Sources stay visible.
When an argument depends on outside reporting, research or data, the source should be traceable from the page. Primary sources are preferred when they are available and relevant.
Facts, analysis and opinion are different things.
A measured result should not be written like an interpretation. An interpretation should not be dressed up as a measured result. Strong language is fine. Fake certainty is not.
Dates matter.
AI products, model behavior and search interfaces change quickly. Time-sensitive claims should be dated so an old observation does not quietly masquerade as current truth.
Research publishes the method.
Original research should state the question, population, collection period, protocol, aggregation choices and limitations. If the method changes materially, the record should say so.
Corrections are part of the work.
Material factual errors should be corrected. Significant corrections should be dated or otherwise made visible when the change affects the meaning of the piece.
AI is allowed to help. It is not allowed to invent.
AI can assist with research organization, source comparison, coding, analysis, editing and commissioned editorial art. It may not fabricate sources, quotations, interviews, data, experiences or claims of firsthand knowledge.
Identity matters.
AI should not impersonate a real person in a way that causes a reader to believe that person personally said, recorded or experienced something they did not.
“Synthetic intelligence should never replace authenticity. It should enhance it.”
Uncertainty stays visible.
If a conclusion is tentative, say so. If a study cannot establish causation, do not imply causation. If the model version is unknown, write unknown instead of guessing.
No content quota.
The publication is not obligated to publish daily, weekly or on any algorithmic cadence. The standard is whether the work adds evidence, a useful argument or an original way of seeing the problem.