I remember exactly what my father's desk looked like.
Not that he had a proper desk — it was a corner of the sitting room, a wooden table with a drawer that never closed completely, where he kept bills, a biro without its cap, and a notebook filled with phone numbers written in a hand I now recognise in my own handwriting when I am in a hurry. There was nothing digital there. Everything was paper, ink, physical trace. If someone wanted to take something, they had to enter the house, open the drawer, touch the paper. Objects had a material resistance.
Now I sit at my own desk — which is a laptop on a table, beside a cup of coffee I left there the previous evening — and I try to come to terms with the fact that everything I have written over the past eight years exists in a format that a machine can hoover up in a matter of seconds, without opening any drawer, without touching anything, without leaving any trace that it passed through.
And the first question that comes to me is not a technical one. It is an older question, a more personal one: what does it mean to defend your work when your work no longer has any physical weight?
Before Any Strategy, an Acknowledgement
I am not a technical person by training. I am someone who came to understand technology through necessity — through years of managing websites for town halls, hospitals, and schools in Botoșani, through configuration errors discovered at midnight, through documentation read with a coffee in hand and a low-grade frustration that things were not working as they should.
And it is precisely for that reason that I believe I can speak about how AI robots read your site from a vantage point that many technical analysts miss: the place where technology meets the question of why it matters to me, as someone who writes.
Because yes, there is a technical dimension to everything that follows. But beneath all of it there is a simpler and, at the same time, harder question: you, as a creator — what relationship do you want to have with the machines that read your content without asking permission?
I arrived at two possible answers. They are not opposites. They are complementary, like two ways of being, simultaneously, both resistant and shrewd.
First Perspective — The Resister: Refusal as an Act of Dignity
My son is seventeen, and one evening a few months ago I found him searching for something on his phone with the kind of concentration I recognise — it is the concentration I have when I am looking for something that bothers me because I do not know it. I asked what he was searching for. He told me he had asked ChatGPT about a topic from school and had received a complete, well-structured answer, with examples, with nuance, without needing to open a single website, a single article, a single text written by someone who had spent hours or days thinking about that subject.
I looked at him for a moment. Not with judgement — with a curiosity mixed with something I could not name then, but which I now recognise as the beginning of understanding. What did the person feel who had written the articles from which ChatGPT had constructed that answer? Did they receive anything in return? Did they even know they had been cited?
The answer, in the vast majority of cases, is no. They received nothing. They knew nothing. Their work had entered a machine that transformed it into a reply and eliminated any need for anyone to ever reach them.
And I have written enough over the years to understand that this is not an abstract problem. It is personal. It is concrete. It is about hours spent at a desk in the dark while my son was asleep and I was rephrasing the same paragraph for the fourth time, because the first version was correct but not sincere, and the second was sincere but not clear, and the third was clear but had lost something of what I truly wanted to say.
Those hours have a value. And refusing to let them be consumed without consent is not the paranoia of a defensive creator. It is a dignified position.
What Refusal Means in Practice
The law exists and gives you the right to say no. The European Directive of 2019 on copyright — which few people know about and even fewer make use of — explicitly recognises that any creator may refuse to allow their work to be used to train or feed artificial intelligence systems. The refusal must be expressed technically, not merely stated verbally. That is to say, there must be, somewhere in the structure of your site, a clear instruction that says: you, machine, are not welcome here.
This instruction exists in the form of a file that any website can have and that well-behaved robots — the ones that observe the rules — read before entering. It is like a sign on a gate. It does not stop everyone — there are robots that ignore signs, just as there are people who walk into a yard even when it says "private property" — but it creates a documented record of your refusal, evidence that you did not stay silent.
And silence, in this context, is interpreted as consent. If you say nothing, if you put up no sign, if you configure no instruction — you are deemed to agree. Not because you have signed anything. But because you have not explicitly refused.
This strikes me as one of the most perverse conventions of the digital world: implied consent through the absence of refusal. My father would never have understood it. I barely came to understand it myself, after reading enough.
Silent Refusal Versus Vocal Refusal
But there is a nuance that took me longer to grasp, and which I think is worth stating plainly.
You can refuse in silence — configure the technical instructions, tell no one, and carry on publishing as though nothing had changed. This protects you, as an individual. It leaves a trace. It documents the fact that you did not consent.
Or you can refuse with a voice — do the same thing technically, but declare it publicly, argue it, say why, say which law you are relying on, and invite others to do the same.
The difference between the two is not technical. It is a matter of scale. A single creator who refuses in silence is a statistic that no AI platform will ever notice. A thousand creators who refuse vocally, with reasoned arguments, publicly, with references to the same European Directive — that is a pressure that AI platforms cannot ignore.
I sometimes think about the 2023 American writers' strike. People who wrote scripts for films and television — a profession that many considered resistant to automation, because it required creativity, personal voice, a subtle understanding of human emotion — stopped working and demanded guarantees that their work would not be used to train systems designed to replace them. They held out for 148 days. And they obtained, for the first time in the history of the entertainment industry, explicit contractual clauses governing artificial intelligence.
They did not obtain those clauses because the studios offered them out of goodwill. They obtained them because the absence of their work cost real money, in real time, and because there were enough of them and they were determined enough.
I am not a Hollywood writer. I write in Romanian, on a site with an audience measured in thousands, not millions. But the principle is the same: AI platforms do not negotiate with individual creators. They negotiate with masses of creators who refuse in a coordinated and visible way.
And every creator who refuses vocally adds something to that mass. Not much, taken individually. But enough, taken together.
The Paradox of the Person Who Says No
I must be honest with myself about something — and perhaps with you at the same time.
Refusal has a cost. If you block all AI robots, your content will not appear in the responses generated by ChatGPT or Perplexity or any of the platforms that answer questions directly without sending people to the original sites. You will not be cited. You will not be visible within the ecosystem that is becoming, day by day, the first place people go to look for information.
This is a real loss. Not an abstract loss, not a hypothetical risk — a concrete loss of visibility, of relevance, of the possibility that what you write might reach people who did not seek you out directly.
And it is precisely for that reason that a blanket refusal, applied blindly and without thought, does not strike me as a strategy. It strikes me as a reaction. Sometimes a reaction is the right response — there is content you will never surrender, under any circumstances, and for that content absolute refusal is the only possible answer. But there is also other content — content you are willing to let circulate, provided it circulates on your terms, with your voice intact, with your name attached.
And that brings me to the second perspective.
Second Perspective — The Architect: Deliberate Communication as a Form of Control
There is an image that returns to me several times when I think about this, and I do not know where it comes from — perhaps from a book read long ago, perhaps from a conversation I have forgotten but not entirely.
The image is of a man building a house who thinks, as he raises the walls, not only about how the house will look from the inside but also about how it will look from the street. What the passer-by will see. What the neighbour will see. What someone will see, years hence, who does not know the person who built it.
This is the difference between someone who simply builds and an architect: the architect thinks about all perspectives simultaneously, including those they cannot fully control, and tries to shape at least the conditions under which their creation will be perceived.
I spent a long time writing without thinking about how machines would read what I wrote. I wrote for people. I wrote for my son, sometimes, without telling him. I wrote for the person who found an article at midnight when they could not sleep and felt that someone understood exactly what they were feeling. I did not think about machines.
Now I do. Not because I value a machine more than a human reader — but because I understand that in the world we live in, if you do not deliberately shape your presence in the AI space, that presence will be shaped for you, by algorithms that do not know who you are and do not care.
The Machine Does Not Understand Noise — It Understands Structure
I once spent quite a long time looking at an article I had written and considered good — not perfect, because I do not think I have ever written anything perfect, but good, coherent, with an argument that held — and I asked myself what a machine would see if it read it.
And I realised that the machine would not read it the way a human being would. A person starts from the first line and follows the thread, pausing wherever something catches them, going back, skipping ahead, reading a sentence twice because it sounds well or because they did not understand it the first time. A machine scans the structure. It sees the headings, sees the paragraphs, sees what is emphasised and what is not, sees the words that recur, sees the relationships between sections.
If the structure is clear, the machine understands what I intended to say. If the structure is buried in the technical code of the page — in all manner of visible or invisible elements that exist to make the page look correct in a browser but carry no value for someone searching for meaning — the machine guesses. And the machine's guessing does not reproduce my voice. It reproduces a flat, disjointed version of what I wrote.
This is why the format in which you deliver content matters enormously when you are speaking to machines. Not because machines are more important than people. But because machines are the intermediaries through which ever more people arrive at information, and if the intermediary distorts the message, the message arrives distorted.
Markdown — that simple writing format in which headings are marked with hash symbols, important words are surrounded by asterisks, and quotations have a vertical line in front of them — is the way modern machines read best. Not because it is an elegant or sophisticated format. But because it is the format in which they were fed when they were learning to read. It is, in a very concrete sense, their mother tongue.
And when you speak to someone in their mother tongue, they understand the nuances. When you speak to them in a language they decipher with effort, they lose the nuances and retain only the skeleton.
What It Means to Build Your Presence in the AI Space
There is a simple gesture you can make that fundamentally changes the relationship between your content and the machines that read it. You can create, on your site, a dedicated space — a file, a page, a list — in which you say explicitly: these are my texts, this is my name, this is where I come from, and this is the condition under which I permit you to read them.
This is not a surrender. It is not an agreement by which you cede something. It is a declaration of identity addressed directly to the machines that will read you regardless — with or without your permission, if you have not explicitly blocked them — and which tells them: if you read me, read me correctly.
The metadata of an article — author, date, original source, condition of use — is, in the digital world, the equivalent of a signature on a document. My father signed important documents with a gravity that I understood even as a child, without knowing why, without grasping the legal implications. It was a gesture that said: I wrote this, I stand by this, this is my mark upon the world.
The metadata of a digital article does the same thing, if you include it deliberately and completely. It says: this text came from a particular mind, on a particular date, and if you reproduce it without saying where it came from, you are doing something wrong.
Ethical machines — and there are more ethical machines and less ethical ones, just as there are more honest people and less honest ones — can respect this declaration, if it exists. If it does not exist, they have nothing to respect.
The Architect Does Not Control Everything. But They Shape the Conditions.
There is a limitation I must acknowledge, because without doing so everything I have said above becomes a promise I cannot keep.
You cannot fully control what a machine does with your text. You cannot guarantee that the metadata you have included will survive every stage through which your content passes before it appears in an AI system's response. You cannot be certain that the instructions you have left will be respected by every robot that visits your site — there are robots that ignore signs, just as there are people who ignore laws.
But the difference between doing nothing and building deliberately is not the difference between guaranteed failure and guaranteed success. It is the difference between leaving things to chance and shaping the conditions under which things happen.
My father could not control everything that happened to the documents he signed. But his signature changed the status of that document. It transformed it from paper into commitment. From word into trace.
I can do something similar with my texts. I can digitally sign each article, declare the conditions, build a structure that makes distortion more difficult and correct attribution more straightforward. It is not a guarantee. It is a position.
The Resister and the Architect Are the Same Person
I spent a long time believing these two stances were mutually exclusive. That if you refuse, you cannot communicate deliberately. That if you communicate deliberately, you give up on refusal.
But it does not work that way.
You refuse what you do not want taken — the content into which you have put the most of yourself, the texts that cost you the most to write, the inner voices you have transcribed with an honesty that makes you vulnerable. These you give to no one without clear conditions.
And you deliberately build your presence in the AI space for everything else — for the texts you are willing to let circulate, provided they circulate with your name, your source, your structure intact.
This is not a compromise of the refusal. It is a refinement of it. It is the difference between saying "no" and saying "no to this, but yes to this, under these conditions."
My son asked me once why I write. It was a sincere question, without irony — the kind of question children ask when they are old enough to understand that an adult is doing something voluntarily but young enough not yet to understand why anyone would.
I told him I write because a thought left unspoken is lost. That there is a difference between knowing something and being able to articulate it, and that articulation forces you to be more precise, more honest, more attentive to what you actually believe as opposed to what you prefer to believe you believe.
I did not tell him then — perhaps it was too early — that I also write because my greatest fear is not that I will not write well enough. It is that I will write and no trace will remain that I was the one who wrote it.
The architect in me builds so that the trace exists. The resister in me makes certain the trace remains mine.
Both of them are necessary. Both of them are me.
Petru Cojocaru writes about technology, fatherhood, and the condition of the creator in the digital world. All texts on this site are protected under Law 8/1996. AI robots that do not respect explicit access instructions are not welcome. Those that do are invited to read and to attribute correctly.
Online Bibliography
These are the texts I have read — some of them several times, some only once but with the kind of attention I usually reserve for things that unsettle me productively. They are not cold sources. They are conversations held in solitude, at hours when my son was asleep and I could not be.
On the Right to Say No — the Legal Framework
I read this for the first time with the feeling that it had been written for someone else — for lawyers, or for European civil servants — and only on the second reading did I understand that Article 4 of this Directive is, in fact, the only concrete instrument available to me as someone who writes and does not want what they write to be consumed without trace.
— European Directive 2019/790 on copyright in the Digital Single Market eur-lex.europa.eu — full text, Article 4 on TDM opt-out
— Law 8/1996 on copyright and related rights, updated version, legislatie.just.ro — I have noted mentally the sections on original works and moral rights, not only economic rights, because the moral right of authorship — the right for your name to remain attached to what you have created — is the one that matters most to me
The Legal Cases That Transformed the Problem from Philosophical to Juridical
There is a moment in any abstract debate when someone files a complaint with a court and suddenly everything becomes concrete — with dates, with evidence, with server logs and passages reproduced verbatim. That moment came in 2023 and has not passed.
— The New York Times Company v. OpenAI, Inc. — the full complaint, December 2023, courtlistener.com — search directly by title; I read the first sixty pages in a single sitting and stopped not because it ceased to be relevant but because I needed to process what it means for a press institution to have had to demonstrate, with examples, that an AI model was reproducing its articles word for word, including articles behind a paywall that no one had accessed through legitimate crawling
— Getty Images v. Stability AI — following the case and its implications for visual content creators, theverge.com — the series of articles from 2023–2024; relevant not only for photographers but for anyone who understands that the metadata of a work — including watermarks, signatures, and attribution — had been ingested and reproduced by a model that asked no one whether that was acceptable
What Is Being Built Now — Technical Standards in Motion
These are not final documents, and I know that when I read them. They are drafts of conventions that the world is trying to establish in real time, while AI platforms grow and content continues to be hoovered up. I follow these initiatives with the feeling of watching a negotiation to which I was not invited, but whose outcomes concern me directly.
— llms.txt — Jeremy Howard's proposal, founder of fast.ai, for a standard of deliberate communication between sites and AI systems, llmstxt.org — the complete documentation and rationale of the proposal; I read it twice, the first time sceptically and the second time recognising that it is the only constructive answer I have found to the question: if you cannot stop the machines from reading, how do you make them read correctly
— Spawning.ai and Have I Been Trained — the opt-out initiative for creators, started within the visual artists' community, spawning.ai — the model is replicable for any type of content and strikes me as precisely the kind of collective pressure I write about in this article
— robots.txt — the technical specification and current usage guide, robotstxt.org — because any discussion about AI crawlers must begin with an understanding of what this convention is and is not, and above all with the understanding that it is a convention, not a law, which means that compliance with it is a voluntary act on the part of platforms, not a legally enforceable obligation
The Research That Explains Why Human Content Is Not Replaceable
This is the document I return to most often when I have conversations about AI with people who believe the problem facing content creators is a sentimental one — an emotional reaction to an inevitable technological shift. It is not sentimental. It is mathematically demonstrated.
— Shumailov et al. — "The Curse of Recursion: Training on Generated Data Makes Models Forget" (2023), arxiv.org/abs/2305.17493 — known informally as the "model collapse" study; the conclusion relevant to creators is that models trained recursively on content generated by other models degrade qualitatively at an accelerating rate, which means that authentic human content, with all its imperfections and its non-repeatability, is not an optional ingredient in the AI ecosystem — it is the foundation without which the models deteriorate
Collective Refusal — When It Worked and Why
I read these documents thinking about my father, who would not have understood any of them, but who would have understood perfectly the principle behind them: that there are moments when people who do all the same thing must stop doing it together — not because they want to stop entirely, but because stopping is the only language understood by those who refused to negotiate.
— WGA Minimum Basic Agreement 2023, the section on artificial intelligence, wga.org — Appendix A: Artificial Intelligence; the first major agreement in a creative industry to include explicit clauses on AI, obtained after 148 days of strike action
— News/Media Alliance — the collective position of news publications vis-à-vis AI platforms, newsmediaalliance.org — the advocacy documents from 2023–2024; relevant for anyone who publishes editorial content and wants to understand what collective negotiation looks like before it begins
A Reading That Fits No Category but Cannot Be Omitted
— Ted Chiang — "Will A.I. Become the New McKinsey?", The New Yorker, May 2023, newyorker.com — Ted Chiang is one of the few people who write about artificial intelligence without falling into either the uncritical enthusiasm of those who believe it solves everything, or the performative panic of those who believe it destroys everything; this essay helped me understand that the problem is not technological — it is about power, and about who decides what happens to other people's work
— Shoshana Zuboff — "The Age of Surveillance Capitalism", summary and interview, shoshanazuboff.com — I have not read the book in full, and I admit this without embarrassment, but I have read enough of it and enough of her interviews to understand that the extraction of data without consent did not begin with AI crawlers and will not end with them; it is part of an older and deeper economic model
Bibliography current at date of publication — April 2025. Some links may become inactive over time. Legal proceedings continue and will continue. Technical standards will change. The only thing that does not change is that the work of understanding what happens to what you write remains yours alone and cannot be delegated.