AI Content Doesn't Rank Lower Because It's AI. It Ranks Lower Because It's Boring.

Here is a belief that hardened into fact almost overnight this week: AI generated pages rank lower, so you should stop using AI to write. The screenshots went around. The takes wrote themselves. Everyone nodded along.

I want to slow that down.

Because the moment a correlation shows up in a study, our whole industry treats it like a commandment carved into stone. AI content ranks lower, therefore AI is the problem, therefore go back to writing everything by hand. That is a clean story. It is also, I think, mostly wrong. Not because the data is fake, but because the conclusion people drew from it skipped three or four steps.

So let me ask the obvious question that almost nobody asked. Ranks lower than what?

What the numbers actually measured

When an analysis says pages flagged as AI written tend to rank lower, sit with what that sentence contains. It contains a detector, which is imperfect. It contains a sample of the open web, which is enormous and mostly junk. And it contains a snapshot in time, which is not the same as cause and effect.

Detectors do not read a page and find a watermark that says "a robot made this." They guess, based on patterns, whether text looks statistically like machine output. That means the pages getting flagged as AI are disproportionately the ones that read like machine output. Flat. Even. Predictable. The kind of prose that never surprises you.

Now think about who publishes that kind of prose at scale. It is people running content at volume, thin affiliate sites, programmatic pages spun up to catch a keyword, and marketers who typed one prompt and hit publish. That is the population the detector is best at catching, because that is the population that made no effort to sound like anything but a language model.

The detector is not measuring AI. It is measuring laziness that happens to correlate with AI.

So when someone tells me AI pages rank lower, my honest reaction is, yes, of course they do, in the same way that pages with no editing, no point of view, and no reason to exist also rank lower. The AI is incidental. The absence of a human decision is the thing.

Correlation is a hammer, and everyone wants a house

A hammer is not a house. A single correlation is not a strategy. And yet every few months our field grabs one finding and tries to build an entire worldview out of it.

Remember when the lesson was that long content always wins because the top results averaged more words? People started padding thousand word answers into four thousand word marathons, and rankings did not follow, because word count was never the cause. It was a side effect of thoroughness. The thoroughness was doing the work. The word count was just along for the ride.

This is the same movie with a new cast. AI detection is standing in for a deeper variable, and that variable is whether a real person made real choices about what to say. When you confuse the marker for the mechanism, you optimize for the wrong thing. You start running your drafts through a detector trying to "humanize" the text, tweaking sentence length until the tool goes quiet, and you feel productive. You have changed nothing that a reader cares about. You have just fooled a classifier.

That matters, because Google does not rank the classifier's opinion. Google ranks whether the page was worth landing on.

Google's actual position, said plainly

Here is what Google has been consistent about, and I think it is more honest than the panic gives it credit for. They do not care how the words got onto the page. They care whether the page is helpful, original, and made for people rather than for a ranking system. Machine assistance is not the sin. Producing content with the primary purpose of gaming search is the sin.

Read that carefully and the AI question mostly dissolves. If you use a model to help you organize research, draft a section, tighten a paragraph, or get past a blank page, and then you bring judgment, experience, and an actual opinion to it, you have not broken any rule. If you use the same model to flood the internet with plausible filler about topics you know nothing about, you have.

The problem was never the tool. The problem is when the tool becomes the whole author.

I have watched this play out with clients on both sides. One team used AI heavily and their organic traffic kept climbing, because a subject matter expert sat between the model and the publish button and refused to ship anything they would not put their name on. Another team banned AI entirely, wrote everything by hand, and still produced forgettable pages that ranked nowhere, because "a human typed it" is not a quality standard. It is just an input.

The social search data everyone skipped past

Buried under the AI headlines this week was something I found more useful, and it got a fraction of the attention. Search Console has been expanding the kinds of reporting and social related data it makes available more broadly. Boring by comparison. No villain, no panic. But it points at a shift that is bigger than the AI detection debate.

Discovery is fragmenting. People do not just type into one search box anymore. They ask a question inside an AI assistant, they watch a video, they read a thread, they get a recommendation from a tool that summarizes ten sources and never sends them to any of them. The old model where you rank a blue link and someone clicks it is still real, but it is one lane on a road that keeps adding lanes.

Why does that connect to the AI content question? Because the reason to have a point of view, a real perspective, a specific take rooted in something you actually know, is not just to beat a detector. It is that generic content has nowhere left to hide. When an AI answer engine is summarizing the web to give someone a direct response, it does not need your restated, average, machine smooth version of the same five facts everyone else published. It already has those. What gets pulled, cited, and recommended is the source that says something the others did not.

Sameness was always weak. Now it is invisible.

The opt out trap nobody thinks through

Part of this week's conversation touched on AI opt outs, the idea that you can block AI systems from using your content. And I understand the instinct. You worked hard on something, a machine hoovers it up, and it spits your insight back to a user who never visits you. That stings.

But think one move ahead before you slam that door. If the way people increasingly find answers runs through AI systems, and you have made yourself unreadable to those systems, what exactly have you protected? You kept them from quoting you, and in doing so you kept them from recommending you. You are not in the room when the decision gets made.

There is a real tension here and I am not pretending it is simple. Some publishers with genuine leverage and a paywall to protect have a legitimate case for restriction. But most businesses are not in that position. Most businesses are trying to get found, and quietly opting out of the fastest growing discovery channel to make a principled point about training data is a decision you can regret for a long time. Being uncopyable and being undiscoverable are not the same goal, and it is easy to buy the second one while thinking you bought the first.

Metadata, and the quiet return of fundamentals

The other thread this week was renewed guidance on metadata, titles, descriptions, the plumbing. And I notice how unfashionable that sounds next to AI headlines. Nobody builds a conference talk around writing a good title tag.

But here is what I keep seeing. The teams obsessing over whether a detector flags their prose are frequently the same teams shipping pages with vague titles, descriptions that describe nothing, and structure that gives neither a human nor a machine a clear signal about what the page is for. They are worried about an exotic threat while the front door is unlocked.

Metadata is not glamorous. It is how you tell both a search engine and an answer engine, in plain terms, what this page is and who it serves. Get it wrong and it does not matter how the words were written, by hand, by model, or by committee. The page is mislabeled. You put a great product on the shelf with the wrong sign over it.

Fundamentals feel like a step backward when everyone is chasing the new thing. They are usually the step forward.

So what do you actually do with all this?

Strip away the week's noise and the guidance is almost embarrassingly consistent with what good marketers already knew. Let me put the useful part in one place, because this is the one spot a short list earns its keep:

• Use AI to remove friction, not to remove yourself. Drafting, research, structure, cleanup. Fine. Authorship, judgment, and point of view stay with a person who knows the subject.

• Stop optimizing for detectors. They are a proxy for quality, and a bad one. Optimize for the reader who is deciding whether to keep reading.

• Say something only you can say. Your experience, your data, your opinion, your mistakes. That is the part no model can generate, because it does not have your life.

• Fix the plumbing before you fear the ghost. Titles, descriptions, structure, clarity. Boring wins more often than clever.

• Think hard before you wall yourself off from where discovery is heading.

None of that is a hack. It is not supposed to be. The whole reason the AI penalty story spread so fast is that it promised a simple lever. Turn off AI, rank higher. If only.

The line that actually holds

Here is where I land. The headline that AI pages rank lower is true and useless at the same time, the way "restaurants with bad reviews get fewer customers" is true and useless. Of course they do. The reviews are downstream of the food.

AI did not make your content worse. A missing point of view made your content worse, and AI just made it cheaper and faster to produce content that never had one. The tool lowered the cost of saying nothing, and a lot of people took the deal.

So no, I am not telling you to fear the robot or to fire it. I am telling you the machine will happily help you produce the most average page on the internet, and the internet is drowning in average pages, and average has never ranked, gotten recommended, or been remembered.

The question was never whether a human or a machine wrote it. The question is whether it was worth writing at all.

Answer that one honestly and the detector debate stops mattering. Duck it, and no amount of humanizing your text will save you.