Every few months someone tells me AI content is finished on Google. Then I look at what actually got hit, and it's almost never "content written with AI". It's a pattern — and it's a pattern you can avoid.
So, does Google penalize AI content? No. Is AI content bad for SEO? Only when it's used in a way that would be bad for SEO whoever wrote it. Here's what that means in practice.
What Google actually says
Google's position has been consistent since it published its guidance on AI content: it cares about whether content helps people, not how it was produced. Using AI isn't against the rules.
What is against the rules is in its spam policies under scaled content abuse: producing lots of pages mainly to manipulate rankings rather than help readers. The policy is deliberately tool-neutral. It covers AI, but it equally covers a room full of people rewriting other sites by hand.
AI matters here only because it makes producing lots of low-value pages very cheap. That's the trap.
The pattern that gets hit
Look at the sites that get wiped out — automated news sites are the clearest example — and they tend to share most of these:
- Rewrites with nothing added. The article is someone else's content in different words. No angle, no context, no reason to read it instead of the original.
- No human in the loop. Nobody read it before it went live, which is how you get invented quotes, wrong numbers and stories about events that didn't happen.
- Fake people. Bylines for writers who don't exist, or "local correspondents" for towns nobody on the team has visited.
- Volume as the strategy. Hundreds of posts a day across dozens of near-identical sites, all chasing the same trending terms.
- No identity. No real person or organisation behind the site, nothing a reader could hold accountable.
Any one of those on its own probably won't sink you. Together they describe a site that exists for search engines rather than readers — exactly what the policy is written to catch.
What survives
The AI-assisted sites that keep their traffic look different:
- A narrow niche with an audience that has a real reason to read you.
- Sourced facts — taken from reading the original material, attributed and linked, not guessed.
- A genuine voice and angle, so your version says something the original didn't.
- A real person reviewing before publication, at least until the output has earned trust.
- Sensible volume. A handful of good pieces a day beats fifty forgettable ones.
- A named publisher — a real person or business readers can see.
Where this leaves automation
The tools aren't the risk. Unattended tools are.
When we built NewsMasher — I'm one of the people behind it — we made the cautious path the default rather than an option you have to go looking for. It reads the publisher's page before writing, writes in your own Writing Profile's voice, and flags anything it couldn't read for you to check. All five automation switches (writing, publishing, social drafts, social sending and ads) start off, so nothing goes live unless you've chosen to let it.
It won't guarantee rankings. Nothing honest can, and the NewsMasher site says so plainly. What it does is make it easy to run the kind of site that survives an update, rather than the kind that becomes someone's case study. If you're starting from scratch, here's how I'd create a news website with AI.
The quick self-check
Before you publish your next AI-assisted article, ask:
- Would a reader who already saw the original get anything from mine?
- Did a human read this before it went live?
- Is every fact in it traceable to a source I've linked?
- Would I be comfortable with my name on it?
If any answer is no, the problem isn't that AI wrote it.
— Wayne