Trust & verification
Editorial standards
The Impact of AI separates what is confirmed, what is claimed and what remains uncertain.
Source hierarchy
We prefer original research papers, regulator publications, official company announcements and direct public records. Reputable reporting can add context, but it does not replace the underlying evidence when that evidence is available.
Automation and human responsibility
Automation supports source collection, duplicate checking, translation assistance and early drafting for the 06:30 edition. It does not remove editorial responsibility. Before release, each report must have a reachable source, a valid date, a recognised sector, clear attribution, an impact assessment and an explicit account of evidence limits. A story is withheld when the source does not support the central claim.
International sources and translation
We seek original sources from outside the UK and the English-speaking technology industry. Editorial translations identify the country and original language, link to the source and are never described as certified legal translations. Material legal or technical interpretations should be confirmed by a qualified reader of the source language.
Claims and uncertainty
Performance figures supplied by a model developer are described as vendor claims unless independently replicated. Early scientific results are not presented as established findings. Forecasts and opinions are labelled and never rewritten as facts.
Reporting labels and source breadth
Every report declares its reporting basis. “Source analysis” means the central record relies on one directly linked source and is not independent corroboration. “Multi-source analysis” uses at least two direct sources. “Original reporting” requires a public note describing the interviews, data work, document review or other reporting performed by this publication.
For a normal edition of four or more reports, at least 25% must be multi-source analysis or original reporting before the publishing system will accept the edition. This is a minimum commissioning rule, not a claim that every subject has been independently verified.
Impact Tracker methodology
Every report receives a stable Impact Record identifier so later evidence can be connected to the same public record. The evidence stage is descriptive: “Announced” means the record is anchored in a primary statement or official publication; “Studied” means a research method and measured result are available; and “Observed” means independent reporting or documented real-world evidence is available.
Confidence is never generated from popularity or a hidden percentage. “Developing” normally indicates a single primary trail; “Supported” indicates research, independent reporting or more than one source; and “Corroborated” requires at least three sources, more than one source type and independent research or reporting. These labels describe the available evidence, not whether a technology is beneficial or harmful.
Each record also shows whether people impact and uncertainty have been documented. A record is marked updated when its checked evidence changes after release. Material changes should preserve the earlier source trail and explain what changed.
My AI Impact Radar methodology
The Radar is an editorial decision aid, not an employment forecast, psychological profile or probability that a job will disappear. It starts with broad task patterns reported in occupational research, applies one visible sector adjustment and expresses the result in five ordinal bands: task exposure, augmentation upside, pace of change, consequence of error and skills urgency.
The region and stated priority add context and practical advice but do not secretly alter the work scores. The evidence-strength label reflects the breadth of public task-level evidence for that broad occupation group. No score uses a person’s name, employer, salary, protected characteristic or browsing history.
Sources are linked on the tool itself. The model should be reviewed when major occupational evidence changes and replaced by observed site or workplace data when a narrower, properly governed assessment is available.
Corrections
Material corrections should explain what changed and when. Updates that add evidence without changing the conclusion are reflected in the article’s updated time.
Responsibility
The Impact of AI Editorial Desk is the publication byline and contact for editorial responsibility, corrections and source concerns. Operator information is set out in the privacy notice.