No assistance
The piece was written without meaningful help from a generative AI tool.
Editorial commitment
AI charter
Le Recul may use artificial intelligence as a working tool, never as a replacement for editorial judgement. This charter sets out what AI is allowed to do, what it must never do, and how our content is checked.
AI can help. The verdict stays human.
Our principle
Le Recul treats AI as a tool, not as an authority.
An AI can help search, structure, summarise, compare, rephrase or spot inconsistencies. It can speed some tasks up and make the work more efficient.
What it does not do is decide on its own what is true, important, useful or reliable.
Everything we publish has to stay understandable, checkable and editorially owned. When a piece of information is uncertain, the page says so. When proof is missing, the verdict stays cautious.
Levels of AI assistance
Le Recul may state the level of AI assistance behind a piece whenever that is useful to the reader.
Light assistance
AI may have helped rephrase, structure, summarise or correct a passage, without producing the substance of the piece.
Partial assistance
AI helped prepare an outline, compare information, synthesise sources or draft a first version of some passages. The piece is then checked, corrected and approved by a human.
Heavy assistance
AI played a significant part in preparing the text, organising the data or producing a first draft. This level requires closer proofreading, source checking and human editorial sign-off before publication.
Supervised automation
Some rankings, benchmark collection or data updates may be assisted by scripts and automated tools. That automation does not replace the editorial method: it has to remain controllable, documented and verifiable.
What AI is allowed to do
AI may be used to help:
- prepare a first structure for an article;
- summarise documents or official announcements;
- compare several sources;
- rephrase a passage to make it clearer;
- spot inconsistencies or repetitions;
- prepare comparison tables;
- assist in collecting public, permitted benchmarks;
- help identify the points that need checking.
AI can therefore make the work faster, but it is never enough on its own to turn unverified information into proof.
What AI must never do
At Le Recul, AI must never:
- conclude in place of human judgement;
- invent a source, a figure or a quote;
- simulate a test that did not happen;
- fabricate a reader account;
- create a screenshot presented as authentic;
- change the meaning of a quoted statement;
- publish without human proofreading;
- turn an unverified claim into an established fact.
An impressive demo does not become proof because an AI phrased it well.
Sources, figures and quotes
Sources, figures and quotes are checked before publication.
When an article mentions a price, an announcement, a feature, a benchmark, a limitation or a technical claim, the source behind it should be identifiable wherever possible.
Our priority sources are:
- official documents;
- product pages;
- technical documentation;
- recognised benchmarks;
- public reports;
- company communications;
- identifiable user accounts;
- tests we ran ourselves, or that can be reproduced.
AI can help find or compare sources, but it is never treated as a primary source itself.
Tests, screenshots and reader accounts
A real test has to correspond to something we actually did.
If Le Recul states that a tool was tested, that test is not invented, simulated or inferred from a description alone.
Screenshots, results, exchanges, exports, emails, files and reader accounts have to correspond to real material. An AI must not generate fake evidence to illustrate a test.
Reader accounts are treated as signals. They can help spot a problem, a trend or a limitation, but they do not replace verification.
AI ranking and benchmarks
The AI ranking may use public data, benchmarks, technical sources and automated processing.
That data may be collected, prepared or normalised with the help of scripts or AI tools. But the ranking has to rest on a method that is explicit, consistent and verifiable.
A score does not sum up everything about an AI model.
Le Recul keeps use cases apart to avoid absurd comparisons: chat, code, image, video, audio, cost/performance, local models, cloud models and agentic systems do not answer the same needs.
When the data is too thin, the site says so: partial coverage, limited data, fragile score, or a result still to be confirmed.
Images, audio, video and synthetic media
Le Recul may use images, visuals, audio or video generated or altered by AI only where their nature does not mislead the reader.
Synthetic content is never presented as real evidence.
An AI-generated image is not presented as an authentic photograph. A generated voice is not presented as a real recording. A manipulated video or screenshot is never used to suggest an event that did not happen.
Wherever there is a risk of confusion, the content is labelled clearly.
Corrections and updates
AI can help spot an error, an inconsistency or an out-of-date piece of information.
The decision to correct belongs to the editorial team.
When a significant error is identified, Le Recul corrects the content concerned. If the correction changes the meaning of an article or a verdict substantially, the update is made visible or stated explicitly.
Content about AI can go out of date quickly. Le Recul therefore distinguishes what is established, what is dated, and what needs re-checking.
To report a factual error, use the Contact page or write to contact@lerecul.com. Reported errors are checked, and an important correction may be noted directly in the article or page concerned.
Editorial responsibility
Final responsibility for anything Le Recul publishes stays human.
An AI can assist the work. It does not sign a verdict. It does not carry responsibility for a publication.
The reader should be able to tell:
- what has been analysed;
- what has been tested;
- what has been sourced;
- what has been automated;
- what remains uncertain.
Transparency is not there to cover us. It is there so the reader can judge how reliable the content is.
In short
Le Recul may use AI, but does not hand its judgement over to AI. Our rules are simple:
- AI assistance allowed.
- Human checking required.
- Sources verified.
- Tests never simulated.
- Evidence never invented.
- Synthetic media labelled where needed.
- Editorial verdict owned.
AI makes promises. Le Recul verifies.