In brief
This site uses LLMs as editorial production tools. This is not hidden — it is the starting point. AI generates a first draft; a human chooses the topics, verifies the facts, structures the arguments, and validates each publication. The model is a lever, not an author. This page explains the choice of transparency and what it implies.
Why say it
Many media outlets, blogs, and newsletters use LLMs in their production. Some mention it in fine print, others don’t mention it at all. This is not an accusation — it is a documented observation: since 2023, major news agencies (Associated Press, Reuters, BBC) have published internal guidelines on AI use, and the question of disclosure to readers is addressed unevenly.
The Associated Press formalized as early as 2023 rules for using generative AI in its newsroom, with an explicit requirement for human oversight of all published content. The BBC produced a 2024 ethical framework distinguishing acceptable uses (production assistance, translation, summarization) from prohibited uses (unsupervised publication). The New York Times, for its part, chose to restrict the use of generative AI in its newsroom while pursuing a lawsuit against OpenAI for unauthorized use of its archives — which illustrates the diversity of positions, even among the most structured players.
The choice here is different: own the tool, document the method, and let readers judge on substance.
What it changes — and what it doesn’t
What it changes
The production speed is real. An LLM can generate in a few minutes an article structure, a synthesis of sources, a reformulation. Without AI assistance, a single author could not cover this density of topics in the same timeframe.
The coverage surface is broader. Technical subjects — model architectures, inference protocols, benchmarks — can be addressed without requiring researcher-level expertise at entry, provided every claim is verified against primary sources.
What it doesn’t change
The quality of an article still depends on the quality of sourcing, the rigor of verification, and the clarity of structure — three dimensions that the model does not guarantee alone. An LLM hallucinates references, confuses dates, presents opinions as facts. Without systematic human review, the error rate is high.
Editorial relevance remains a human judgment. Deciding what to cover, from what angle, with what depth — this is not a task delegable to a language model.
How this site is produced
The process is sequential and documented:
- Topic selection — human. Topics are chosen based on their relevance to the target audience, not based on what the model spontaneously proposes.
- Source collection — mixed. Primary sources (papers, documentation, reports) are identified by a human. The model helps synthesize their content.
- First draft generation — model. The LLM produces a structure and content from the provided sources.
- Verification and revision — human. Every factually loaded claim is checked against the cited source. Vague or unsourced sections are reworked or removed.
- Publication — human. The final decision to publish, in the published form, is made by a human.
This is no different from using a search engine, a spell checker, or a news aggregator. The tool changes; editorial responsibility stays in the same place.
In short: think of a journalist using a dictaphone to transcribe interviews. No one asks whether the dictaphone “writes” the article — it is the tool, the journalist remains the author. The LLM plays the same role here: it accelerates raw material production, but the editorial choice, the verification, and the final responsibility stay human. Transparency about usage is precisely what keeps the chain of responsibility visible.
The authenticity problem
The recurring question: “is an article produced with AI authentic?”
It’s a real question, but it’s framed wrongly. The authenticity of editorial content does not hinge on the absence of tools — it hinges on honesty about sources, rigor of verification, and assumed responsibility for what is published.
Research in communication and information science shows that reader trust in content is more correlated with transparency of methods than with the absence of technological assistance. An unsourced article written without an LLM is no more reliable than a sourced and verified article produced with AI assistance.
The real problem is not the tool — it’s what goes unsaid. Content presented as entirely “human” when it is mostly generated without supervision is deceptive. Content that is generated, verified, and acknowledged as such is honest.
Why not “100% human”
The alternative would have been to say nothing and produce content presented without mention of any tool. This positioning is adopted by the majority of publications that use LLMs — out of commercial caution, fear of stigmatization, or because the question was never settled internally.
That is not the choice made here, for two reasons.
The first is pragmatic: readers who can spot LLM-generated text are increasingly numerous. Silence on method is a trust debt — when it comes due, it comes due with interest. The second is principled: if the method doesn’t hold up to scrutiny, it’s the method that needs changing, not the communication about the method.
Recent work in media studies shows that proactive disclosure of AI use in editorial production does not necessarily erode reader trust — provided the disclosure is accompanied by a clear description of the human verification process. It’s not “AI or human”: it’s the degree of supervision that matters.
In short: transparency is not a moral comfort — it is a debt that gets repaid. A site claiming to be “100% human” while it actually generates mostly with an LLM accumulates distrust capital. When that comes to light, the erosion is sharper than honest upfront disclosure. The choice here is pragmatic: own up now to preserve trust later.
What readers can expect
Every article published on this site:
- cites its sources at the end of the page, in bibliographic format
- flags unverified claims ([UNVERIFIED])
- flags structurally absent data ([DATA ABSENT])
- does not attribute statements to individuals without a verifiable direct quote
- does not speculate without marking it explicitly
This is not a guarantee of infallibility. It is a traceable method. If an error is identified, it is corrected and noted.
Key takeaways
- This site uses LLMs as production tools, with human supervision at every step.
- Much online content is produced the same way, without saying so.
- The authenticity of editorial content depends on transparency of methods and rigor of sourcing, not on the absence of tools.
- Editorial responsibility remains human: topic selection, fact verification, publication decision.
- Readers are invited to judge on substance: cited sources, verifiable claims, documented corrections.