Does the research support your briefing?
Check more than the reference

Start with the statement colleagues might use, not the bibliography's appearance. Does it describe what a study tested, suggest a workplace application or claim that an approach works today? A sentence can combine these without making the joins visible. Separate them before looking for support. For the reference itself, we propose comparing the title, authors and publication record with the original publisher or research repository. Identify the version actually available. A link that opens establishes access to a page, not that the page is the cited work. A search result or the assistant's summary can help locate a publication; neither replaces opening it. If identity remains unresolved, record that rather than declaring the reference fabricated.

Finding the right publication answers only the identity question. Next ask which passage supports which part of the briefing. A paper about answering public questions with citations does not, merely by existing, establish that an assistant can reliably review supplier documents. ALCE makes a useful distinction between fluent writing, answer correctness and support from cited passages. In its historical experiments, answers generated without retrieved documents could perform well on correctness measures yet have poor citation support when references were attached afterwards. That is a reason to examine the connection between sentence and source, not just add a bibliography to a finished draft.

Open the relevant finding and read enough surrounding material to understand what the researchers asked and how they assessed it. Was the task generating an answer, finding information or judging whether a citation supported a sentence? What material and systems were used? Does the briefing preserve those boundaries, or quietly turn a research task into a workplace result? ALCE studied question answering using selected public reference collections. Providing retrieved passages produced strong results among its tested approaches; selecting generated answers by citation support also improved citation quality, including in human evaluation. These are useful positive findings, not evidence that citations are universally unreliable. The authors also acknowledge imperfect automated assessment, incomplete coverage of open-ended answers and difficulty recognising partial support. The generative-search study separately asked human annotators whether sentences had full support and whether individual citations offered full, partial or no support. Several sources could collectively support one sentence. A citation that supports only one part should not lend borrowed authority to the rest.

The coordinator can retain a source link, version and section or passage location beside the wording being reviewed. Note what the passage supports and what it leaves unanswered. If a sentence combines a historical finding with a claim about present supplier-document work, preserve the finding without presenting the application as observed evidence. The generative-search evaluation found generally fluent, apparently helpful responses, and many citations did support their associated statements. It also found unsupported statements and inaccurate citations. Its first complete search responses were collected in early 2023, not from today's assistants or internal briefings. Different rates of declining to answer further complicate direct comparisons. Where interpretation needs subject expertise, the briefing owner can seek an appropriate reader. Where the original passage is unavailable, keep support unresolved. Another fluent explanation from an assistant does not supply the missing passage.

We propose a short reading note connecting the briefing sentence, identified publication, relevant source locations, supported wording and open questions. This is an editorial aid, not a procedure tested or required by either study. The briefing owner can retain wording that accurately describes the original finding, qualify a broader inference, remove unsupported claims or keep circulation pending. If several citations are needed, make their separate contributions understandable. Before sharing, compare the revised sentence with the note: has an observation about historical systems become a claim about current reliability during editing? Even a faithful account of a paper does not independently establish that its findings are true or applicable to the intended decision. The generative-search authors explicitly distinguish source verifiability from factual truth. The useful handoff therefore tells colleagues what the source says, under which conditions, and what still needs clarification. It does not label the briefing automatically verified.