Authorship in the age of AI: some thoroughly human reflections [ESSAY]

Two comments pushed me to write this article – both posted on the blog in recent months. One from the archives, one from yesterday – possibly bot-generated. Both said roughly the same thing:

“Content written by AI, and poorly at that.”

The comment itself isn’t the problem for me. It’s a symptom of a larger shift I’m watching: we’re starting to judge authorship by the surface of a text, not by responsibility for the thought behind it.

Out of curiosity, I ran the article in question through a few AI detectors. The result: 95% likely human, 5% AI-generated. Can you trust that? No – detectors are material for a separate post.

The analysis did show, though, exactly what I’ve been building deliberately for years: varied sentence length, rhetorical questions, a personal voice, short sentences closing out sections. On top of that, an orderly structure I’ve loved ever since the days when I sat and hand-coded HTML. And yes – unluckily in the context of that comment – AI “likes” this too.

But that isn’t the end of the discussion. Because the question of authorship in the age of AI is hard to settle with a binary answer. And I don’t have one either. I’ve been looking for it for a good two years.

Why does the question of authorship in the age of AI move me so much?

You could say I have two wolves inside me. One is analytical, SEO-minded, for seventeen years immersed in algorithms, code, documentation. That wolf is excited, because it sees what LLMs are changing in search, in content production, in the way machines process knowledge about people and brands. AI is a tool that opens up possibilities I couldn’t even have dreamed of five years ago.

The other wolf is more… authorial. Creative. And it’s the one that hasn’t let me rest for two years. I’ve written six books, and four of them came before the era of LLMs. Those were texts written by hand, partly on paper, with no language model in the background. And this wolf is the one pushing me toward the question:

Where does authorship end and AI ghostwriting begin?

My last two books came about differently. With AI in the research and proofreading process. The blog you’re reading now is made this way too. So I have a comparison that a good portion of the people commenting on the topic don’t have. I know both modes from my own experience. And that tension between my excitement and my doubt is exactly what I want to write about today.

What does expert writing with AI look like from the inside?

Most people picture writing with AI like this: prompt, finished text, publish. Three clicks and the article’s done, right? But when expert texts are involved… it doesn’t look like that. At least not for me. And not for any expert I know and respect.

I keep a record of the time I spend preparing books and materials.

How much shorter has that time become? For books, by 20–30%. For articles, up to 40%.

(I’m not counting here the time spent getting the LLM to know my voice, my approach, and my views on various topics – “feeding” the model took a while too).

That’s a sizable improvement, but is it what you expect when you think or write “this article was definitely made with AI“?

What hasn’t changed?

In the age of AI (btw, it’s scary these days to use the phrase “in the age of,” just like short sentences or dashes in a text), I’m still the one who decides what I write about.

I’m the one who makes the argument.

I’m the one who chooses which point to make and which to reject.

I’m the one who thinks about the structure.

I’m the one who says “that’s not me” and rewrites an edited passage, because it sounds good but… it doesn’t sound like me.

I’m the one who reads the finished text a third, fourth, tenth time (my articles go through about 20 revision versions in WordPress before publishing, and often a few more after).

I’m the one who fixes the sentences that “don’t sound right” to me, that have no rhythm.

And I’m the one who takes responsibility for what I publish under my name.

What has changed?

Research that once took me a dozen-plus hours of digging through PDFs and documentation now takes an hour, because I know what I’m looking for. It’s different when I know a topic less deeply, as with Common Crawl (so far the article I’ve spent the longest on for this blog, consulting specialist friends along the way).

The structure of an article, which once filled numerous A4 sheets of scribbles, now takes shape in a “dialogue” with an LLM. And then the real work begins: I write, cut, rewrite, change direction, add what’s missing, remove what sounds generic.

Every article that lands on this site is at least a few hours of my work. Not a few hours of generating content. A few hours of writing, reading, fixing, verifying sources, making editorial decisions, and writing passages again from scratch, because the ones from the model don’t capture what I want to say.

Could I get the process done in less time? Definitely. Some of the edits are pure perfectionism. But I publish under my name, and I have an internal “quality barometer.”

Does this mean I write “on my own“? Not in the traditional sense of the word.

Does it mean the LLM writes for me? Also no.

I’m somewhere in between. And I suspect there are more and more people in that place – people trying to work honestly and efficiently at the same time. Only few talk about it, because… it’s scary to stick your neck out.

The starting threshold that everyone who has it knows

There’s one more thing I want to talk about gently, but honestly. An LLM affects decision paralysis – in ADHD (and not only). It helps with the hardest part: starting.

For people who don’t have ADHD, this sounds like a trendy label. For those who know the subject, it’s everyday life: you know exactly what you want to write, you have the idea, you have the structure in your head, and yet… you don’t move. Or you move with a big delay and under pressure. Not because you can’t, or have nothing to say. The starting threshold is simply high for you. And it shouldn’t be treated as an excuse – more as context for the role of a tool when writing is done by people who are often very creative.

A tool that lowers the barrier to entry makes it easier to start the work. Many of the texts that now exist on this blog wouldn’t exist without me. But they probably also wouldn’t exist without a tool that helps me, every day, get going on a topic I wouldn’t otherwise have had the mental resources for.

Where’s the line? The questions I don’t have ready answers to

Every one of my books carries the name of a proofreader and a managing editor. Before publication, the text passes through several pairs of hands before it goes to print. And yet no one has ever questioned my authorship of those books. No one wrote “content written by the editor, but poorly.”

Now the same question, with one element swapped: if a language model does what a human research assistant used to do – searches sources, pulls quotes, prepares a draft of a section – then… why does authorship suddenly become doubtful?

And this isn’t purely a rhetorical question! I won’t lie if I say it’s with me almost every day. I have more of them:

I’m not claiming I know the answer to all these questions. I won’t attempt legal analysis – that’s not my role and not my expertise. What’s more, the law isn’t keeping pace with technological development; we’ll see how Article 50 of the AI Act changes things. In theory, then, I could put my name to AI slop (as of the moment of writing this post). But ethically, in terms of quality, in the sense of trust halo… no longer.

I do believe, though, that the very act of posing these questions changes the way we think about authorship. And it leads us to some interesting reflections. Because the old division of “wrote it themselves” vs “someone else wrote it for them” is too simple to describe what’s happening now.

Research – a separate layer of working with text

In the discussion about “writing with AI,” one issue tends to get skipped: research. When someone says “text written by AI,” they picture:

prompt → finished article → publish

And they skip the hours of work that separate the question from the answer.

I write about RAG and generative-search mechanisms. I write about Google’s patents. I write about personal-brand entities and composing text for citation by LLMs. And none of these topics jumps out of a prompt like a rabbit out of a hat. In theory it could; in practice, without research it wouldn’t meet my quality bar and wouldn’t avoid the model’s hallucinations.

Behind every article and every conclusion I draw is a decision: check this patent, not that one. Check what the schema.org documentation says in the section on sameAs. Let me set this SQRG quote against what Google published in May. Now compare it with what Liz Reid said at I/O. No, I disagree with that one; from experience, I think it’s this way and that.

This is expert work. I have to know what to look for, where to look, and what not to look for, because it’s a dead end. An AI tool shortens the time it takes to reach a source. But it doesn’t replace a dozen-plus years of reading documentation, analyzing algorithms, and learning what’s a signal and what’s… nonsense.

Internet cafés and old modems – or how we pretend there’s no revolution

I remember the days of internet cafés. Yes, I went to them with friends from school, back when home internet was still a luxury. I’ll draw a veil of silence over the Interia chat rooms I visited back then (Interia being one of Poland’s big web portals of the era).

I remember the sound of an old modem – it’s etched into my memory, tied to the excitement that I’d soon have internet access and fire up Gadu-Gadu (Poland’s instant messenger of the day, our answer to MSN Messenger). If you’ve gone nostalgic, then please, have a listen.

I remember people who said the internet was a fad and that “real work” happens in the physical world. That virtual reality would never replace meeting face to face. Here I was, in theory, still a bit too young to have my own firm opinion on the matter. Though, knowing me, I already had one – I just don’t remember it.

Some of those people were right about the details. But it turned out they were wrong about the direction.

Just as you couldn’t pretend the internet didn’t exist, you can’t pretend AI doesn’t exist. You can try, of course, but in a few years the world will verify it, and the discussion will probably look completely different. I don’t know how – I can only guess.

You can ignore AI from the position of “real experts write on their own,” which sounds, by default, noble.

At least until you check how much of what you read every day passed through some language model at some stage of production. Or until you start wondering how much of the text you’ve written is inspiration you draw, even unconsciously, from other sources (people, texts you’ve read – this isn’t even about LLMs anymore).

No one questions an SEO audit done on the basis of data from Screaming Frog, Ahrefs, Sitebulb. No one asks whether a visibility analysis is “yours” when Search Console or Senuto collected the data for you. Because in those contexts we intuitively understand that a tool is a tool, and the value lies in interpreting the results, not in collecting the data.

Why should writing be any different?

But here too there’s a second “but”: ignorance works both ways. Mindlessly pasting whatever the model generated isn’t writing.

Prompt → text → publish with not a single change – that isn’t expert work with a tool. It’s outsourcing thinking to a machine that… doesn’t think, but predicts probability. And that’s how thousands of generic AI slop pieces come into being.

The solution doesn’t lie in the extreme. It doesn’t lie in the polarization between “using AI is cheating” and “AI writes for me and I just publish.” It lies in the middle: in the conscious use of a tool by someone who knows what they’re looking for, can judge what they get, and can say “that’s not me.” Just as in the ’90s the solution lay neither in “the internet is nonsense” nor in “I dump everything online without a second thought.”

Let’s not pretend AI doesn’t exist. But let’s not pretend it replaces thinking either…

I don’t have an answer for you. I do have questions.

I’ve written six books. Four without AI, two with AI in the analysis and editing process. I’ve run blogs for years. I run an agency. I have seventeen years of practice in an industry that changes faster than I can describe the previous change.

And much as I’d like to, I don’t have a clean answer to the question “is writing with AI honest?”.

What I do have is practice and a comparison. I have questions I ask myself (and now publicly), because I think it’s honest toward readers. Toward the people who read me, cite me, learn from my texts.

I write with AI. I also write without AI – sometimes I have inspiration and just sit down and write. But more often I write efficiently, with a tool that shortens my path from thought to text. It doesn’t invent things for me, but it does mean that things that would otherwise stay only in my head (because a day has twenty-four hours, and I don’t have a research assistant) make it here, onto the site.

People who have clean answers about authorship in the AI era often aren’t deep enough in the subject. Those who say “it’s cheating” often don’t have a comparison between working without AI and with AI in the creative process. Those who say “it doesn’t matter” often don’t ask themselves the questions they should.

I’m somewhere in between myself. And I think that’s a place that allows for dialogue.

You don’t have to have all the answers. But the very fact that you’re asking the questions sets you apart, in my view, from a lot of people on both sides of this dispute.

I’ll leave you with that thought, as I rush off to prepare two training sessions and a new company-website project, and get ready for the launch of my book Marka osobista w czasach AI i generatywnego wyszukiwania (currently available in Polish). And I’ll be disappearing from the blog for 2–3 weeks, because AI hasn’t yet made the day magically longer.

P.S. This text was made with Claude’s involvement. In building the structure of my idea, in editing typos and stylistic errors, in styling the boxes. Unusually, it didn’t take part in the research – here it simply wasn’t needed.