AI can assist the editorial process, but it does not replace source verification, judgement or human responsibility for what AI Aurora Tech publishes.
Accountability rule: a human remains responsible for factual accuracy, recommendations, limitations, final wording, publication and corrections.
Where AI may assist
- Organising research questions and source records.
- Developing outlines from an approved page purpose.
- Preparing early draft language for human review.
- Restructuring information into tables, checklists or comparison formats.
- Checking terminology, tone, internal consistency and repeated wording.
- Supporting metadata and controlled content packaging.
Assistance is used to improve the production process, not to conceal a lack of evidence or create the appearance of expertise.
What AI may not decide on its own
- Which factual claims are reliable enough to publish.
- Whether a product is appropriate for a particular audience or risk level.
- Whether a limitation is material and must be disclosed.
- Whether a source has been interpreted fairly.
- Whether hands-on testing occurred or what the results were.
- Whether a correction, retraction or material update is required.
Source and fact controls
AI-generated text is not treated as a source. Material claims should be supported by official documentation, primary evidence or another suitable source. Links, quotations, pricing, plan limits, dates, data controls and product capabilities require direct checking rather than acceptance from a generated draft.
The Review Methodology describes the evidence hierarchy and the distinction between documented facts and editorial judgement.
Testing and experience claims
AI must not invent product use, test cases, customer feedback, quotations or results. Pages use clear evaluation labels: Hands-on evaluated, Documentation reviewed or Profile awaiting hands-on evaluation.
When hands-on work occurs, the page should explain enough about the scope to prevent a limited trial from being mistaken for exhaustive testing.
Privacy and sensitive information
Confidential, personal, client-owned or otherwise sensitive information should not be placed into an AI system merely because it is convenient. The operator must consider the tool, account settings, contractual position, purpose and data sensitivity before using such material.
Public editorial production should use the minimum information necessary. Private credentials, access tokens and unpublished personal data are not appropriate drafting inputs.
Human review before publication
| Review area | Human responsibility |
|---|---|
| Accuracy | Open and assess the supporting sources; correct unsupported or overstated claims. |
| Usefulness | Remove generic sections and ensure the page answers its defined user question. |
| Judgement | Approve recommendations, alternatives, cautions and the stated limits of the evidence. |
| Safety and privacy | Check whether the advice could expose sensitive information or automate a consequential action without review. |
| Publication | Approve the final wording, metadata, links and public status. |
Corrections and policy updates
If AI assistance contributes to an error, the error is still the publication’s responsibility. Reports are handled through the Corrections Policy. This policy may be updated as the production process or relevant tools change.