Publish Anyway: Where the Moat Goes Once Content Stops Being One

TL;DR. The strategic answer to AI commoditizing expertise is not to publish less. Withholding expertise builds no authority and earns no trust. Publish the method generously, and demonstrate judgment through specific real cases instead of generic checklists. Then invest in the assets an AI answer cannot copy: a direct relationship with an audience, a recognized name, and original first-hand material. Being ignored is a bigger risk than being copied.

Part 3 of a three-part series on what happens to expertise once it goes public in the AI era.

Part 1 argued that publishing expertise is the price of building authority, and that the same act of publishing partially trains the systems that commoditize it. Part 2 drew the line more precisely: only the explicit, reproducible layer of what you know transfers into those systems, while the tacit judgment underneath does not. This piece is about what to do with that split once you accept it as permanent.

The false choice this series has been circling

Every conversation about expertise and AI eventually collapses into two options, both of them bad.

Option one: publish everything you know, and watch a model absorb it, restate it, and hand it back to a million people who will never know your name. Option two: withhold the good stuff, keep your best thinking behind a wall, and protect the asset by keeping it scarce.

We reject both, because both misread where the value actually sits. The first assumes the published artifact is the asset, so giving it away is a loss. The second assumes scarcity is protection, when in practice scarcity mostly produces silence. Neither framing survives contact with what is happening in the market.

Here is the uncomfortable part. The choice was never really between being copied and being safe. It was between being copied and being invisible. And invisible is worse.

The risk you should actually fear

Start with the asymmetry, because the prescriptions only make sense once you feel it.

Most people who worry about AI copying their expertise are solving for a problem they do not have yet. The far more common failure, the one that quietly ends most expert brands, is that nobody encounters the work at all. Invisible expertise builds no authority. It earns no trust. It generates no business. A brilliant insight that stays in your head, or in a private deck, or behind a form nobody fills out, is functionally the same as no insight.

Being copied at least requires that you were worth copying, and that you were visible enough to copy from. That is a problem of relevance. Being ignored is a problem of existence, and it is much harder to recover from.

So the strategic answer is not to publish less. It is to change what you think you are publishing. Stop treating the artifact on the page as the asset. Start treating it as a costly signal that points at assets the artifact itself can never contain.

What does the signal point to? Four things. The relationship with the people who read it. The recognized identity of the source behind it. The speed at which you move to the next insight before anyone else has caught up. And the judgment you demonstrate without ever fully handing it over.

Publish the method, demonstrate the judgment

Part 2 described the method and judgment split as a fact about how knowledge transfers. Now we want to use it as a decision rule.

The method should be published generously. The method is the explicit, reproducible layer: the framework, the steps, the checklist, the way you approach a class of problems. This is precisely the part that models absorb well, which is exactly why withholding it earns you nothing. The commodification of the general method is already happening in the ambient environment, with or without your contribution. Guarding your version of a widely known approach protects an asset that has already been priced to zero. You pay the full cost of silence and get none of the authority.

The situated call should be demonstrated, not generalized. The situated call is the tacit judgment: knowing which method applies to this messy case, when to break your own rule, what the client is really asking, why the obvious answer is wrong here. This does not compress into a checklist. The moment you try to generalize it into transferable instructions, you strip out the very thing that made it valuable, and you hand the reproducible husk to the machine.

So demonstrate judgment through specific, real cases. Show the actual decision on the actual problem, with the constraints and the tradeoffs and the thing that almost went wrong. A reader can watch you reason and still not be able to reproduce the reasoning on their own hard problem. That gap is the point. It is durable precisely because it does not transfer cleanly.

Put simply: give away the map, and let people watch you navigate. The map is cheap now. The navigation is not.

The market is already pricing this distinction

This is not an aesthetic preference. It is where distribution and citation are visibly concentrating, even in a field flooded with generated text.

Graphite analyzed roughly 65,000 URLs published between 2020 and 2025. AI-generated content briefly overtook human-written content in raw volume in late 2024, then settled near parity. You would expect that flood to drown out human sources. It did not. 86% of articles ranking on Google's first page were still human-written, and 82% of the sources cited by AI answer tools such as ChatGPT and Perplexity were human-written (Graphite, 2025, reported by Axios in October 2025).

Read that carefully. The volume of generic content roughly doubled toward parity, and the share of attention going to recognized, original human sources barely moved. That is what a repricing looks like. The premium did not disappear when content got cheap. It relocated. It moved off the raw artifact and onto the things that make an artifact worth surfacing: originality, attribution, a source a system is willing to name.

When the supply of something goes up and its value stays flat, the value was never really in that thing. It was in the scarce input around it. The scarce input here is not words. It is a credible, identifiable point of view.

The reward is structural, not sentimental

It would be easy to read the Graphite numbers as a temporary lag, a nostalgia for human writing that the machines will eventually erase. The platform behavior points the other way.

Following Google's core algorithm updates in late 2025 and March 2026, which explicitly rewarded first-hand experience, verifiable authorship, and original insight, a synthesized industry analysis spanning more than 600,000 pages found a clear split. Sites publishing original data and first-hand research saw search visibility increase by an average of 22%, while mass-produced AI content saw sharp traffic declines (industry analysis of Google core updates, 2025 to 2026).

We treat this as evidence of a durable repricing, not as an SEO headline to chase. The specific update names and percentages will age. The direction will not. Distribution systems, whether search engines or answer engines, have a structural incentive to demote infinitely reproducible content and surface material that is hard to fake: your data, your case, your genuinely contrarian read. Generic content is now a commodity input to those systems. Original, attributable material is the scarce thing they compete to include.

This is the constructive version of the warning from Part 1. Yes, publishing feeds the machines. But the machines increasingly reward exactly the kind of publishing that also builds your authority, and increasingly ignore the kind that does not. The incentives are less opposed than they looked at the start of the series.

The asset the algorithm can't give you

There is one more asset, and it is the one you have to build most deliberately, because no platform will build it for you.

Suppose you do everything right. You publish original research, you rank on the first page, you get cited by name inside an AI answer. You are visible. Is that enough? The data says no.

The 2025 Pew Research Center study we referenced in Part 1 found that only 1% of visits to a search results page with an AI summary resulted in a click through to a cited source (Pew Research Center, 2025). Being cited by the summary is real, and it is better than being absent from it. But as a business asset, a citation someone reads and never clicks is thin. Ninety-nine times out of a hundred, the value of the answer accrues to the platform delivering it, not to the source that made it possible.

That is why the durable asset cannot be search visibility alone. It has to be a direct relationship with an audience, a channel that does not depend on being the clicked link inside someone else's summary. An email list. A community. A body of readers who know your name and come back on purpose, not by accident of routing. The distinction matters more every year: being the answer inside a system you do not own is fragile, and being the reason people leave that system to find you is not.

Publish generously, then, but publish toward a relationship. Every artifact should make it a little more likely that a stranger becomes someone who knows who you are and where to find you directly. The content is the invitation. The relationship is what you are actually building.

Where the moat goes

So we can answer the question in the title. Content stopped being a moat because content became reproducible, and a moat made of a reproducible thing is not a moat. Fine. The value did not evaporate. It moved.

It moved to the judgment, which you demonstrate but never fully hand over. It moved to the name behind the judgment, the recognized source a system is willing to cite and a reader is willing to trust. And it moved to the relationship with the people who come back for both. Publishing generously is not a threat to those three things. It is the only way to build them. The method you give away is the signal that points at the judgment you keep. The visible work is what earns the name. The steady output is what turns readers into an audience you actually own.

Which means silence was never the safe option. Withholding your expertise protects nothing, because the thing worth protecting was never the sentence on the page. It was you, doing the thing, in public, often enough that people learn to trust how you think. Publish anyway. That was always where the moat was going to be.

FAQ

If AI can absorb my published expertise, why should I publish it at all?

Because withholding it protects an asset that has already lost most of its value. The general, reproducible method you might guard is being commoditized in the ambient environment whether you contribute or not. The scarce, durable assets are your demonstrated judgment, your recognized name, and your direct relationship with an audience, and none of those can be built while your work stays invisible. Publishing is how you build all three.

What is the difference between publishing the method and demonstrating judgment?

The method is the explicit, reproducible layer: frameworks, steps, and repeatable approaches. Publish it generously, because guarding it earns nothing once it is already widely available. Judgment is the tacit ability to know which method fits a specific messy case and when to break the rule. Demonstrate it through real, detailed cases rather than generalizing it into a checklist, because a reader can watch you reason without being able to reproduce the reasoning.

Does AI content actually outrank and out-cite human content now?

The evidence points the other way. Graphite's 2025 analysis of roughly 65,000 URLs found that even after AI-generated content reached near parity in volume, 86% of first-page Google results and 82% of sources cited by tools like ChatGPT and Perplexity were still human-written. Distribution and citation are concentrating on recognized, original, human-attributed sources, not diluting toward generic content.

If my content gets cited inside an AI answer, isn't that enough?

Citation is real but weak as a standalone business asset. The 2025 Pew Research Center study found that only 1% of visits to a search page with an AI summary produced a click through to a cited source. Being named inside an answer helps, but the value mostly accrues to the platform. A direct channel to an audience, like an email list or community, is the asset that does not depend on being the clicked link in someone else's summary.

What is the single biggest risk this series is warning against?

Being ignored, not being copied. Most people over-index on the fear of having their expertise absorbed by a model, but the far more common and more damaging failure is that nobody encounters the work at all. Invisible expertise builds no authority, earns no trust, and generates no business. Being copied at least means you were visible and worth copying. Being ignored is much harder to recover from.

Further reading

• Axios, "AI-written web pages haven't overwhelmed human-authored content, study finds" (2025), reporting on Graphite's analysis of ~65,000 URLs (2020–2025) — 86% of Google-ranked articles and 82% of ChatGPT/Perplexity citations were human-written. axios.com
• Graphite, "AI Content In Search & LLMs" — Ongoing tracking of AI-generated vs. human-written content share online. graphite.io
• Industry analysis of Google's March 2026 core update (synthesizing Ahrefs, Semrush, Originality.ai and other data across 600,000+ pages) — Sites publishing original data/first-hand research saw a 22% average visibility increase; mass-produced AI content saw sharp traffic declines. wyomingnews.com
• Pew Research Center, "Google users are less likely to click on links when an AI summary appears in the results" (2025) — Only 1% of visits to a page with an AI summary resulted in a click on a cited source. pewresearch.org