Compile a Research Brief from Multiple Sources

A research brief sounds simple but takes a surprisingly long time when done manually. You read twelve sources, take notes, try to remember which PDF said what, and end up with a document that is half synthesis and half hedged paraphrase. Cowork handles the reading and structuring pass. Deciding which sources are credible and whether the conclusions hold up is still your job.

Assembling the Source Folder

Create a folder called something like research-[topic]-[month]. Into it, drop:

• PDF articles, reports, or studies you have downloaded
• Text or Markdown files (notes you have already taken, copied article text)
• Saved web pages (most browsers can save a page as .html or .mhtml from File > Save Page As)
• Your own notes, even rough ones

If you have a URL that you have not saved yet, download the page. Cowork reads files in its connected folder; it does not browse the web on its own unless you have a web connector active. The connector approach works, but saved files are simpler to audit because you can see exactly what Cowork was working from.

One optional but useful addition: a file called sources-overview.txt in the same folder. Write two or three sentences explaining the topic and what you are trying to learn. This gives Cowork a frame for what counts as a relevant finding versus background noise. If you skip this file, the prompt tells Cowork to infer the topic from the content, which usually works fine but occasionally produces a brief that starts from an angle you were not expecting.

The Prompt

With the folder ready and Cowork pointed at it:

Read all the files in this folder. They are research sources on the topic described in sources-overview.txt (if that file exists) or inferred from the content. Write a research brief saved as research-brief.md. Structure it as: 1) Topic and Scope (two or three sentences), 2) Key Findings (bulleted, with a source citation for each claim), 3) Conflicting or Uncertain Points (note where sources disagree or where evidence is thin), 4) Gaps (what is not covered by these sources that would strengthen the brief), 5) Source List (filename, title if identifiable, and one-line description of each source). Do not invent claims. If something is only in one source, say so.

The instruction “do not invent claims” matters more than it might seem. Without it, a language model doing synthesis sometimes fills gaps with plausible-sounding but unsourced statements. The instruction, combined with the per-claim citation requirement, keeps the output anchored to what is actually in the files.

Reading the Output

The research-brief.md will have five sections. Here is what to look at in each.

Topic and Scope. Read this first. If Cowork has misread what the brief is about, the rest of the document is built on a wrong frame. Fix the scope description and re-run before reading further.

Key Findings. Each bullet should have a citation in parentheses pointing to a filename. Open a few of the source files and spot-check a few of the specific claims. This is not paranoia. It is normal editorial practice. Cowork synthesizes accurately most of the time, but “most of the time” is not the same as “always,” and a brief that goes out with a wrong statistic attributed to a source that does not say that is a problem.

Conflicting or Uncertain Points. This section is often the most valuable part of the brief. It shows you where the evidence is thin or contested. If you are writing marketing content from this brief, these are the claims you should either not make or make with appropriate hedging.

Gaps. Cowork will note what the sources do not cover. This is a reading list for your next round of research if you need to deepen the brief.

Source List. Verify that all your files appear here. If a file is missing, Cowork may not have been able to read it (a locked PDF, a corrupt file, a format it does not support). Check the missing file separately.

Verifying Claims Before You Use Them

The research brief is a synthesis of your sources, not a primary source itself. Before you use a specific claim in a campaign, a pitch, or a published piece, trace it back to the original source and read the passage in context. A single-source claim flagged in section 2 deserves more scrutiny than a claim supported by three independent sources.

If a finding is important enough to anchor a campaign around, it is important enough to read the original paper or report yourself, not just the brief. Cowork is good at reading and organizing; it is not a substitute for editorial judgment about what is strong enough to stake a claim on.

Iterating on the Brief

The first brief is rarely the final one. Common follow-up prompts:

“Expand the Key Findings section for [specific subtopic]. Pull more detail from the source files.”

“The two sources that disagree on [claim] are [file-a.pdf] and [file-b.pdf]. Read both passages and describe exactly what each says and where they differ.”

“Add a section to the brief called Implications with three to five practical takeaways for a B2B SaaS marketing team.”

Each of these refines the same document rather than starting over. Cowork will update research-brief.md in place, which keeps your folder clean.

When the Sources Are Paywalled or Restricted

Cowork reads files you give it. If a source is behind a paywall and you have legitimate access, download the PDF and put it in the folder. If you only have the abstract, the brief will only reflect the abstract, which Cowork will note. Do not paste full text from sources that prohibit it; that is a separate problem from the workflow itself.