AI Can Now Copy Everything Except Your Data and Your Opinion

The math of content marketing changed the moment a language model could write a competent blog post in seconds. Whatever your team produced to explain a concept, define a term, or walk a reader through a process, a model now writes a passable version instantly, at no marginal cost, in any quantity you want. The floor for adequate content dropped to zero.

That gets discussed mostly as a threat, and for plenty of content operations it is one. It also clears something up. Once the generic is free and unlimited, the only content left with real economic value is the content a machine cannot produce, which comes down to a genuine point of view and original data the model has never seen. Everything else is a commodity now, and commodities do not build authority. They fill space that was already full.

So the case for thought leadership stops being about taste and becomes structural. You publish a strong position not because it reads as more confident but because it is one of the few things still scarce.

What a model can and cannot make

The line between commodity content and defensible content runs right along what a model can generate from its training data alone.

A model can synthesize anything that has been said enough times. It is very good at the average of what is already known, which is exactly why explanatory and definitional content lost its edge: the model has read ten thousand versions and will produce the next one on request. What it cannot do is generate a dataset it never ingested, or hold a position that cuts against the consensus it was trained to reproduce. Original research and real conviction sit outside its reach almost by definition.

This is not just a tidy theory. It shows up in what the engines actually cite. A Search Engine Land study found 52.2% of AI-cited passages contained original or owned data, far above the rate at which original data appears in content overall. The Princeton-originated GEO research on AI visibility found that adding statistics lifted citation likelihood by roughly 41% and citing sources by about 40%, among the biggest and most consistent gains of any tactic tested. The same systems commoditizing generic writing are reaching, disproportionately, for the specific and the original. They want the thing they cannot make.

Content that generates its own data is more citable than content that summarizes someone else's, for the simple reason that the model has to come to you for it. The rest it can write itself.

An opinion with nothing under it is also a commodity

There is a failure mode worth flagging, because "have a strong point of view" gets misheard as permission to be loud. A hot take with nothing behind it is as easy to mass-produce as a listicle. Models generate provocative opinions fine. What they cannot generate is a provocative opinion that turns out to be backed by data no one else has.

The 2025 Edelman-LinkedIn B2B Thought Leadership Impact Report is direct about the quality bar. Decision-makers read a lot of thought leadership and are mostly unimpressed, with only a small share rating what they read as very good. When it does clear the bar, it works. 81% of hidden buyers say strong thought leadership helps them see challenges or opportunities they had missed, and 86% of decision-makers prefer content that challenges their assumptions. The audience is asking for a point of view that reframes how it thinks, but one it can test. As the research gets summarized in the field, thought leadership without original research is just opinion, and the research is what gives a perspective enough weight to act on.

That is the combination the moment calls for. Not opinion in place of data, and not data in place of opinion, but a position only your evidence could support. The conviction earns the read. The proprietary data earns the belief, and gives the model a reason to cite you instead of paraphrasing the consensus.

For Q4: fewer pieces, sharper, sourced

If generic content is free, producing more of it is paying to enlarge an infinite pile. What follows from the data is the opposite of volume. It is fewer assets, each carrying a real argument and real evidence.

That fights a lot of content-team muscle memory, which was built for cadence and throughput, for keeping the calendar fed. The research on high-performing programs points the other way: the brands growing fastest make fewer, more deliberate assets and treat research as an ongoing program instead of a one-off. When the average is automated, the return moves to the exceptional, and you do not get the exceptional by producing the average faster.

The test for any planned piece this quarter is one question: could a model have written this from what it already knows? If yes, you are adding to the commodity pile, and the pile does not need the help. If no, because the piece carries a position you will defend and data only you have, you are making one of the few things that still holds value, and one the model has to come to you for.

An opinion backed by proof is the last scarce asset in a category where everything else just went free. The companies that see that will spend the next few quarters becoming the source. The ones that don't will spend them adding to a pile nobody wants more of.