Somewhere in the past year, the applause stopped. Not the usage, mind you. Pew Research found that about half of American adults now use AI chatbots, up from a third in 2024 and less than a quarter the year before that. A full 24 percent of Americans use a chatbot every single day. By any adoption measure, AI is winning. But the mood around it has changed in a way that anyone paying attention can feel.

A couple of years ago, every software company seemed determined to stuff a “Write with AI” button into every empty box on your screen. AI was the feature everyone bragged about. This July, LinkedIn went the other direction and added a button that lets users report a post because it “Seems like AI slop.” At the same time, the company killed its own AI rewrite feature and replaced it with a simple proofreader, and it announced that generic content produced with little human involvement will get reduced distribution beyond your own network. Wired has reported on writers deliberately changing their style so they won’t be mistaken for AI, avoiding em dashes, restructuring sentences, and intentionally leaving more imperfection in their work as proof of humanity.

Anthropic went even further. As of August, its new Claude models embed invisible watermarks directly into the text they generate, part of complying with the EU AI Act’s transparency rules, and the marks apply worldwide, not just in Europe. The watermark travels with copied text and survives light editing. That has already kicked off the next argument: what happens when AI didn’t actually write something, but merely helped edit or proofread it? Even Anthropic admits a detected mark only proves that Claude touched the content somewhere, not that Claude authored it.

This article is a prime example. I used AI to do the research for the content.I told AI what to do with that research and how I wanted to present this information. It is based on my thoughts and ideas. Both ChatGPT and Claude have been used to refine my concepts into the written article. Claude did the final editing, therefore I am sure it is watermarked and would be flagged as AI-generated content. Still, the article is mine.

We have gone from people hiding the fact that they didn’t know how to use AI to people hiding the fact that they did.

And writing is only one front. People are worried about AI taking their jobs. Artists and authors are worried about their work being used to train models. Teachers and parents are worried that students will stop learning how to think. Communities are fighting data centers over electricity and water. Businesses are worried about privacy and liability, and consumers increasingly question whether anything they see online is real. The tide has clearly turned. But I don’t think the story is as simple as saying people have become anti-AI, and the numbers back me up on that.

We’re Using AI More and Trusting It Less

Here is the strange part. The same Pew research showing record adoption also found that 63 percent of Americans believe AI is advancing too quickly, and when asked about its impact over the next twenty years, 40 percent expect a negative effect on society while only 16 percent expect a positive one. Gartner found something even more pointed for businesses: half of U.S. consumers said they would prefer to do business with brands that don’t use generative AI in consumer-facing messages, advertising, and content, and 49 percent believe generative AI has actually made the quality of available content worse.

So people aren’t rejecting AI. They’re rejecting what AI is doing to them, and that is an important distinction.

I have a somewhat unusual seat from which to watch all of this. I make a living creating and selling AI tools, and I believe AI can ultimately have a tremendously positive impact on individuals and society. I’ve also written books like Bluff and The Assimilation Protocol that explore some of the ways it can go badly wrong. I don’t think those positions contradict each other. Believing that AI can improve our lives doesn’t require pretending there aren’t legitimate reasons to worry about it. In fact, I’ve come to think this backlash may be necessary.

AI Slop Isn’t Really an AI Problem

Let’s start with the phrase everyone loves right now: AI slop. There is certainly plenty of it. AI has reduced the cost of producing average content to nearly zero, which means we now have the ability to create an essentially unlimited amount of average content. That probably wasn’t the breakthrough anyone was hoping for.

But the real problem isn’t that machines learned to write. The problem is that humans discovered they could publish without thinking. Spam didn’t make email bad, clickbait didn’t make websites bad, and robocalls didn’t make telephones bad. Each was an abuse of a technology that dramatically lowered the cost of communication, and AI has simply lowered that cost again. Someone can now crank out fifty LinkedIn posts in the time it used to take to write one, but that doesn’t mean they suddenly have fifty times as many worthwhile ideas. That gap between output and ideas is exactly where slop comes from.

I worry that we’re starting to confuse the tool with the behavior. Consider the poor em dash. Writers have been using it for hundreds of years, and now someone can drop one into a perfectly human-written article and immediately be accused of using ChatGPT. I used to use them myself, and I gave them up for precisely this reason. We’ve reached the absurd point where writers are intentionally making their work less polished in order to prove a human wrote it, which means we may actually start rewarding worse writing because imperfection has become evidence of humanity.

That isn’t progress.

The Backlash Is Healthy, Until It Becomes Superstition

There is a good side to what’s happening. For the first couple of years after ChatGPT exploded onto the scene, the dominant question was “what can AI do?” That’s the fun question. Now we’re asking a much harder one: what should we let AI do? Should it make hiring decisions, diagnose diseases, write school assignments? Should companies replace entire customer service departments with bots? Should someone be allowed to generate thousands of articles and flood the internet with them? Should a company be able to build a massive data center that consumes enormous amounts of electricity and water without the surrounding community having much say in the matter?

Those are reasonable questions, and the data center one in particular is not something we should wave away as anti-technology hysteria. This past July, Reuters covered the first nationwide protest campaign against data center construction, with 142 demonstrations across 42 states in a single day. A Reuters/Ipsos poll from June found that only about a third of Americans approve of the current pace of data center construction, and just 14 percent would support one being built in their own community. Businesses can’t simply respond by saying “but AI is the future.” People still have to live next to the future, and the technology industry needs to acknowledge that there are real tradeoffs here.

Where I get concerned is when healthy skepticism curdles into something closer to contamination theory. AI touched this, therefore it’s bad. AI helped write this, therefore it’s dishonest. AI reviewed this document, therefore the author didn’t write it. That logic doesn’t hold up either. A watermark can tell us that an AI system participated somewhere in the process, but it cannot tell us where the idea came from, how much human judgment was involved, or whether the final result is worth reading. Those are very different questions, and we keep collapsing them into one.

The Job Question Is Different

The fear about jobs deserves special attention, because frankly, the technology industry created it. For several years, some of the loudest promises surrounding AI have been about reducing headcount, replacing workers, and doing more with fewer people. Then we act surprised when workers don’t celebrate. If you tell people that a technology’s greatest benefit is that companies won’t need them anymore, you’ve done a pretty good job of building an anti-technology movement.

I’ve said for a long time that if you view AI as a means to replace human effort, you will be fearful, but if you see it as a tool for amplifying human potential, you will be inspired. Those aren’t two marketing angles on the same technology. They are two completely different philosophies, and they produce completely different products.

Consider a coach. A coach can only personally work with so many people, because there are only so many hours in a day. The ReplicAItor Coach platform I developed uses AI to extend a coach’s expertise beyond those hours, making their methodology available between sessions, evaluating client submissions, maintaining continuity, tracking progress, and providing support when the coach isn’t personally available. The point isn’t to make the coach disappear. The point is to make the coach’s expertise more useful, and that distinction is going to matter more and more. People will embrace technology that makes them more capable long before they embrace technology explicitly designed to make them unnecessary.

Businesses Are Going to Have to Change How They Talk About AI

This backlash has real implications for companies that have invested heavily in AI, and the first one is simple: stop assuming “AI-powered” is automatically a selling point. For the past few years, companies have slapped AI onto product descriptions the same way companies once put “.com” after everything. That worked while AI itself was exciting. Given that half of consumers now say they’d rather do business with brands that skip generative AI in their customer-facing content, it may now have the opposite effect. Customers don’t actually want AI. They want an answer faster, better service, fewer mistakes, something personalized to them. They want to save time, make more money, and get help making decisions. AI is simply one way of delivering those outcomes, and if the technology creates the outcome without creating a new trust problem, nobody will complain.

The second change is that businesses need to preserve human agency. Gartner’s research on AI shopping is instructive here: even in low-stakes categories like household supplies, only 11 percent of consumers were willing to let AI actually make the purchase for them, while roughly three times as many were comfortable letting AI narrow their choices. People are telling us something. Help me, don’t take over. That may become one of the most important design principles of the next generation of AI products.

Businesses should also start distinguishing between AI-assisted, AI-generated, and AI-autonomous, because those are not the same thing. If I write an article and ask AI to find a typo, did AI write my article? Obviously not. If I hand AI my ideas, my stories, and my argument, ask it to organize them into a first draft, and then completely rewrite that draft, who wrote it? That one’s harder. And if I type “write me an article about AI” and immediately click Publish, that’s something else entirely. Yet we keep trying to shove all three into the same bucket. We need a more intelligent vocabulary for this, and the companies and platforms that develop one first will earn trust the others won’t.

Maybe AI Needed the Backlash

Technology tends to be easiest to love before we understand what it changes. The first stage of the AI revolution was fascination, and the second was adoption. I think we’ve entered the third stage: legitimacy. AI now has to prove that it deserves the role we’re giving it.

That means businesses will have to demonstrate value instead of simply announcing that they use AI. It means AI companies will have to take electricity, water, copyright, transparency, and privacy seriously rather than dismissing them as barriers to innovation. It means writers will have to contribute actual ideas rather than generating endless variations of whatever everyone else is already saying. It means employers will have to decide whether AI exists primarily to eliminate their people or to make their people dramatically more capable. And it means those of us building AI products need to ask a harder question than whether something can be automated. We need to ask whether it should be.

I still believe AI can have a net positive effect on society. I actually believe that more strongly than ever. But the companies that win the next phase of AI probably won’t be the ones using the most AI, talking about it the loudest, or finding clever ways to hide the fact that they used it. They’ll be the ones that use AI where it genuinely makes people better at what they do, keep humans involved where judgment and responsibility and empathy matter, give customers control, and stay transparent enough to earn trust without treating AI assistance as some kind of confession. They’ll remember that technology is supposed to serve people, not the other way around.

So yes, the tide has turned. But I don’t think it has turned against AI. It has turned against unquestioned AI, and that is a much higher bar. It’s also the right one.