Every few weeks a founder asks me some version of “is it safe to publish content written with AI?” Usually they’ve seen a screenshot of a traffic chart falling off a cliff and a caption blaming Google’s latest spam update.
The fear is understandable. The diagnosis is wrong.
Google is not banning AI content. Google is banning scaled content abuse, and its own spam policy says so in plain language. The policy doesn’t care whether a human or a model typed the words. It cares whether you generated hundreds of pages that exist to rank rather than to help anyone.
That distinction matters, because it tells you exactly what to do. This post covers what Google is actually cracking down on, the 4 signals that get a site flagged, who’s getting wiped out, and how we publish AI-assisted content at Hill Tribe Solutions without losing sleep over it.
What Google actually says about AI content
Google’s spam policies list the things that get a site demoted or removed. Cloaking. Doorway pages. Expired domain abuse. Hidden text. Link spam. And, added a couple of years ago, scaled content abuse.
The definition is worth reading closely. Scaled content abuse is producing many pages with little or no value for users, regardless of how they were made. Google explicitly names AI as one way people do it, alongside scraping, spinning, and cheap outsourcing. But the violation is the scale and the emptiness, not the tool.
Google’s separate guidance on AI-generated content says the same thing from the other direction. Content made with AI is fine if it’s helpful, original, and made for people. Content made to manipulate rankings is not, however it was produced.
So the question was never “did AI write this.” The question is “does this page have a reason to exist beyond ranking.”
The 4 signals that get a site flagged
Google can’t read intent. What it can do is measure patterns. From what we’ve seen across client sites and our own, 4 signals do most of the work.
1. Content sameness
High semantic similarity across pages. The classic version is a service page templated 200 times with only the city name swapped. Every page says the same thing in the same order with the same sentence structure. A model can spot that in milliseconds, and a Google model has been trained to.
This is the one that catches the most well-meaning people, because templating used to be the only way to produce location or category pages at volume. It’s not anymore. If you have an AI that can write each page from scratch with real local detail, there’s no excuse for the template.
2. Behavior signals
Someone lands on the page, reads for 4 seconds, and clicks back to the search results. Do that at scale and Google learns the page didn’t answer the query. Engagement is the single biggest tell. A site people stay on is a site Google can trust. A site people bounce off is a site Google tests, then demotes.
3. Site patterns
Hundreds of pages published overnight. No named author. A generic About page that could belong to any company. No phone number, no real address, no photo of a person. These are the fingerprints of a site built to be a site rather than a business.
4. No experience
Nothing on the page that only someone who’s done the work could have written. No real price range. No specific mistake you’ve made. No named client pattern, even anonymized. Google’s quality raters are literally instructed to look for first-hand experience, and the algorithms are tuned to what those raters reward.
Two ways you get hit
There’s a quiet way and a loud way.
The quiet way is algorithmic demotion. No warning, no message in Search Console. Traffic drops 50 to 80% over a few days and stays there. This is what most affected sites experience, and it’s what the scary charts show.
The loud way is a manual action. A real person at Google reviews the site and flags it. This is rarer, but it’s also more final. I’ve seen a site get manually removed from the index entirely. Not demoted. Gone, with no realistic path back.
That second one is worth sitting with, because it means you have 2 audiences to pass. The algorithm, which is math crawling your site looking for patterns. And a human, who might open your About page and decide in 10 seconds whether this is a real business. You have to pass both.
Who’s actually getting wiped out
The casualties are consistent across every update we’ve watched.
Niche information sites with hundreds of thin AI pages and no author. Content farms pushing volume every day. Affiliate review sites that never touched the product. And the one closest to home for a lot of businesses: location pages that are the same page with the city swapped.
Notice what’s not on that list. Company blogs publishing a well-researched piece every week or 2. Service pages that describe what you actually do, with real prices and real examples. Documentation. Case studies. Product pages with original detail.
Those sites are using AI too. Most of them heavily. They’re fine, because they’re not doing anything the policy describes.
Why I’m not worried about this, and why you shouldn’t be
Every competitor that gets demoted for churning slop is a page-one position that just opened up. Google is doing the field-clearing work for anyone who publishes with care. The lazy operators go first. That’s a moat for the rest of us, not a wall.
We’ve been building with AI for over a year now, and the lesson from a year of building with AI applies directly here. The tool doesn’t make the decision. It makes the decision cheaper to execute. If your decision is “generate 500 pages tonight,” AI helps you do that faster and Google punishes you faster. If your decision is “write one page that actually answers the question,” AI helps you do that better.
How we publish AI-assisted content that holds
Here’s the operating playbook we use for our own site and for client sites. None of it is clever. All of it is about removing footprints.
Every page is semantically unique. No templates. If we build 30 location or category pages, each one gets written from its own brief with its own local references, its own examples, and its own structure. The model makes that cheap. We just refuse to skip it.
We publish at a human pace. A site that has posted twice a month for a year doesn’t suddenly post 10 times a day. If you’re building out a large site, build it, stage it, review it, and launch it as a finished thing. Don’t let Google watch you dump 300 pages in a night.
Real About and Contact pages. A named founder with a photo. A real address. A phone number that gets answered. This sounds obvious. It’s the thing most flagged sites are missing.
Original detail on every page. A real price range. A real client pattern, anonymized. A specific thing that went wrong on a build and what we changed. This is the “experience” signal, and it’s the part AI can’t fake for you. You have to supply it.
Engagement is the goal, not a byproduct. Every page has one job: answer the question the searcher had well enough that they don’t go back to Google. If a page can’t do that, it doesn’t go up.
Standalone sites, not networks. If you run more than one site, they don’t link to each other in a pattern. Each one stands on its own as its own business. Cross-linked networks are the oldest footprint there is.
What this means if you’re a founder using AI to write
Use it. Seriously. The risk was never the tool.
But treat every page like it has to survive a human opening it and asking “is this a real business that knows what it’s talking about?” If the answer is yes, the algorithm will get there too. If the answer is no, no amount of prompt engineering will save it.
A lot of teams get this wrong in the same way. They learn the tool before they learn the judgment. That’s why our AI training for teams spends more time on what to publish and what to kill than on how to prompt. The prompting is the easy part.
The short version
Slop dies. Useful survives. That’s the whole update.
If you’re publishing with AI and you’re nervous, the fix isn’t to stop. It’s to make every page unique, add the experience only you have, and make sure a stranger could tell within 10 seconds that a real business is behind it. Do that and the crackdown works in your favor, because it’s clearing out everyone who didn’t.
And if you want a second pair of eyes on a site before you scale it, we run audits on exactly this. Better to find the footprints yourself than have Google find them for you.