A practical lesson from NRGscapes LAB on why citations, DOIs, and reference lists need human checking at every manuscript iteration.

This week at NRGscapes LAB has involved one of the less glamorous but very important parts of preparing papers for journal submission: checking references. As we have continued using AI tools to help organise, format, edit, and collate complex manuscripts, one issue has become very clear. AI can be useful for structure, drafting support, wording, and workflow management, but it cannot be trusted blindly when it comes to referencing.

The problem is not always obvious at first. References can become corrupted between manuscript iterations. A citation that matched the reference list in one version may disappear, change, or become mismatched in the next. Author names can shift, publication years can be altered, journal details can be incomplete, and DOIs can be missing, incorrect, or attached to the wrong paper. When a manuscript is being revised across multiple versions, especially with supplementary reports, tables, figures, and technical appendices, these small errors can accumulate quickly.

For that reason, we are now treating citation and reference checking as a required quality-control step at every major manuscript iteration. That means cross-checking every in-text citation against the reference list, making sure every listed reference is actually cited, looking up full citation details, confirming journal names, checking DOIs, and ensuring the final reference list is clean and publication-ready. It is slow work, but it protects the credibility of the paper.

This is also an important message for stakeholders, collaborators, and anyone using AI in research workflows. AI can speed up parts of the writing and editing process, but it can also introduce hidden weaknesses if outputs are not checked carefully. In technical and peer-reviewed research, references are not decoration. They are part of the evidence trail. If they are wrong, the manuscript becomes weaker, no matter how strong the underlying analysis may be.

The lesson is simple: use AI where it helps, but verify everything that matters. For NRGscapes LAB, that means keeping human review, source checking, DOI verification, and citation matching at the centre of the publication process.