Build a Useful Evidence Packet From a Long Report

A 146-page impact report arrives with a request for a 500-word authored article by tomorrow, the kind of piece that later needs a full fact-check of a long bylined article. Search finds 37 mentions of “access,” several tables use different populations and the executive wants a strong opening number. Sending the whole PDF to a writer technically provides the source. It does not make the evidence usable. Build an evidence packet instead: one card per proposed claim, each with the exact passage, its table or footnote, and a note of what the passage does and does not support.

An evidence packet is a focused bridge between the report and the deliverable. It gathers the passages most relevant to proposed assertions, preserves enough context to interpret them and records what each passage can support. The full report remains available for inspection; the packet never replaces it.

With the assignment, not the report: list the claims the piece is likely to make, then search the report for the passages that bear on each one, including the ones that limit it. Reporters are watching for exactly this: 72% told Cision's 2025 State of the Media survey↗ they worry about factual errors in AI-generated PR content¹.

Start with the article brief, intended audience and likely argument. List the questions a writer must answer: What changed? For whom? Over what period? How was it measured? What can the executive reasonably infer? Keep the brief in its own section. A request to lead with broad impact is an editorial goal, not evidence that broad impact occurred.

Turn likely sentences into atomic assertions. “The program expanded access and improved completion” contains at least two claims and may hide a third about cause. Search the report for each concept, including the report’s own terminology. A relevant keyword hit is a candidate. Read the page, section and notes before placing it in the packet.

Include contradictory or limiting passages. A regional result may sit beside a methodology note excluding certain participants. Omitting that note makes the packet faster to scan at the cost of changing the meaning. The writer needs the constraint early, not during final approval. The BBC treats such omissions as an accuracy problem: its editorial guidelines↗ warn that leaving out significant details can make content inaccurate².

The proposed claim, the source passage with enough surrounding context to read it correctly, and two separate decisions: whether the passage supports the claim and whether it may be used publicly.

Proposed assertion

Write the claim in plain language and label its status as proposed. Identify the entity, population, period and comparison it concerns. If the wording contains two checkable ideas, split it into separate cards even when the final prose may combine them.

Source context

Record report title and version, page, section, table or figure, exact relevant passage and enough surrounding text to understand it. Include definitions and footnotes, because a footnote can change what the number means. For a chart, retain labels, units, legend and any note explaining missing or adjusted data.

Support decision and use decision

State what the passage supports, what it does not establish and any unresolved conflict. Name the reviewer and decision date. Separately record whether the information and wording may be used publicly, with its decision owner and conditions.

QuoteIt does this for every document a team uploads. Each fact becomes an item that keeps whose result it is, its value, scope and period beside the passage it came from, so the item reads correctly on its own. A draft is written from the items chosen for the brief, and each claim in it is checked against them, so a sentence that merges enrollment with an outcome is flagged before review.

It splits one sweeping sentence into the separate claims the report actually supports, each with its own population and denominator.

Imagine a fictional training charity whose report says 620 learners enrolled during the year. A results table shows 410 completed at least one module, while a footnote explains that completion data was unavailable for one partner site. Elsewhere, a survey of 180 respondents shows that 72 percent felt more confident applying for work. These invented figures are for demonstration only.

A proposed sentence says, “The program gave 620 people the confidence to find work.” The enrollment count does not establish a confidence outcome, the survey covers a smaller responding group and neither passage establishes that anyone found work. The packet should create separate cards for enrollment, module completion and the respondent-reported confidence result. It should retain the missing-site note and survey denominator. For survey findings, the American Association for Public Opinion Research↗ lists what should be disclosed with results, including the dates of data collection and the sample sizes³, and both belong on the card.

A supportable proposal might say, subject to review and public clearance: “During the year, 620 learners enrolled. Among 180 survey respondents, 72 percent reported greater confidence in applying for work.” The writer can then offer the executive’s approved interpretation without converting it into an observed employment outcome.

As big as the deliverable needs: a handful of cards for a reactive statement, a broader set for a campaign, ordered by the piece rather than by report page.

Order cards according to the deliverable, not report pagination. Put the strongest relevant evidence near the section it informs and flag cards requiring a decision. Add a short source map showing where definitions, methodology and limitations live. Give the writer access to the complete original so they can inspect material beyond the extract.

A compact packet takes time to assemble. For a short reactive statement, prepare only the cards required for its few claims. For a long campaign, a broader packet can reduce repeated searching, while every reused card still receives a currency and permission check, the same check that keeps approved company facts current over time. Cards that went unused are worth keeping too; there are ways to make unused research useful on the next brief. The packet succeeds when another reviewer can reproduce each support decision from the cited original page. The BBC asks its content makers to keep research records in a way that allows double-checking by another member of the team², and that is a fair standard for a packet.