How We Use Artificial Intelligence

We use AI tools. We think readers are entitled to know exactly where, and exactly where we don’t.

Where we use it

AI assists with organising research, drafting and structural editing, identifying gaps or inconsistencies in an argument, suggesting questions worth pursuing, and checking pieces against our own published standards. Some published prose begins as AI-assisted drafting before being rewritten and verified.

Where we do not

AI is never treated as a source. It does not supply facts, determine conclusions (except as set out below for the Cass Review appraisal), or make editorial decisions. It is never cited as an authority.

Every scientific claim, quotation, statistic and legal reference that reaches publication is checked against the primary or authoritative source by the author. There is no exception to this and no threshold below which it is skipped.

In our evidence reviews

Search strategies, inclusion and exclusion decisions, data extraction and certainty grading are performed by the author against the published method. AI is not used to screen studies, extract outcome data or assign any grade. Where AI has assisted in organising or formatting review material, the underlying judgements remain entirely human and are documented so they can be checked.

In the Cass Review appraisal

In the Cass Review appraisal, an AI model makes its own calls on some judgements, as a check on Eden’s. These are the judgements that need subject expertise, and any judgement that turns on a study Eden has graded before. Eden records her call first; the model then makes its call without seeing hers, and the two are compared only once both are fixed. Eden’s call is the one that counts. Where she makes no call on such an item, the model’s call is recorded and marked as the model’s. Every disagreement is published, and agreement is never counted as support.

The rules for this are fixed in the appraisal’s registered protocol. See Appraising the Cass Review.

Responsibility

Eden Openly is written and edited by one person. There is no second editor. Every published conclusion, interpretation and factual claim is that person’s responsibility, and so is every error.

We do not attribute mistakes to artificial intelligence. A tool cannot be at fault for what we chose to publish.

Why we say this

AI is now part of most publishing and is very rarely disclosed. We would rather be specific about it than let a reader discover it and wonder what else we left out.