Here's the ELI5.

You know how your phone suggests the next word when you're typing a text? ChatGPT is that, but way more powerful. It reads everything you've written so far, guesses the most likely next word, writes it down, then does it again. And again. And again. That's what tools like ChatGPT and Claude are doing under the hood. Really good guessing, one word at a time.

This is super useful for a lot of things. Summarizing a legal document? You want the most probable interpretation. Writing a SQL query? There's a correct answer and the model is good at converging on it.

"Most probable" is a good thing when the goal is accuracy/correctness.

But marketing's entire job is to be the least probable thing in someone's feed. The ad that makes you stop scrolling. The homepage copy that sounds like nobody else in the category. The voice so unique you could identify it without looking. Marketing works by breaking patterns, and an LLM works by predicting them.

So all that generic content your AI is creating... It's a feature. Not a bug.

AI is a Consensus Machine.

Just like you wouldn't blame a calculator for being bad at poetry, we need to stop blaming AI for not being innovative.

But the debate about whether AI can write good copy is the wrong debate. Better prompting helps. A lot, actually. You can tune a model to produce content that matches an established brand voice, and the output will be decent. But writing on-brand copy was never the hard part. Any competent copywriter could do that once the brand existed.

The hard part was always crafting the voice to begin with. The strategic stuff. The stuff that requires taste and judgment.

Yet that's exactly where most companies are now pointing the Consensus Machine. They're not just asking AI to write the copy. They're asking it to figure out what to say, how to position, what voice to use. They're using a consensus engine at the strategy layer, and then wondering why their strategy is producing shitty results.

The evidence

Two layers here. The first one is table stakes. The second is where it gets interesting.

Good ol' startup blue

RSAC 2026, the big cybersecurity conference, happened a few weeks ago. A team from Highwire analyzed every booth there. The numbers: 45.7% of companies led with AI messaging. Almost the exact same percentage used blue as their primary brand color. 22.2% did both. And over 50% of the value propositions had meaningful overlap with at least one other brand.

Half the booths could have swapped signage and nobody would have noticed.

Maybe cybersecurity is an extreme case. Technically complex products, hard to differentiate at the messaging layer. But I don't think it's as extreme as we'd like to believe.

Here's a fun game you can play while you're waiting for that Claude query to finish up: Pull up the websites of any five competing SaaS companies in project management, CRM, analytics, whatever. Cover the logos. Try to tell them apart from the copy. You probably can't. The Consensus Machine isn't just producing consensus content. It's producing Consensus Strategy, because when every marketing team uses the same tools trained on the same data optimizing for the same metrics... I mean wtf do you think is going to happen?

Meanwhile, consumer tolerance for AI-generated content collapsed from roughly 60% favorable in early 2023 to 26% as of early 2026, per a study from Billion Dollar Boy via Digiday. And Gartner found that 50% of consumers now say they actively prefer brands that avoid using generative AI in their consumer-facing content.

An IRL A/B test

When Italy temporarily banned ChatGPT in 2023, researchers got something incredibly rare in social science: a natural experiment with a clean control group. One country lost access to AI; neighboring countries with similar profiles kept it.

The results: when AI was removed, content became 15% less lexically similar across the platforms they studied. Engagement rose 3.5%. And this happened despite posting frequency and average content length both declining.

Fewer posts. Shorter posts. More variety. Higher engagement. The math only makes sense if AI was actively compressing the space of what people were creating, and removing it let the space expand again.

It's the closest thing we have to a controlled experiment on this question. Not perfect (VPNs exist, the ban was short, etc), but it points hard in one direction.

Some other related research while we're at it: Doshi and Hauser found that AI makes each individual feel more productive, but when everyone in a group uses it, the group's total output gets less creative and less diverse. A creativity illusion at the individual level that has a real creativity tax at scale.

The real divide

So if the mechanism is real, and if consensus works great for some tasks but hurts others, where's the line?

Think of it like a restaurant. Every restaurant has three layers.

The kitchen: prep, inventory, scheduling, supplier logistics. You want speed and consistency here. Consensus is fine. "Most probable" equals "most correct." AI shines here.

The dining room: what guests actually experience. The plating, the menu descriptions, the lighting, the staff. This is where consensus starts to hurt, because "most probable presentation" looks like the chain restaurant down the street.

But then there's the concept. What kind of restaurant is this? Why would someone choose you over the four places on the same block? What's the identity, the point of view, the thing that makes this place this place? Two restaurants can share a kitchen supplier and have completely different identities. The concept is the differentiator.

And the concept is where the Consensus Machine does its worst damage. Because right now, most companies aren't just using AI to write the menu descriptions (dining room). They're using it to figure out what kind of restaurant to be in the first place (concept). They're asking a consensus engine to design their positioning, their voice, their strategy. And the most statistically probable restaurant concept is a mid-tier Italian place with exposed brick and a cocktail menu that says "handcrafted." You've been to that restaurant. Everyone has. It's fine. But you also don't remember it.

The mistake isn't using AI. The mistake is using AI the same way at all three layers.

Companies are starting to figure this out, and you can see it in the job market. Growth Unhinged reported that storyteller job postings doubled over the past year. Anthropic posted a Head of GTM Narrative role at $300K+. The companies building AI are hiring humans to do their marketing.

While everyone else is celebrating how many marketers they just laid off. Sit with that for a second.

The widening gap

Here's the forward-looking part, and I think it's the most important implication of the whole thesis.

If AI pushes most companies toward the center, and if audiences are migrating toward the edges, then the companies that maintain distinctive positioning don't just have a static advantage. They have a compounding one. Every quarter that the middle gets more crowded, the edges get more valuable. Every company that hands its concept to the Consensus Machine makes it easier for the companies that don't.

This is the opposite of how competitive dynamics usually work. Usually, differentiation gets harder over time. Competitors copy you. The shelf life of new tactics gets shorter and shorter. Categories mature and consolidate. But AI has inverted this: the force pushing companies toward sameness is now so strong that simply not doing that has become a form of differentiation. The Distinctiveness Dividend.

And the Dividend compounds. A company with a recognizable voice and a specific point of view doesn't just win today. It builds an audience that seeks out that authenticity, which makes the next piece of content more valuable, which attracts more of the right audience, which further separates it from the companies still publishing probability-weighted consensus content.

And the gap between distinctive and generic is only going to widen.

So what do you do with this?

I'm not going to give you a framework. (The frameworks are what got us here.) The whole point of understanding the mechanism is that it makes the solutions obvious. AI equals consensus. Marketing equals differentiation. These are structural opposites. Once you see that, you can figure out where to draw the lines in your own company. You know your kitchen. You know your dining room. You know your concept. The question was never "should we use AI?" It was always "which layer am I on, and does consensus help or kill me here?"

The only strategic question is whether you have the discipline to treat them differently.

Most companies won't. Most will keep letting the Consensus Machine design the concept because the output is fast and clean and nobody gets fired for choosing the safe thing. And every one of those companies will make it a little easier for yours to stand out.

If you choose to.