, not a single switch that flipped.
Here's what nobody's explaining. The mark doesn't prove AI wrote something. It proves AI touched it. Anthropic has acknowledged the mark can appear even when the model only proofread or translated text a person wrote. And plenty of AI writing carries no mark at all.
There's also no public detector. You can't check a competitor's page, and there's no cross-lab standard that would let one company's tool read another's mark.
None of it changes anything for your business, because Google's position hasn't moved. It doesn't care who typed it. It cares whether the thing exists anywhere else.
Everyone teaching this has nothing to teach with
I've been pulling transcripts of videos about information gain for a few weeks now. Nearly all of them are the same concept, the same handful of examples, close to the same script. All telling you to stop saying what everyone else is saying.
That's the whole problem in one sentence.
Explaining information gain is easy. Having any is hard. Most people teaching it don't run a business, don't have numbers, don't have a job that went wrong last Tuesday. So they teach the concept, because the concept is the only thing they own.
I run a junk removal company in Connecticut. What a cleanout costs me, what I charge for it, what I got wrong — that isn't a content strategy. It's the only thing I've got that nobody else can post.
You have the same thing and you're sitting on it. It isn't your opinions. It's your invoices.
Frequently Asked Questions
Does Google penalize AI-written content?
No. Google's published guidance focuses on content quality rather than how it was produced. What it acts against is content produced mainly to manipulate rankings, which was already against the rules before AI.
Is information gain a confirmed Google ranking factor?
No. Google holds a granted patent describing the approach — "Contextual estimation of link information gain." A patent shows what a company was thinking about, not what it runs in search today. Anyone stating it as a live ranking factor is going beyond the evidence.
What is the OriginalContentScore?
A field found in the 2024 Google documentation leak, whose description indicates it applies to pages with little content. The commonly cited 0-to-127 range is one part of it. No real scores from it have ever been published.
What can I publish that AI can't generate?
Your own numbers. What a job actually costs you, what you charge, how long it takes, what went wrong and what it cost to fix. None of that exists online until you put it there.
Adimize does AI-Search SEO for home service businesses. See what we do →