I Don't Use AI to Write This

A prospect asked if I use AI to write my blog posts. I said no. There was a silence on the call that lasted about three seconds, which in sales call terms is roughly the length of the Hundred Years' War. \"Why not?\" she asked. And I realized I had never actually articulated the answer.

\n\n

I'd felt the answer. I'd felt it every time I read an AI-generated blog post and experienced that peculiar sensation of consuming something that technically qualifies as content the same way a gas station sandwich technically qualifies as food - all the components are present, the bread is bread, the meat is meat-adjacent, but the thing itself is missing something that you can't name but can absolutely taste. I'd felt it every time a competitor's website got a sudden injection of thirty new pages in a month and I knew, without checking, that every single one of them read like it had been written by a very polite entity that had never had a bad day, never lost a client, never sat in a meeting watching a decision-maker ignore data that took three weeks to compile. I'd felt it. But I'd never said it out loud, and saying things out loud is, as my therapist has pointed out on multiple occasions, a different skill than feeling them.

\n\n

So I thought about it. On the call, in those three seconds, I thought about it with the speed of someone who suddenly realizes they're about to have to defend a position they've held so instinctively that they never bothered to build the defense. And what came out of my mouth was something like: \"Because writing is how I think, and if I outsource the writing, I outsource the thinking, and the thinking is the product.\" Which is true. But it's not the whole answer. The whole answer is longer, and more complicated, and involves some things I'm not sure I believe all the way, and some things I'm very sure about, and the line between those two categories moves depending on the day.

\n\n

This is the whole answer. Or as close to it as I can get.

\n\n

The Convergence Problem

\n\n

Every large language model is trained on essentially the same corpus. GPT-4, Claude, Gemini, Llama - the details differ but the mass of it is the same: the internet, Wikipedia, books, academic papers, code repositories, the collective output of several billion people writing things down over several decades. When you ask any of these models to write a blog post about, say, technical SEO for B2B SaaS companies, they all produce something from the same substrate. The word choices differ. The structure varies slightly. But the substance converges, because the substance is drawn from the same pool.

\n\n

This is not a flaw. It's a feature of the technology. LLMs produce the statistically likely next token, which means they produce the statistically likely next idea, which means they produce the consensus. They produce what is, on average, correct. What is, on average, well-phrased. What is, on average, the kind of thing that has been said before about this topic by people who wrote about it in the training data. They produce the mean. And the mean is, by definition, average.

\n\n

Now think about what this means for content marketing. The whole point of content marketing - the entire strategic premise - is differentiation. You produce content to distinguish yourself from competitors. To demonstrate unique expertise. To say something that nobody else is saying, or to say something everyone is saying but in a way that is recognizably, irreducibly yours. The value of content marketing is the extent to which your content is not like everyone else's content.

\n\n

AI-generated content is converging. Not slowly. Rapidly. Because every company is using the same models (or models trained on the same data) with the same prompts (or prompts that produce the same outputs), and the result is a vast, expanding ocean of content that sounds like it was written by the same competent, characterless, unfailingly polite entity. An entity that has read everything and experienced nothing. An entity that can tell you the best practices for E-E-A-T but has never sat across from a client who just got hit by a core update and doesn't know if they should panic or wait. An entity that can produce a perfectly structured pillar page about link building strategy but has never built a link through a relationship that started as an argument at a conference bar at 11 PM in Salt Lake City (which is how I built one of the best links my site has ever received, and I'd tell you the story except it involves a disagreement about baseball statistics that somehow turned into a guest post opportunity, and that sentence is the kind of thing AI doesn't generate because AI doesn't go to conference bars and argue about baseball).

\n\n

When everyone's content converges, nobody's content differentiates. And when nobody's content differentiates, content marketing stops working. Not stops working a little. Stops working at the structural level, at the level of the strategy itself, because the strategy depends on differentiation and the tool being used to execute it is an anti-differentiation machine.

\n\n

I watch this happening in real time. I look at competitor blogs in the SEO consulting space and I can see the shift. There was a time - not long ago, maybe 2022 - when each competitor's blog had a voice. Some were technical and dry. Some were casual and full of screenshots. Some were academic and citation-heavy. Some were personal and anecdotal. You could read a paragraph and know whose site you were on. That is increasingly not the case. The blogs are merging into a single register: clear, professional, well-structured, thoroughly researched, and completely interchangeable. They are better-written than they used to be, in the narrow sense that the grammar is cleaner and the structure is more logical. And they are worse-written than they used to be, in the deep sense that they say nothing that couldn't be said by anyone else, because they weren't said by anyone at all.

\n\n

What Google Knows That You're Hoping It Doesn't

\n\n

I want to set aside the philosophical argument for a moment and talk about the practical one, because some of you are reading this thinking \"that's nice, Amos, but I need traffic, not a writing manifesto,\" and you're not wrong to think that, and the practical argument is, if anything, stronger than the philosophical one.

\n\n

Google's helpful content system - which was first introduced in August 2022 and has been updated several times since, most recently in a way that folded it into the core ranking system itself - is explicitly designed to identify and demote content that exists primarily for search engine purposes rather than to genuinely help people. The system looks for signals of human expertise, experience, and firsthand knowledge. It looks for what Google internally calls \"information gain\" - does this page contain something you can't get from other pages on the same topic? It looks for author authority, for specificity, for the kind of granular detail that comes from doing a thing rather than reading about a thing.

\n\n

AI-generated content, by its nature, fails most of these signals. Not because it's bad. Because it's derived. It is, by definition, a synthesis of existing information. It cannot contain firsthand experience because it has no firsthand experience. It cannot contain original data because it generates no data. It cannot contain the kind of specificity that comes from the author having personally done the thing they're writing about, because the author has never done anything. Every claim it makes is a statistical average of claims other people have made, which means its information gain is approximately zero, which means it is precisely the kind of content the helpful content system was built to detect.

\n\n

Google has been cagey about whether they can detect AI-generated content at the textual level - whether there's a classifier running on the content itself that flags it as machine-generated. I suspect they can (the technology exists, watermarking aside, and Google has the talent and the data to build a very good one). But honestly, it doesn't matter whether they can detect it at the textual level, because they can detect it at the signal level. Content that provides no information gain. Content that lacks firsthand experience markers. Content that covers a topic comprehensively without adding anything to the topic. Content that reads like a very good summary of everything that already exists on the first page of Google for this query. All of these are signals that the helpful content system already evaluates, and all of them correlate nearly perfectly with AI-generated content, whether or not Google knows (or cares) about the generation method.

\n\n

The practical argument is this: AI-generated content is optimized for the wrong signals. It is optimized for coverage, for comprehensiveness, for topical completeness. Those were valuable signals in 2018. They are not the signals that matter in 2026. The signals that matter now are experience, expertise, originality, and perspective - and those are the four things that AI is structurally incapable of providing, because they all require having been a specific person in a specific situation, which is not something a language model can fake, no matter how sophisticated the prompt engineering gets.

\n\n

The Counterfeit Painting Problem

\n\n

I keep coming back to this analogy, and I apologize for it because it's not original (nothing I say is original - the original people are the ones making things, and I'm a consultant, which means I comment on the things other people make, which is a humbling realization that I try not to think about more than three or four times a day).

\n\n

AI-written content is like a very convincing counterfeit painting. It has all the right elements. The brushstrokes look correct. The composition follows the rules. The color palette is appropriate for the period. If you showed it to most people, they would say it's a real painting, and they might even say it's a good one. But it was made by someone - something - that has never felt anything, and if you look at it long enough, you can tell.

\n\n

You can tell because the painting is too correct. A real artist makes choices that don't follow the rules, not because they don't know the rules but because they know something the rules don't capture. A real artist puts a shadow where there shouldn't be one because the shadow makes you feel something. A real artist leaves a line unfinished because the unfinished line creates tension. A real artist makes a mark that is, technically, a mistake, and the mistake is the thing that makes the painting alive.

\n\n

AI doesn't make those mistakes. And the absence of meaningful mistakes is itself a tell. Not to everyone. Most people scroll through AI-generated content the way most people walk through a museum gift shop - they see the reproduction and it's fine, it's nice, it serves its purpose. But the people who matter - the readers who become clients, the industry peers who become referral sources, the journalists who become amplifiers - those people can tell. They have been reading enough content for long enough that they have developed an instinct for the difference between something written and something generated, the same way an art dealer has an instinct for the difference between a painting and a print, even when the print is very good.

\n\n

And here's the thing about a counterfeit: its value depends entirely on nobody knowing it's a counterfeit. The moment someone suspects, the value doesn't decrease. It collapses. A counterfeit that's been identified is worth less than nothing - it's evidence of dishonesty. And we are approaching the moment where AI-generated content, on a professional services website, will be received the same way. Not \"oh, they used AI to write this, that's efficient.\" More like \"oh, they used AI to write this, so they don't actually have expertise, and everything on this site is potentially a synthesis of other people's ideas presented as their own, and I can't trust any of it.\" The liability is not that the content is bad. The liability is that the content is untrustworthy, and trust, once lost, does not come back in the same shape.

\n\n

The Business Argument That Nobody Wants to Hear

\n\n

I consult for companies that range from startups to publicly traded enterprises, and across that entire range, there is one thing they all want from their content marketing: a competitive advantage. They want their content to do something that their competitors' content doesn't do. They want their blog to be the one people bookmark. They want their newsletter to be the one people actually read. They want their thought leadership to be the thing that makes a prospect think \"these people know something the other vendors don't.\"

\n\n

You cannot build a competitive advantage on a tool that is available to all of your competitors. This is so obvious that it feels stupid to type it, and yet entire content strategies are being built on exactly this premise. \"We'll use AI to produce more content, faster, cheaper.\" Great. So will your competitor. And their competitor. And the company that launched last Tuesday. You are all using the same tool to produce content from the same training data, which means you are all producing the same content at the same speed, which means you have optimized your way into a perfectly level playing field, which is the opposite of a competitive advantage. It is a competitive wash.

\n\n

The competitive advantage in content has always been the same thing: saying something only you can say. Having an insight that comes from your specific experience. Making an argument that requires your specific expertise. Telling a story that only you were there for. These are the things that make someone read your post instead of your competitor's post, link to your post instead of your competitor's post, remember your name instead of your competitor's name. And these are precisely the things that AI cannot do, because AI does not have specific experience, does not have specific expertise, was not there for any specific story.

\n\n

I realize this sounds like I'm romanticizing writing. I'm not. I'm making a business argument. The business argument is: if your content is interchangeable with your competitor's content, it has no strategic value. It may have tactical value - it fills pages, it targets keywords, it gives you something to post on LinkedIn every Tuesday. But strategic value requires differentiation, and AI is a differentiation-erasing technology. The more you use it, the more you sound like everyone else who uses it. The more you sound like everyone else, the less reason anyone has to choose you.

\n\n

Writing as Thinking

\n\n

Now I'm going to say the thing I said on that call, the thing that fell out of my mouth before I'd thought it through, and I'm going to try to say it better because I've had time to think about it and because writing this (by hand, word by word, the slow way) is itself the process of thinking it through, which is sort of the point.

\n\n

Writing is not the output of thinking. Writing is the thinking. When I sit down to write about, say, why AI SEO isn't a real discipline, I don't start with a finished argument that I then transcribe into prose. I start with a hunch, a feeling, a half-formed irritation, and the act of writing is what turns that hunch into an argument. The sentences do the work. The process of finding the right word forces me to clarify what I actually mean. The process of structuring a paragraph forces me to understand the logical sequence of my own ideas. The process of writing a transition between two sections forces me to discover the connection between two thoughts that I didn't know were connected until I tried to connect them.

\n\n

This is not a mystical claim. It is a well-documented cognitive phenomenon. Writing is a form of externalized thinking. The physical (or digital) act of putting words in sequence engages a different cognitive process than thinking silently, and the output of that process - the ideas, the connections, the arguments - is different from what you would produce if you just sat and thought. Every serious writer knows this. You don't know what you think until you write it. The writing is the discovery.

\n\n

When I use AI to write, I skip the discovery. I get the output without the process, and the process was where the value was. I get a blog post that sounds like my ideas but isn't - it's a statistical approximation of my ideas based on whatever the model absorbed about the topic, and the approximation is close enough to be satisfying and far enough to be wrong in ways I might not catch, because the wrongness is subtle, it's not factual errors (those are easy to spot), it's the absence of the connections I would have made if I'd done the thinking myself, the insights I would have discovered in the act of writing that I'll never discover because I outsourced the act to a machine.

\n\n

I am not productive when I skip the writing. I am less productive. Because the writing is not a cost. It is an investment. Every piece I write teaches me something I didn't know I didn't know. Every argument I construct reveals a gap in my thinking that I can fill. Every analogy I reach for shows me a connection between two ideas that becomes part of my permanent mental model. Outsourcing this to AI is not \"freeing up my time for higher-value work.\" It is eliminating the highest-value work I do and replacing it with time I'll probably spend on email.

\n\n

The Objections, Because I Know What You're Thinking

\n\n

You're thinking: \"This is a luxury. You can afford to hand-write everything because you're a solo consultant with a small blog. If I'm running a content operation that needs to produce fifty pages a month, I can't write them all by hand.\"

\n\n

You're right. Sort of. If your content strategy requires fifty pages a month, then yes, you need tools to scale, and AI is a powerful scaling tool, and I'm not here to tell you that you should hire fifty writers instead. But I want to push back on the premise. Why do you need fifty pages a month? Because your SEO strategy says so? Because \"more content = more traffic\" was true in 2019 and nobody questioned whether it's still true in 2026? Because your agency (the one with the AI service tier) told you that volume is the game?

\n\n

Volume is not the game. Not anymore. Google's helpful content system is explicitly designed to reward quality over quantity. Sites that publish massive amounts of thin or derivative content are being demoted. Sites that publish less frequently but with more depth, more originality, more genuine expertise are being elevated. The game has changed, and the change favors exactly the approach I'm describing: fewer pieces, more substance, written by people who know what they're talking about.

\n\n

If your content strategy requires fifty AI-generated pages a month, you don't have a content strategy. You have a keyword-filling strategy, and the era in which that worked is ending, not slowly, but with the kind of algorithmic violence that Google periodically visits upon strategies that game the system rather than serve the user. I have seen it happen before. I have seen it happen to clients who came to me after it happened. It is not fun. It is the kind of meeting where nobody makes eye contact and everyone's water glass stays full because nobody can swallow.

\n\n

You might also be thinking: \"OK, but AI as a first draft. AI as a starting point. You edit it, add your voice, add your experience, and it's still faster than writing from scratch.\"

\n\n

I've tried this. Multiple times. And here's what I found: editing AI output into something that sounds like me takes longer than writing from scratch. Because the AI has already made structural choices - about what to cover, in what order, with what emphasis - and those choices are wrong, not factually wrong but strategically wrong, and unwinding them requires more effort than making them fresh. It's like trying to renovate a house by tearing down walls that shouldn't have been built in the first place. At some point you're not renovating. You're demolishing and rebuilding, and you'd have been better off starting with an empty lot.

\n\n

This is not true for everyone. If your writing is functional - if the purpose is to convey information clearly and you don't care about voice or perspective - then AI as a first draft is genuinely useful. But for content that is supposed to represent your expertise, your personality, your point of view? The first draft is where those things are born. The first draft is where the voice emerges. If you skip it, you spend the rest of the process trying to inject life into something that was born dead, and it never quite works, and you end up with content that is 80% as good as what you'd have written yourself, which sounds efficient until you realize that the 20% you lost is the only 20% that mattered.

\n\n

What I'm Actually Saying

\n\n

I am not saying AI is bad. AI is a tool, and like every tool, it is good for some things and bad for others. I use AI in my work - for data analysis, for code, for research synthesis, for generating ideas I can push against, for doing things that are genuinely mechanical and benefit from automation. I'm not a Luddite. I'm not smashing the loom. I'm saying the loom is bad at making paintings.

\n\n

I am not saying everyone should hand-write everything. For some content - product descriptions, support documentation, internal communications, routine reporting - AI is the right tool. It produces acceptable output at extraordinary speed, and \"acceptable\" is all those contexts require. I am talking specifically about content that exists to represent your expertise and differentiate your brand, the content that is supposed to make people choose you. That content needs to be the thing that only you can produce, and AI cannot produce the thing that only you can produce, because AI is not you.

\n\n

I am saying that the rush to automate content creation is, in many cases, a rush to eliminate the only thing that was making the content valuable. It's like automating the flavor out of food because the automation makes the food cheaper. The food is still food. But nobody's driving across town for it anymore.

\n\n

Every competitor in my space is publishing more than I am. Substantially more. Some of them are publishing five, ten, twenty times more. Their sites are filling up with well-structured, well-optimized, perfectly adequate content that covers every topic in our industry with the thoroughness of a textbook and the personality of a hallway. They are winning the volume game. They are losing the voice game. And in a market where every buyer has access to the same information - where the textbook is free and the hallway is infinite - voice is the only thing left that creates preference.

\n\n

I write this blog by hand. Every word. It takes me between six and twelve hours per piece, depending on the topic and depending on how many times I delete the first three paragraphs and start over (the answer is usually twice, sometimes three times, once memorably seven times for a piece about attribution models that I still don't think I got right). It's slow. It's inefficient. It cannot scale. And it is the single best business development investment I make, because every piece sounds like me, and sounding like me is my competitive advantage, and my competitive advantage is not something I'm willing to hand to a machine, even a very good machine, even a machine that could probably write something 80% as good in 2% of the time.

\n\n

The 80% is not the same as the 100%. And the 20% gap is where the trust lives.

\n\n

She asked me why not. Here's why not. Because the last defensible advantage in a world of infinite content is being the person who actually wrote it, who actually thought it, who actually sat in the chair and did the work and came out the other side knowing something they didn't know when they sat down. That's not romanticism. That's strategy.

\n\n

The counterfeit hangs in the gallery. It looks right. But the artist isn't in it, and if you look long enough, you can feel the absence. You can always feel the absence.