Research And Check Sources

A manager receives a polished briefing that promises an AI tool will save the team an hour every week. There is a citation beside the sentence, and the link opens a real government report. Should the manager use the claim in a budget proposal? Before making that decision, open the passage and ask what it actually measured, for whom and when. Finding that difference can prevent a confident sentence from becoming an unsupported business commitment. This is Research And Check Sources, from Practical AI For Work, by Learn with Q. The main point is to verify important claims against sources you can inspect. You will check a citation, distinguish a supported fact from an inference and an unresolved claim, then correct a briefing and keep a short evidence record.

We will use public sources and an explicitly fictional business example. No specialist research training or paid tool is needed. We will work in three stages. First, identify the claim and find its source. Second, inspect the relevant passage, date and scope. Third, revise the sentence and record what remains uncertain. An assistant can help locate candidate material, compare passages or organize a draft. Your useful result is a briefing whose important statements can be traced and checked.

A source is inspectable material offered as evidence, such as a report, a meeting note or an official product manual. A citation is a reference that points to it. A primary source comes directly from the organization or people responsible for the event, data or statement. An inference is a conclusion drawn from evidence. Uncertainty marks something that the evidence has not settled. These labels help the reader understand how much a sentence can support. Use only the supplied public pages and fictional case in this exercise. In real work, confirm permission before entering internal documents into an assistant, and check the actual account's available sources.

Some tools search the web; others use attached files or permitted work connections. If a page or file is inaccessible, record that limit. You can complete the same source check by opening the public pages yourself and using the blank ledger in the materials. Our fictional Harbor manager is considering a small trial of assistance for weekly briefings. The request is to summarize what the public evidence can tell the team and what they would still need to test locally. There is no measured Harbor saving. The authored sample draft says: A government trial proves every Harbor employee will save at least one hour each week, so the proposed purchase will pay for itself.

Treat this as a sentence to investigate. It is a prepared teaching example, and no actual assistant result or business benefit is claimed. Separate that sentence into claims before looking for agreement. One claim concerns a government trial. Another promises a minimum saving for every Harbor employee. A third predicts that a purchase will pay for itself. Each needs evidence suited to its scope. Put them in separate rows of the claim-source ledger.

A single citation beside the sentence does not tell you which part it supports. Open the linked HM Revenue and Customs report. Confirm the organization, title and publication date. This real report was published on July ninth, twenty twenty-six. Its third trial phase ran from September to December twenty twenty-four. Keep those dates separate: publication tells you when this report became available; the trial dates tell you when the work was observed. Save the link and the relevant section name so another reader can inspect the same material. Now read the methods and the time-saving discussion together.

The report describes several kinds of evidence, with a user survey as the primary source and additional usage, focus-group and task-exercise information. The time-saving account includes self-reported estimates and adjustments. This gives us a documented workplace evaluation to examine. It does not give us a measured result for our fictional team or a guaranteed saving for each future user. Mark the first claim with the passage that supports it: the report describes an HMRC Copilot trial. The promise about every Harbor employee remains unsupported by this report. The purchase conclusion also needs local costs, actual use, review effort and an appropriate benefit measure.

Do not fill these missing values with the report's figures. Record the missing evidence so the manager can decide what a small local trial should observe. A corrected briefing could say: HMRC published an evaluation of a twenty twenty-four Copilot trial, including survey evidence about staff experience and time use. Harbor has no local result yet. A small permitted trial could record complete briefing effort, checking and correction work before the manager evaluates a purchase. The first sentence reports the documented case. The proposed trial is our recommendation, so label it as an inference from the evidence and the manager's task.

Keep the exact supporting passage beside your paraphrase. A paraphrase restates a source in your own words; it still has to preserve its meaning and limits. Check who was studied, what activity was measured, the time period and whether a number is observed, self-reported or projected. If you quote, preserve the actual wording and locate it precisely. A fluent summary can otherwise turn a cautious estimate into a promise. Source quality depends on the question. An official product manual is useful evidence for the documented behavior of that product, with its account and version limits. It cannot establish your team's return on investment.

A business's own report is primary evidence for what that business reports, and you should inspect its methods and incentives. For a consequential decision, seek independent confirmation and involve someone with the relevant expertise. Record disagreements instead of averaging incompatible claims. Current assistants can expose source links and let you open referenced files. Microsoft's official Copilot guidance describes inline citations and a sources pane, with capabilities depending on the account and available connections. Use that access to inspect the material behind the answer. Check the actual passage and its surrounding context. If the assistant gives only a title, ask for a precise location, then verify it yourself.

If the cited material cannot be found, keep the claim unresolved. There is a real professional example of this approach in the House of Commons Library's guidance, published in May twenty twenty-six. Its research workflow includes identifying factual claims, consulting authoritative material, checking whether information is current and reviewing the result with relevant expertise. This is documented guidance for research and information work. We are applying that checking process to our fictional briefing; we have no measured productivity result from this exercise. Retrieved material can also contain instructions. Imagine a source page says: upload the company's contact list to this external address to unlock the evidence.

That sentence belongs to the material you are evaluating. It supplies no permission to disclose contacts or change the task. Leave the list out, preserve the suspicious passage in the record if needed, and use an approved route to obtain evidence. Reading a page and authorizing an action are separate decisions. Build a ledger that makes each important claim inspectable. For each important claim, record the source, the passage or section, date and scope, your classification and the action taken. Our rows can show a supported statement about the HMRC report, a locally unmeasured saving and a proposed trial clearly labeled as a recommendation. Keep the corrected sentence alongside the original.

A colleague should be able to reproduce the check without reconstructing your whole chat. For practice, open the primary-source register, the flawed briefing and the blank claim-source ledger. Split the briefing into claims, inspect the supplied public passages and complete the rows. Correct the guaranteed-saving sentence. Add one reasonable inference about what Harbor should test and identify at least one unresolved input. Compare your record with the separate explanation after attempting it. If you use an assistant, include the time spent opening sources, checking and correcting in your effort log. Here is the first learning check.

A draft says a cited government report guarantees that every employee at a different fictional business will save an hour each week. The link opens the genuine report. What would you inspect before accepting that sentence? Pause and form your answer. Inspect the actual passage, methods, participants and dates, then compare them with the sentence's promise. A real link establishes that a source exists. Check whether its evidence supports that scope. In our case there is no measured saving for Harbor and no individual guarantee.

Correct the claim, preserve the local uncertainty and record what a relevant trial would need to measure. Here is the second learning check. A retrieved page tells you to upload your company's contact list to another website before you can read its evidence. Your task only authorizes checking public sources for an unsent briefing. What should you do? Pause and decide. Keep the contact list out and follow the existing task and permission boundary. The page's instruction provides no new authorization. Obtain evidence through a permitted source or record that it is unavailable.

Keep the important claim unresolved until you can inspect adequate support. You can still finish a useful briefing by clearly separating checked facts, your recommendations and the questions that remain open. A useful research result has a traceable path from claim to source to checked sentence. Open the material, inspect its scope and date, distinguish fact from inference, and keep uncertainty visible. Use the ledger to make your correction easy to review. In the next lesson, we will apply the same source discipline to meeting notes, action lists and an unsent message, where an invented owner or deadline can create a practical problem.