You will be able to use an assistant to find, sort and summarise sources while you keep the judgement.
Write down one question from your job where someone will later ask you to back up the answer, such as what your competitors charge or which rule applies to your team. Keep it beside you as you read. The method described here works best for questions you will need to defend, and in most jobs those are the research questions that matter.
Kavitha researches competitor pricing for a company that looks, from her sources, to be in workplace health services. After Lesson 4.2, she found that checking every claim in her research-mode report took longer than the report had saved her, and she wondered whether to go back to a search engine and a notepad. Neither going back to manual searching nor relying on the AI report solves her problem. What helps is a different order of work. The assistant speeds up finding and reading sources, and your starting point becomes a set of sources you have already checked, instead of an AI answer you then have to unpick claim by claim.
Turn search on, describe your question, and ask the assistant to suggest sources instead of giving you an answer. Ask for likely primary sources, meaning the agencies, official publications, annual reports and industry bodies that would hold the information. For each one, have it note what the source probably contains and why it is relevant to your question. What comes back is a reading list, and you should treat every item on it as a lead you still have to confirm.
Then open each suggestion yourself. Assistants sometimes propose a report with a believable title from a real organisation that was never actually published, or a page that has moved since. Anything you cannot find comes off the list.
Kavitha's list for her pricing question held the competitors' own websites and published rate cards, a government agency page on workplace health, an industry association's annual survey, and two news articles. On checking, one report attributed to the association did not exist, so she removed it, and the remaining sources all opened.
When you have sources you have opened and judged reliable, hand them to the assistant directly by uploading the PDFs or pasting in the relevant pages. Keep to your company's data rules about what you can share. From here, ask about those documents and leave the wider web out of it. The assistant's job shifts from searching and summarising whatever turns up to reading material you picked.
For every statement it makes, ask for the exact quote behind it and the page or section where it appears. Lesson 5.2 of Working with AI assistants, "Make it quote and point to the source", walks through how to phrase that request. Check a few of the quotes against the documents with your own eyes, because a model with the sources in front of it can still paraphrase loosely or pin a quote on the wrong document.
Done this way, a long annual report or a dense regulation takes a fraction of the time that reading it cover to cover would, and each statement you get back carries a pointer to the page you can follow up.
Often the more useful question to put to a set of sources is "where do they disagree?" rather than "what do they say?" Ask the assistant to go through the main points, compare the sources on each, and list the differences with quotes from both sides.
Sources can give different figures because they count in different ways, cover different years, or draw the boundaries of the market differently. One may describe a rule that another reports has changed. A competitor's website might show a price that a news article says has since risen. Each of these counts as a finding in its own right, and it is frequently the one that changes a decision.
Ask for one smooth summary and these differences tend to disappear. Given a request for a single answer, a model will often take one source's version, or blend the sources into a statement that none of them made. Requesting the disagreements directly puts them in front of you.
Kavitha's sources split on whether competitors had shifted towards annual subscriptions. Two of the three competitors' rate cards listed only per-session prices, while the industry survey reported that most providers now offer subscriptions.
With sources, quotes and disagreements in hand, close the assistant and write the conclusion yourself, in two or three sentences. Say what the evidence shows, how confident you are in it, and what it means for the decision in front of your team.
Collecting sources is something anyone can do, and the conclusion is the part of research your employer is paying you for. It depends on judgement: choosing which source to trust when two disagree, working out what a figure means for your company, and deciding whether the evidence is strong enough to act on. Letting the assistant write the conclusion hands over the one part of the job that needed you, and you may not spot that an AI-written conclusion rests on the weakest source in the set.
Kavitha wrote that subscriptions are common across the industry but have not yet reached their closest competitors, and that this gap might be an opening worth testing.
Now try the method on a question from your own work. Start with the reading list, and keep only the sources you have opened and are willing to stand behind.
Collect three primary sources on a work question, ask the assistant where they agree and disagree with quotes, and write your own two-sentence conclusion.
Junxiong-WFG Organisation is an authorised representative of AIA Financial Advisers Private Limited (Reg. No. 201715016G).