AI Didn't Make Content Cheap. It Made Being Wrong Cheap.

For about two years now, the loudest promise in marketing has been simple. AI makes content free. Push a button, get a blog post, fill the calendar, go home. The bottleneck is gone, the robots write, and all of us can finally scale the way the gurus always said we should.

I want to poke that balloon.

Because Google quietly updated its guidance recently to say it wants AI assisted content fact checked, and most people read that as a minor compliance note. It is not. It is Google telling you, out loud, where the real cost of AI content actually lives. And it is not where you think.

The cost didn't disappear. It moved.

Here is the thing nobody wants to admit at the AI cheerleading rallies. Writing was never the expensive part of good content. Being right was.

Think about how a decent article used to get made. Somebody who knew the subject sat down and typed it. The knowing and the typing happened inside the same head at the same time. The expertise and the output were fused. So when we talked about the cost of content, we lumped it all together and called it writing.

AI pried those two things apart. Now the typing is basically free and instant. The knowing is not. The model will produce fourteen paragraphs of confident, grammatically perfect prose about your industry whether or not a single sentence is true. It does not know the difference between a fact and a very likely sounding guess. It never did.

AI did not make content cheap. It made being wrong cheap, and being right just as expensive as it always was.

So when Google says fact check your AI content, read it plainly. They are not adding a new rule. They are pointing at the bill that AI handed you and quietly slid under the door. The production cost went to zero. The verification cost went up, because now you have a firehose of plausible text and no human brain attached to any of it.

Plausible is not the same as true, and the gap is widening

A hammer is not a house. A sentence that sounds correct is not a sentence that is correct. Those feel like obvious statements until you watch a content team ship forty AI drafts in a week and realize nobody actually read them closely, because reading them closely would have taken longer than writing them the old way.

That is the trap. The speed of generation seduces you into skipping the slow, boring, unglamorous work of checking. And the model is really good at sounding sure. It will cite a statistic that does not exist. It will attribute a quote to the wrong person. It will state a regulation that changed three years ago as if it is current. Not maliciously. It just pattern matches toward whatever looks like an answer.

Now scale that. One wrong fact in one post is an embarrassment. Four hundred posts generated at volume, each carrying two or three confident errors, is a reputation problem with compound interest. You are not publishing content. You are publishing liability, nicely formatted.

The brands that win the next few years will not be the ones who generate the most. They will be the ones who verify the fastest.

Why Google actually cares, and why it is not about Google

It would be easy to treat the fact checking guidance as another ranking hoop. Jump through it, appease the algorithm, move on. That reading is too small.

Google cares about accuracy because its own product is now built on top of your content. The AI answers at the top of the results page summarize the web. If the web fills up with confident nonsense, those summaries become confident nonsense, and Google's whole value proposition rots from the inside. They are not policing you for sport. They are protecting their supply chain.

And here is the uncomfortable part for anyone who has been treating search traffic as a given. Desktop click through rates slid in the second quarter. Not catastrophically, but the direction is clear and it has been clear for a while. When the AI answer gives people what they came for, they do not click. The answer is the destination now, not the doorway.

So let me ask the obvious question. If fewer people are clicking, and the AI is answering on your behalf, what exactly are you optimizing for?

You are optimizing to be the source the AI trusts enough to repeat. That is the game now. And an AI, whether it is Google's or anyone else's, is going to lean on sources it can rely on to be factually clean. Accuracy stopped being a virtue and became a distribution strategy.

Gemini's UTM tags are a bigger tell than they look

Buried in the same batch of news was a small, nerdy detail. Links coming out of Gemini started showing up with UTM tags attached. If you do not live in analytics, that sounds like a footnote. It is not.

A UTM tag is how traffic announces where it came from. It is the little luggage label on a visitor that says, I arrived from this specific source. When Gemini starts tagging its outbound links, it means the referral is becoming visible. You can finally see, in your own reports, when an AI assistant sent someone your way.

The moment AI referrals show up in your analytics is the moment getting recommended by AI stops being a theory and starts being a number.

I have been saying for a while that answer engine optimization is not a buzzword you can safely ignore until later. This is the proof. The channel is becoming measurable. And once something is measurable, it gets managed, budgeted, and fought over. We went from, nobody can prove AI sends traffic, to, here is the tag, in a very short window.

Connect the two stories and the shape of things gets clear. Google wants your content accurate so its AI can repeat it. The AI is starting to label the traffic it sends you. Accuracy is the price of admission, and the referral is the reward. Those are not two separate news items. They are the same story told from both ends.

The lawsuits, and the quiet permission slip

There was also legal news. A judge dismissed two of the AI related suits. I am not a lawyer and I am not going to pretend the whole thing is settled, because it is not, and there will be more cases.

But notice the mood it creates. Every time a case gets tossed, the industry exhales and treats it as permission. See, the robots are fine, carry on. And that collective shrug is exactly how teams talk themselves out of doing the hard verification work. If nobody is going to sue me, why slow down to check?

Because the court is not your audience. Your customer is. Legal clearance to use AI is not the same as earning the right to be believed. A judge dismissing a copyright claim tells you nothing about whether the content you shipped last Tuesday is accurate, useful, or worth a human being's trust. Those are different courtrooms entirely, and the second one never adjourns.

What fact checking actually looks like when you mean it

Let me get practical, because I do not want this to read as hand wringing. Fact checking AI content is not glamorous and it is not complicated. It is just work that most people skip. Here is where the real checking happens, in rough order of how often AI gets it wrong:

• Numbers and statistics. If the model states a figure, assume it is invented until you find the real source. It is the single most common failure, and the most damaging.
• Quotes and attributions. Who actually said this, and did they say it this way? AI confidently assigns words to the wrong mouths.
• Anything time sensitive. Rules, prices, features, dates. The model's knowledge has a horizon and it will state stale facts as current ones.
• Claims about your own product. Nothing erodes trust faster than your own blog describing a feature you do not offer.

None of that requires a special tool. It requires a human who knows the subject reading slowly and asking, how do I know this is true? That person is your actual cost center now. Not the writer. The checker. And if your content operation does not have one, you do not have a content operation. You have a confident guessing machine with your logo on it.

The new scarcity

For twenty years the scarce thing in content marketing was the ability to produce. Writing was hard, so people who could write well had leverage. That scarcity is gone. Anyone can produce now. The floor collapsed.

So the value moved up a floor. The scarce thing now is credibility. The ability to say something and have it be reliably, checkably true. In a world drowning in plausible text, the rarest and most valuable asset is being right on purpose.

When everyone can generate, the only edge left is being trustworthy.

That is why Google's little guidance update matters more than the headline suggests. It is not asking you to add a step. It is telling you that the step you removed, the knowing, the verifying, the standing behind your words, was the only part that was ever worth paying for.

AI did not make content free. It made the cheap part free and left the expensive part exactly where it was. The companies that understand that will treat verification as a feature, not a chore. The ones that do not will flood the web with confident, polished, slightly wrong garbage, wonder why the AI never recommends them, and blame the algorithm.

The robots can write now. Fine. The question that decides who wins has not changed in a hundred years, and no model has answered it yet.

Is it true?