There is a specific kind of pain that knowledge workers know well.
You need to make a decision. Or write a brief. Or update a stakeholder on something complicated. You have been reading about it for days — articles, reports, PDFs, meeting notes, half-remembered conversations. You have the information. What you do not have is the synthesis.
So you spend your afternoon doing the work that computers are actually quite good at: taking many things and turning them into one coherent thing.
Here is how to stop doing that by hand.
The Setup
The research synthesizer is a two-pass process. First pass: you feed. Second pass: you shape.
It works for any synthesis task: competitive landscape, market update, technical architecture review, policy analysis, post-mortem documentation, investment thesis, project status update. The content changes. The structure does not.
What you will need:
That last one matters. The worst briefing documents are “everything we know about X.” The best ones are “here is what you need to know to decide Y by Friday.”
Intake and Extraction (5 minutes)
You are not asking for a document yet. You are asking for the scaffolding. This step catches bad sources before they end up in the final brief — and surfaces the gaps, which is often more valuable than what you found.
I need to synthesize research on [topic]. Here is my source material: [Paste all your sources here — articles, notes, excerpts, URLs] Do not write the final document yet. First: 1. List the 5-7 most important facts, data points, or claims across these sources 2. Identify where sources agree and where they conflict 3. Flag anything that seems uncertain, dated, or poorly sourced 4. Note what is missing — what question does this research NOT answer that the reader will probably ask? I will review your extraction, then we will write the document together.
Why this works: Most people skip straight to the final output. The extraction step gives you a chance to review before any bad information gets laundered into an authoritative-sounding document.
The insight: What is missing from your research is often more important than what you found.
Draft the Document (10 minutes)
Once you have reviewed the extraction, run this. The explicit format matters: without constraints, language models write very thorough documents that nobody reads.
Good. Now write a briefing document for [audience: your manager / the board / a new hire / a client who has 10 minutes]. The document should enable them to [specific decision or action]. Format: - One-paragraph executive summary (what they need to know in 60 seconds) - 3-5 key findings with supporting evidence - One section on open questions or risks - One recommended next step Use the factual claims from our extraction. Do not add claims we did not source. Flag uncertain points with "(unconfirmed)" rather than stating them as fact. Target length: [1 page / 3 paragraphs / 500 words — whatever fits your context]
Why this works: The format constraint forces a document that matches how your audience actually consumes information. The "(unconfirmed)" flag keeps you honest without burying the document in caveats.
The goal: A document that enables a specific decision — not a document that proves you did research.
When Your Sources Conflict (use when needed)
If Pass 1 surfaces real disagreements between sources, run this before finalizing. Most people skip this step. The result is a document that sounds more confident than the underlying evidence supports — which is fine until someone in the room has read the conflicting report.
You flagged a conflict between [Source A] and [Source B] on [claim]. Before I finalize the document, help me think through this: - What would explain this disagreement? (Different time periods? Different methodology? Different definitions?) - Which source do you think is more reliable and why? - How should I represent this uncertainty in the final document without losing the reader?
Why this works: Conflicting sources are not a problem — they are information. This prompt turns the conflict into a reasoning exercise rather than a coin flip.
The protection: A document that acknowledges uncertainty honestly is more credible than one that papers over it.
The Pattern Worth Noticing
There is something consistent across the prompts in this series: the two-pass structure appears everywhere.
Generate, then evaluate. Draft, then critique. Extract, then synthesize. The instinct is to ask for the final output immediately. The better approach is to ask for the intermediate work first — the outline, the extraction, the first pass — because that is where the errors enter and where you can catch them cheaply.
The research synthesizer is a clean example of this. The first pass costs you five minutes to review. Missing a bad source in the final document costs you credibility with the person you sent it to.
20 min total (two passes) vs. an afternoon of manual synthesis
A brief that enables a decision — not just a document that proves you did research.
Try It This Week
Pick one report, brief, or summary you have been putting off because it requires synthesizing multiple sources. Run the two-pass process. See how long it actually takes.
The goal is not to see if AI can do it. The goal is to find out how much time you were spending on the mechanical part of thinking — and reclaim it. Reply to this email if you have a synthesis challenge you want us to build a prompt for.
That’s 12 issues. 12 workflows. If you’ve been using them, you have built something most people haven’t — a practical AI skillset.