Research synthesis across PDFs, transcripts, and notes

The problem with research isn't finding information. It's synthesising it. You can download twenty PDFs, pull a dozen articles, export three months of meeting transcripts — and then spend hours reading across all of it trying to connect what matters. Cowork doesn't read faster than you. It reads everything simultaneously and produces a structured output that lets you start at "what does this mean" instead of "what does this say."

Research synthesis is Cowork's answer to the context window problem. In Chat, uploading multiple documents fills the window quickly. At ten files, you're already managing what gets included. At twenty, the model is working with a truncated view. Cowork reads the full folder — processing documents sequentially, referencing earlier findings as it goes, synthesising across the complete set without you curating what fits.

The use cases

Reading across 20+ PDFs and producing a structured brief. The analyst's core use case. Drop everything into raw/ — research reports, articles, whitepapers — and ask for a structured brief covering the key findings, conflicting evidence, and open questions. You get a document you can actually act on, not 400 tabs.

Meeting transcript analysis. Export your transcripts into a folder and ask Cowork to extract themes, decisions, action items, and follow-up drafts. It works across multiple meetings, spotting patterns in decision-making or recurring blockers you might miss session by session.

Personal knowledge synthesis. If you keep notes — in Notion exports, markdown files, text files — Cowork can mine them for patterns you've missed. "What are the common objections across my sales call notes from the last quarter?" is a synthesis task that takes minutes, not hours.

Due diligence. For a company, investment, or real estate decision: dump the available material into raw/ and ask for a structured assessment. Financials, news coverage, competitor analysis, reviews — Cowork reads across the pile and produces a brief with flagged risks and open questions.

Contract review. Feed contracts into raw/ and ask Cowork to extract key terms into a structured table: parties, duration, payment terms, renewal clauses, termination conditions, unusual provisions. The output is a table in out/, not a paraphrase in the chat window. Have a lawyer review anything consequential, but using Cowork to pre-digest the structure saves hours of that review.

The prompt pattern

Use this structure on every research synthesis task:

"Read everything in raw/. Produce out/brief.md with sections for [X, Y, Z]. Extract all decisions into out/decisions.csv. Note conflicting information in out/open_questions.md."

Three outputs, always:

1. The brief — the structured synthesis, with explicit sections you specify
2. The decisions extract — structured data you can act on directly (a CSV means it's sortable, filterable, importable)
3. The open-questions log — where Cowork flags conflicts, gaps, and things it couldn't resolve

The open-questions log is not a failure mode. It's a feature. Research tasks almost always have conflicting sources or incomplete information. You want that surfaced explicitly, not averaged away into confident-sounding prose.

The validation step

Always tie your success criteria to a specific file existing with specific sections:

"Success: out/brief.md exists and contains sections for Executive Summary, Key Findings, Conflicting Evidence, and Recommended Next Steps. out/decisions.csv exists and contains columns for decision, date, source, and owner."

This gives Cowork a checklist to verify against before finishing. It also gives you a clear rubric when you review the output — you're not asking "is this good?" but "does this have what I asked for?"

Without a success criterion, Cowork produces whatever structure seems reasonable. With one, it produces what you specified and flags gaps in out/open_questions.md instead of silently filling them.