There's a guy in one of my SEO Facebook groups who I've started thinking of as "Panic Guy Pete." Not his real name, obviously, but every time Google so much as breathes near an algorithm update, Pete is the first one posting in all caps. Last March, after the core update rolled out, Pete announced he'd deleted 400 blog posts overnight because he'd "heard AI content was getting nuked." A week later, he was writing every single article himself again, by hand, at 1 a.m., convinced that typing slower was somehow the ranking signal Google had been waiting for.
I get the instinct. I really do. When you've built a content operation and you see traffic charts cratering across your industry, the scariest explanation is also the simplest one: "Google can tell it's AI, and it's punishing me for it." That story is clean. It has a villain, a victim, and a fix. Delete the AI content, write everything by hand again, problem solved.
Except that's not what's actually happening, and after running content pipelines that mix AI drafting with human editing for a while now, I can tell you the panic and the fix are both aimed at the wrong target.
Google Isn't Detecting Your AI. It's Detecting Your Effort (Or the Lack of It)
Google's own position on this has been remarkably consistent since it first addressed generative AI content back in 2023: the focus is on the quality of the content, not how it was produced. That line hasn't changed, and I've watched enough sites rise and fall to believe them. Google's guidance for site owners specifically says generative AI is genuinely useful for researching a topic and structuring original content, but using it to crank out page after page without adding real value for readers can trip their "scaled content abuse" policy.
Read that again, because it's the whole ballgame: the violation is the scale without value, not the AI. A single well-researched, fact-checked, AI-assisted article sitting next to your genuinely helpful human-written ones isn't the problem. Five thousand near-identical "best plumber in [city name]" pages generated overnight and shipped without a human ever reading them? That's the problem, and it was the problem back when people did it manually with spun content, too. AI just made the bad version faster to produce.
Google's spam policies even spell out what this looks like in practice, and the pattern is depressingly familiar to anyone who's been in this industry a while: template articles with a few swapped nouns, scraped content run through a rewriter and posted unedited, programmatic filler pages built to catch long-tail traffic rather than answer a question a person actually has. None of that requires AI to be a violation. AI just removed the friction that used to slow it down.
What Actually Gets Sites Hit
I've sat with clients after traffic drops enough times to notice the pattern, and it rarely comes down to "we used AI." It comes down to a handful of things that show up again and again:
Publishing faster than a human could ever review. If your content calendar has you shipping fifty articles a day across a dozen sites, nobody on your team read those before they went live. Google's raters are specifically trained to look for main content created with little to no effort, little to no originality, and little to no added value — and volume without oversight is the tell.
No one's expertise is actually on the page. This is the E-E-A-T problem, and it's the part I think gets misunderstood the most. AI can synthesize what's already been written about a topic. It cannot have personally used the product, treated the patient, or run the experiment. If your health article reads like a summary of other health articles, with nobody's name, credentials, or firsthand experience attached to it, that's a trust gap AI alone can't close, and Google's raters are trained to notice the absence.
Zero editorial layer. The sites that survived the recent core updates almost universally used AI for the draft and structure, then had an actual person fact-check, add original data or quotes, and cut the generic filler. The ones that got flattened skipped that step entirely.
Fabricated authority. Bylines with fake credentials, AI-generated "expert" quotes, invented statistics that sound plausible but don't trace back to a real source. This one isn't a gray area. It's a straightforward trust violation whether a human or a model typed it.
What I Actually Do With Every AI Draft Now
I'm not going to pretend I write every word by hand anymore. I don't, and neither does most of the industry, whatever Panic Guy Pete believes at 1 a.m. What changed is what happens between the draft and the "publish" button.
Every AI draft gets a pass where I check the facts against a primary source, not just against whatever the model was confident about. Numbers, quotes, dates, anything that could be wrong in a way that matters gets verified before it goes anywhere near a live URL. I add something the model genuinely couldn't: a detail from having actually done the thing, talked to the person, or tested the product. I cut the sentences that exist only to hit a word count, because those are exactly the sentences that make an article read like it was "obviously written by ChatGPT," and honestly, they usually were right about that part. And I keep bylines honest. If a real person with real experience reviewed the piece, their name goes on it. If nobody did, that's a signal to slow down before publishing, not a technicality to paper over.
None of this is about outrunning an AI detector. Google has said plainly it isn't running content through one and grading you on the result. It's about whether the page in front of a reader actually helps them, which turns out to be a much harder standard to fake than people want it to be.
The Uncomfortable Part
Here's the thing Pete and I actually agree on, even if we got there from opposite directions: publishing less and checking more is the right move. We just disagree on why. He thinks it's because Google can smell the robot. I think it's because volume without a human paying attention was always a bad bet, long before language models existed to make it cheaper.
I don't know if that distinction matters to anyone deleting 400 posts at midnight. But it's the only version of this I've found that actually holds up the next time an update rolls through and half my feed starts posting in all caps again.
You May Also Like: AI vs. Human Content: Does AI Content Cause Google Ranking Loss?
____________________________________
Ron
Digital marketing expert and professional writer
Stoycope.com