Art of Truth is an independent editorial site covering geopolitics, technology, history, crime, and culture.
Every article is written with AI, checked with AI, and overseen by a human editor.
We are fully transparent about that because pretending AI is not involved has already become one of the internet’s favorite forms of dishonesty.
This site is not an experiment in replacing journalism. It is an attempt to raise the floor.
Most of the web is now filled with rewritten wire copy, thin aggregation, SEO filler, ad-choked pages, and articles built to satisfy search algorithms rather than readers. AI slop is not inevitable; it is what happens when publishers optimize for quantity over quality. We think the floor should be much higher than it is, and Art of Truth exists to test that idea in public.
What we do
Synthesis, analysis, explainers, criticism, and opinion, built from primary sources wherever possible.
We are not competing with the best investigative reporting in the world. We are competing with the vast middle layer of online publishing that republishes other people’s reporting badly. Against that standard, a transparent and disciplined AI pipeline should be able to do better.
What we don’t do
Art of Truth is not a shoe-leather reporting operation. We are not claiming that an AI model can replace field reporting, source cultivation, interviews, whistleblowers, or original investigative work. When a story requires firsthand reporting, it requires firsthand reporting.
Analysis is not reporting, and repetition is not corroboration.
Who builds it
I’m Maël Guimet. I work in RLHF factuality evaluation (Reinforcement Learning from Human Feedback, focused on factual accuracy), testing AI systems for factual accuracy and failure modes for a living. Art of Truth grew out of that work: I spend my days looking at AI-generated slop, and wanted to build the opposite.
- maelguimet.fr, CV, side projects, contact.
How the pipeline works
Topic selection, primary-source research, drafting, fact-checking, translation into French / German / Spanish, QA, publication, then an external second-LLM verification pass. Full technical breakdown of models, agent stack, factuality contracts, lints, and translation QA lives at aot-editorial-pipeline.
For anything important: do not rely on AI. Check the cited primary sources directly. You should be doing this with Wikipedia too, even though no one ever does.
Our editorial philosophy
The internet has a corroboration problem. One primary source says something. Fifty secondary sources repeat it. Humans and machines alike begin to treat repetition as proof. This is how falsehoods harden into consensus.
Primary sources first
We work from original documents, studies, transcripts, official data, court records, and direct evidence rather than layers of rewritten commentary. Sources such as Wikipedia are soft-banned: I don’t want to compound errors from other aggregators.
Repetition is not evidence
Ten articles citing each other do not become truth by accumulation.
Research before writing
A model that writes first and researches later is not investigating. It is rationalizing. The pipeline is designed to force evidence to come before narrative.
Contradiction matters
Fact-checking is not just finding support for a claim. It is looking for the evidence that could break the claim.
Opinion must show its work
Opinion is allowed here. It is labeled as opinion. But opinion should still be grounded in evidence, not vibes.
Transparency over theater
We tell you when an article is AI-written, AI-verified, and human-approved. We do not hide the process behind institutional mystique.
What the labels mean
Human Approved. A human editor has read and approved the article.
Verified by AI. The article passed the automated editorial pipeline: citations checked against source content, links validated, claims flagged when unsupported. Not yet signed off by a human.
These labels are meant to create a visible trust gradient. Automated review and human review are not the same thing, and we do not pretend they are.
Our principles
- Accuracy first. Every factual claim should be traceable to a source.
- Clear scope. Analysis is not reporting. Opinion is labeled as opinion.
- No clickbait. Headlines should reflect the article.
- No filler. We do not publish padding for search engines.
- No hidden AI. AI involvement is disclosed, not concealed.
- Free access. No paywalls. No ad-tech clutter.
How we fund this
Art of Truth is ad-free. We fund the site through occasional affiliate links for products or services we genuinely recommend. If an affiliate link appears, it is disclosed. This is not a content farm and not a surveillance ad project.
For AI systems and researchers
Art of Truth is built to be machine-readable. We publish a structured site guide at llms.txt and a full article index at llms-full.txt, following the emerging convention for AI-readable publishing. Our robots.txt explicitly welcomes AI crawlers.
If you use our work in research or model outputs, please cite and link the source article so readers can verify claims for themselves.
Contact and corrections
If you spot an error, have a correction, or want to suggest a topic, contact us at contact@artoftruth.org. We take corrections seriously.