Before LLMs could generate a passable blog post in minutes, I wrote an article for Vimeo about how yoga instructors could make money online.
Someone quickly published a version to compete with it in search. Beyond paraphrasing the copy, they used my name. They also described me as a yoga instructor.
I was never a yoga instructor, but I was slightly annoyed.
The piece had just enough of the original to look like it belonged, but none of the decision-making that made the original worth reading.
AI didn’t invent these habits. It removed the friction that used to keep some of them in check.
Reusing the shape instead of doing the work
Copycat content is nothing new. Even amid zero-click search and AI overviews, the core approach to SEO still goes something like: study competitors, identify patterns, and make sure your content covers them too.
Competitive research is useful because it shows you what already exists—the signals that make a piece look like it belongs. What competitive research can’t tell you is what’s true, useful, or worth saying for your reader now.
The yoga clone had the topic, the search intent, and enough of the original signals to look credible, but it lacked the editorial judgment and positioning to realistically compete.
Today, copying isn’t limited to manual effort or what shows up in a search result. It’s anything you can paste into an LLM interface.
Feed it a writer’s work and ask for their cadence or argument style. Tell it to mimic the contrarian opening, the sharp conceptual label, or the confident close.
The output may not reuse exact sentences, but it can reproduce the observable choices that made the original compelling: the title, the angle, the structure, the voice, and even the confidence.
This isn’t inherently unethical—people have always studied work they admire—but there’s a big difference between taking inspiration from a shape and forcing your content to fit someone else’s context-dependent choices.
An LLM lets us make those editorial decisions before we’ve necessarily done the work of learning when and why to make them.
Reference materials show you the field. They don’t give you proximity to the reader’s problem, original evidence, a real observation, or a point of view you’ve actually developed.
That’s the part worth spending your time on.
Writing for the approval chain
Behind every lifeless asset there’s a marketer trying to survive the approval process.
The content pre-empts every objection, leans lazily on pre-approved messaging, and makes room for every potential stakeholder concern. By the end, everyone inside the company is satisfied, but nobody outside it has much reason to care.
There’s often a point in there somewhere, but it’s surrounded by enough caveats, qualifications, and consensus language that readers won’t bother to find it.
Stakeholder involvement is often essential. An editorial review process can catch unsupported claims, protect accuracy, and keep a company from making promises it can’t keep.
Trouble starts when protecting internal alignment becomes safer and more important than saying something that matters. Or when “on-brand” becomes shorthand for “nothing we produce can risk making anyone uncomfortable.”
A marketer who ships something sharp that a stakeholder vetoes has a bad week. A marketer who ships something bland gets nothing—no risk, no reward.
That’s why optimizing for the approval chain becomes a habit. Actually putting readers first starts to look like a radical act.
It doesn’t help that AI is exceptionally good at serving this incentive structure.
Give it enough pre-approved messaging—language the company already likes—and instructions about what not to say, and you can generate an endless supply of perfectly safe, milquetoast content.
That’s how you get blog posts that sound like a Confluence page no one’s glimpsed in three years. Internal documentation wearing an article’s clothes.
The draft may start with something to say. Then every objection or concern pulls it a little closer to the version nobody can object to.
Editorial review should pressure-test the thesis, not sand it down until all the tension is gone.
Limiting ideation to the public web
I once found a statistic in an RSS feed that seemed like a viable seed for a thought leadership post.
I needed qualitative data to substantiate it, so I checked sales calls and combed through internal docs, attempting to build a piece from a number that sounded like it should matter.
The pitch got rejected for good reason: nobody in the field actually believed it.
I had found something “real,” but it wasn’t meaningful enough to construct a useful argument.
That’s the risk when brainstorming starts with search. You latch onto what’s easily surfaced, then try to build insight around it.
The resulting work can read well—and even get some things right—and still have a weak relationship to what people in the field recognize, worry about, or need to decide.
But that’s increasingly how content gets made: scan the SERPs or your competitors’ blogs, find the gaps, then add some vague layer of brand personalization.
AI makes that process almost frictionless. The content can look legitimate before it has a reason to exist.
Starting with a search bar makes sense until every competitor, writer, and large language model starts from the same material.
Public information is a common input, not a competitive advantage. You can’t reverse-engineer the invisible work that happens before a finished artifact exists.
You have to mine for insights within the organization. Talk to customers. Listen to calls. Find evidence. Sit with an uncertain idea long enough to discover that it might be wrong—or that there’s something more interesting underneath it.
You have to get close enough to the underlying problem to have something real to say.
Repeating the same conviction without giving it a new job
Every company has a few things it needs the market to understand. But when one of those core convictions becomes the entire content calendar, is the organization delivering useful messages, or just repeating the one?
A company may believe that AI needs more than a powerful model—it needs context, controls, workflows, and human judgment around consequential decisions. Fine.
That belief could become a useful article for a technical buyer evaluating governance, an executive assessing implementation risk, or an operator deciding where human review belongs.
But if every version reaches the same conclusion for an undefined audience—models need context, controls, and people—the company isn’t building a body of thought for readers. It’s reproducing a sales deck.
The language often gives this away. Faster outcomes. Increased profits. More productivity.
These results may be real and achievable, but without a reader attached, they sound like executive dashboard copy.
Faster for whom? What stops happening? What becomes possible?
Instead of asking what the audience needs to understand next, the team starts asking how else it can say what the company’s already decided to say.
- Same claim: a model needs systems, context, controls, and people around it
- Same audience: everyone, or an undefined enterprise buyer
- Same awareness: broadly explains why the category matters
- Same evidence: internal assertions about capability
- Same payoff: “Therefore, our approach is sophisticated.”
This is not a content strategy. This is a product thesis repeatedly repackaged for a calendar.
Treating output as evidence of strategy
A full content calendar proves that work happened.
That’s essential to an organization. When a marketing team can point to the number of articles, emails, videos, sales assets, or social posts shipped, it demonstrates its value. Or does it?
As marketers, we know that much of what we do—from deep work to operations—doesn’t produce convenient artifacts.
You can spend hours talking to customers or SMEs and realize the original premise was wrong. You can decide not to publish because the evidence isn’t there.
You can spend a week figuring out what one audience actually needs to understand before deciding what to make.
The work is real, but none of it looks like output. And when output is the easiest part of the work to measure, it’s easy to confuse it with the work itself.
Low output? Not pulling your weight. Consistent posting schedule? Productive. Even when nobody can say who each piece was for, or why that reader should care.
So marketers maintain a flurry of activities. Nobody believes more content is inherently better, but we all agree it looks like work.
But what happens when production stops being the bottleneck?
When a model can generate three plausible articles before lunch, producing more becomes an easy substitute for figuring out what deserves to exist.
Marketing became too easy
Better prompting can make a draft more fluent, a style guide can make it sound more like a brand, and a workflow can make the process more reliable.
None of those things can decide why a piece should exist. That decision comes first.
Who is this for? What can you show them that they couldn’t get from the same public sources, familiar templates, and increasingly capable models everyone else is using?
You still need to recognize when a premise is wrong, when an argument isn’t finished, when the evidence doesn’t support the claim, or when a perfectly competent piece doesn’t give the reader a reason to care.
These are judgments, not production problems. And you need to be willing to make them before you start asking how efficiently the thing can be produced.
AI can help execute an answer. It can’t responsibly invent one. Those decisions still need to happen, and they should be visible in the work.
