AI Content Detection: Does Google Actually Care? (After 10 Years in SEO, Here Is My Take)

We get asked this almost every week. A client forwards a Slack message, a founder emails us at 11pm, an agency partner pings us between calls. The wording changes but the fear is identical: "We used ChatGPT to help with a few blog posts. Are we about to get nuked by Google?"

Short answer, and we'll happily put our name on it: no, Google does not penalise content for being made with AI. What Google penalises is content that's rubbish. Those are two very different things, and the entire panic around AI content detection comes from confusing them.

We've run SEO campaigns through every major Google update of the last decade. We've watched what actually moves rankings on real sites, not what gets shouted about on LinkedIn. So let's get into what's real, what's noise, and what we'd genuinely do if it were your money on the line.

Does Google penalise AI-generated content? The honest answer

It does not. And this isn't us reading tea leaves. It's Google's stated, written position.

Back in February 2023, Google published its guidance about AI-generated content, and the line everyone should tattoo on their forearm is this: their "focus on the quality of content, rather than how content is produced, is a useful guide." Google goes on to say that appropriate use of AI or automation is not against its guidelines, and that helpful, original content satisfying E-E-A-T can do well in Search "however it's produced."

Read that again. However it's produced. Google is not running a detector on your draft and docking points if a model scores it above some secret threshold. It never has. That's not how the ranking systems work, and it's not what Google has ever claimed to do.

So where does the fear come from? Mostly from people misreading the one rule that does exist, which we'll cover next.

What Google actually penalises (and it isn't the robot)

There is a rule, and it's worth understanding properly because it's the thing people trip over.

Google's spam policies define scaled content abuse as "when many pages are generated for the primary purpose of manipulating search rankings and not helping users." The policy then names the obvious culprit directly: "using generative AI tools or other similar tools to generate many pages without adding value for users."

Notice what's doing the work in that sentence. It isn't the word "AI." It's "many pages," "without adding value," and "primary purpose of manipulating search rankings." The AI bit is incidental. You could do the exact same thing with a sweatshop of underpaid writers and Google would treat it identically. In fact it has, for years, long before anyone could spell GPT.

Here's what we've actually seen drag real sites down in our campaigns:

• Scaled junk. A client comes to us after spinning up 400 near-identical location pages overnight. No human ever read them. Traffic spikes, then craters. That's not an AI penalty, it's a thin-content penalty wearing an AI costume.
• Swapped-variable duplication. "Best plumber in [CITY]" repeated 60 times with only the town name changed. Google's been squashing this since the doorway-page era. This is the exact trap we keep flagging in our work on programmatic SEO, where AI just makes it cheaper to produce the same mistake faster.
• Content with no expertise behind it. A 350-word page about a medical symptom, no clinician involved, no sources, no first-hand experience. For YMYL topics especially, that's a quality failure that has nothing to do with the tool used to type it.
• Pages that exist to rank, not to help. Keyword-stuffed, answers the question worse than what already ranks. Always been penalised. Always will be.

None of those are "AI problems." They're effort problems. AI didn't cause them, it just removed the friction that used to stop lazy operators from making them at scale.

What Google's quality raters were told about AI in 2025

This is where the nuance lives, and it's the bit that gets mangled in panicky blog posts.

In January 2025, Google updated its Search Quality Rater Guidelines to formally address generative AI. As Search Engine Land reported, raters are now told to assign the Lowest rating when a page's main content is "copied, paraphrased, embedded, auto or AI generated... with little to no effort, little to no originality, and little to no added value."

That phrasing matters enormously. The penalty trigger is the chain of "little to no effort, originality, added value." AI is listed in the same breath as copying and paraphrasing, because the sin is the same: pushing out content that adds nothing. The guidelines also, for the first time, formally define generative AI and explicitly call it "a helpful tool for content creation."

A few important caveats here. Quality raters do not directly change your rankings. Their scores train and calibrate Google's systems, they aren't a live penalty button. And the test they apply is originality and value, not "did a machine touch this." Our take: this update is Google telling raters to spot lazy regurgitation, not to play spot-the-robot.

Why AI content detectors are mostly snake oil

We've tested the lot. GPTZero, Originality.ai, Copyleaks, the free ones, the paid ones. Here's the uncomfortable bit nobody selling a detector wants printed.

They're guessing. We've fed them blog posts we wrote by hand, every word, and watched detectors flag whole sections as "likely AI." We've fed them obviously machine-written sludge and watched it sail through as "human." If a tool can't reliably tell the difference in a controlled test, it certainly can't be the basis of a Google ranking system.

And this isn't just our anecdote. A Stanford study by Weixin Liang and colleagues, published through Stanford HAI and in the peer-reviewed journal Patterns, found that GPT detectors misclassified more than half of TOEFL essays written by non-native English speakers (around 61%) as "AI-generated," while being near-perfect on essays by US-born students. The detectors lean on signals like "perplexity," which basically punishes simpler, more constrained language. The full paper is on arXiv if you want the methodology.

Sit with that. These tools systematically flag people who write in plainer English. If your business ran hiring or editorial decisions off that, you'd be filtering out humans and calling them robots. Google, to its credit, has never built its ranking on anything this shaky.

The question detectors try to answer, "was this written by AI?", is the wrong question. The question that decides your rankings is "does this help the person who searched?" Those are not the same, and only one of them pays your bills.

So can AI content actually rank? Here's the data

This is the part people really want to know, so let's use numbers instead of vibes.

Search Engine Land published a 16-month experiment that's about as close to a clean test as you'll find. The setup: 2,000 AI-generated articles across 20 brand-new domains, left to fend for themselves.

The early results looked tempting. Around 71% of pages were indexed within 36 days, and 80% of the sites ranked for at least 100 keywords in month one. If you'd taken a screenshot at week four you'd have declared AI content a free lunch.

Then reality arrived. By month three, only 3% of pages were still in the top 100, down from 28% at the start. Most of the traffic happened in the first 2.5 months and then flatlined. The pages stayed indexed but became effectively invisible. The researchers' conclusion lines up exactly with what we tell clients: early gains "will likely fade within a few months" without genuine SEO strategy and human guidance.

Our reading of that experiment: Google didn't deploy an AI detector and zap those pages. The pages simply had no expertise, no authority, nothing original, so once the novelty-indexing wore off they had nothing to hold a ranking with. The decay wasn't a punishment. It was the absence of value showing its true level.

What we do not get penalised for (and use every single day)

We'll be transparent: AI is part of our workflow at SEO Engico, and it's part of our clients' workflows. Here's what's never caused a problem, in our experience or anyone's data:

• AI-assisted drafts with real human editing. A model helps with research and a first pass, then a human who actually knows the subject rewrites, corrects, adds opinions, and fact-checks. The output is human-quality because a human shaped it.
• AI for the boring technical bits. Drafting meta descriptions, alt text, first-pass schema markup, product descriptions from a spec sheet. Repetitive, factual, low-creativity tasks where a model genuinely helps and the output is verifiable.
• AI for ideation and structure. "Here are eight points I want to cover, suggest a logical order." We still decide the points. The model just helps arrange the furniture.

If your SEO programme uses AI this way, you are not at risk. You're just working faster than the agency next door that's still typing meta descriptions by hand.

How we actually use AI to write content that ranks

Since people always ask for the recipe, here's roughly how we produce a post that's meant to win, step by step.

1. Research by hand first. We pull real data from Search Console, DataForSEO and Ahrefs, read the current top results, and map the gaps. A model can't do this well because it doesn't have your live data or competitive context. This step is judgement, not generation.
2. Outline with the human in charge. We decide the angle and the points based on what we've actually seen work. AI helps tidy the flow. It does not get a vote on the strategy.
3. Draft with heavy human input. Opinion and experience sections we write ourselves. Explanatory sections might start as an AI draft and then get rewritten. The mix depends on the topic, and anything in the healthcare or finance bucket gets far more human hands on it.
4. Edit every sentence. Would we actually say this out loud? Is it accurate? Does it beat what's already ranking? If not, it's cut.
5. Add what a model can't fake. Real client outcomes (anonymised), numbers from campaigns we've run, dry jokes, strong opinions, the occasional "we got this wrong once and here's what we learned." This is the experience layer Google's E-E-A-T is hunting for, and it's the one thing no detector and no model can manufacture.

That last step is the whole game. We've written more on how the engineering-minded side of using AI for SEO fits into a wider system, and how it overlaps with getting your brand cited inside ChatGPT and AI Overviews, which is increasingly where the attention is going anyway.

The uncomfortable truth nobody selling detectors will tell you

The best AI-assisted content is undetectable because, by the time it's done, it is human content. It had a model involved in the process, the same way spellcheckers, grammar tools and research assistants have been involved in writing for thirty years. Nobody ever binned a piece because Word fixed a typo.

This very post had AI assistance in places. Can you point to which paragraphs? Probably not, because the whole thing went through our editorial process, carries our genuine opinions, and reflects what we actually see in campaigns. That's the point. The line was never human versus AI. It's good versus bad, and that line is decades old.

If you're chasing a high "human score" on a detector, you're optimising for a number that doesn't influence Google and isn't reliable anyway. You're polishing a metric that doesn't exist on the scoreboard.

The four questions to ask instead of "will this get detected?"

This is what we tell every client who's nervous about AI content. Stop checking detection scores. Check these four things before anything goes live:

• Does this answer the searcher's question better than what currently ranks? Not as well. Better.
• Does it include expertise only someone with real experience could provide?
• Would you happily put your own name on it in public?
• Does it back up its claims with authoritative sources where they're needed?

Four yeses, and it genuinely does not matter whether AI was in the room. Google won't penalise it, your readers won't care, and your rankings will reflect the quality of what you shipped. One no, and that's your problem to fix. The AI tool was never the issue.

If you'd rather not run that gauntlet alone, that's exactly the kind of content and quality system we build for clients. Have a look at how our SEO service approaches content, or just tell us what you're working on and we'll give you a straight answer about whether your AI workflow is safe. We've yet to meet a content problem that was actually caused by the robot rather than the brief behind it.