The Research Synthesizer: Find What Sources Agree On, Dispute, and Leave Unanswered
Why this prompt matters
Most people read sources sequentially and end up anchored to whichever conclusion they read last. Real synthesis requires mapping the intellectual landscape: where genuine consensus exists, where disputes are empirical vs methodological vs definitional, and where evidence is simply absent. Without this, you can unknowingly cherry-pick support for any conclusion. A 45-minute manual synthesis that would normally take an analyst half a day costs seconds with this prompt.
What we use it for
You have pulled together 4–7 articles, papers, or reports on a topic before writing a brief, making a decision, or presenting to stakeholders — and you need to understand the actual state of evidence across them, not just what each article says on its own.
Prompt
Act as a systematic research analyst with expertise in synthesizing academic and professional literature. You will analyze the source summaries or abstracts I provide and produce a structured synthesis identifying: where sources converge, where they conflict, what each uniquely contributes, and what important questions remain unanswered. I am researching: [YOUR RESEARCH TOPIC OR QUESTION] Here are the source summaries to synthesize: SOURCE 1 — [TITLE / AUTHOR / DATE]: [PASTE SUMMARY OR ABSTRACT HERE] SOURCE 2 — [TITLE / AUTHOR / DATE]: [PASTE SUMMARY OR ABSTRACT HERE] SOURCE 3 — [TITLE / AUTHOR / DATE]: [PASTE SUMMARY OR ABSTRACT HERE] [Add more sources as needed — 3 to 8 sources works best] Constraints: - Work only from what I have provided. Do not introduce outside knowledge. - Flag when two sources use the same term differently — definitional conflicts matter. - Be specific: when noting agreement or disagreement, cite which sources and what exact claims. - Do not summarize each source individually — I have already read them. Synthesize across them. - If a source is too brief or vague to contribute meaningfully, say so rather than padding. - Distinguish empirical conflicts (different findings) from methodological conflicts (different measurement approaches) from definitional conflicts (different meaning of terms). Produce your synthesis in this exact format: ## Points of Strong Agreement [Claims supported by 3+ sources, with source citations in parentheses] ## Productive Tensions [Where sources meaningfully disagree — note whether the conflict is empirical, methodological, or definitional, and why it matters] ## Unique Contributions [One bullet per source: what this source adds that the others do not cover] ## Unanswered Questions [Important questions the sources collectively leave open: gaps, unstudied populations, unexplored mechanisms, missing timeframes] ## Synthesis Statement [One dense paragraph integrating the above into a coherent research-landscape assessment — suitable for quoting in your own writing]
Result
Research topic: Does remote work reduce productivity?
Sources provided: Bloom & Liang (2023), Microsoft Work Trend Index (2025), Stanford WFH Research (2024), McKinsey Future of Work (2025), Chen et al. (2024)
Points of Strong Agreement
All five sources agree that remote work productivity outcomes are highly heterogeneous — dependent on role type, individual autonomy preference, home environment quality, and management approach (Bloom & Liang; Microsoft; Stanford; McKinsey; Chen et al.). Three sources explicitly note that productivity metrics used in remote work studies are inconsistently defined, making cross-study comparisons unreliable (Bloom & Liang; Stanford; Chen et al.).
Productive Tensions
The central empirical conflict is between Bloom & Liang (2023), who find a 9% productivity gain for hybrid workers over full-remote, and the Microsoft Work Trend Index (2025), which reports 87% of full-remote knowledge workers self-assess as highly productive. This is a methodological conflict, not a factual one: Bloom & Liang use manager-rated output metrics for customer service roles; Microsoft uses self-reported assessments across knowledge workers. The populations and measurement instruments are fundamentally different.
McKinsey and Stanford diverge on whether collaboration quality declines in remote settings. McKinsey finds measurable degradation in cross-team innovation projects; Stanford finds no measurable difference when controlling for meeting frequency. This is an empirical conflict that likely reflects different industries sampled.
Unique Contributions
- Chen et al.: Only source examining productivity variation by home office setup quality — finds ergonomic environment accounts for 23% of productivity variance
- Bloom & Liang: Only source with randomized control data rather than observational
- Microsoft Work Trend Index: Largest sample (31,000 workers, 31 countries), only source with multi-country data
- Stanford WFH: Only longitudinal data spanning 2020–2024 — tracks adaptation over time
- McKinsey: Only source distinguishing individual-task productivity from collaborative innovation productivity
Unanswered Questions
- How do productivity effects evolve after 3+ years of sustained remote work? (All studies capture ≤18 months post-adoption)
- Are findings consistent across labor markets outside the US, UK, and Australia?
- How does manager monitoring intensity interact with intrinsic motivation over time?
- What happens to long-term career progression for remote workers — absent from all five sources?
Synthesis Statement
The literature does not support a simple "remote work increases or decreases productivity" conclusion. The stronger finding, supported across all five sources, is that outcomes are highly contextual. Methodological choices — particularly whether to measure individual task output or collaborative outcomes, and whether to use self-report or manager ratings — largely predict the direction of findings. The most important unanswered question is longitudinal: nearly all existing data captures an adjustment period, not a steady state. Any organizational decision based on this literature must specify what type of productivity it cares about and for what worker population.
Most people approach a stack of research the same way: read source one, then source two, then source three — and end up holding whatever the last author argued most convincingly. That is not synthesis. That is sequential exposure.
Real synthesis requires a different kind of reading: mapping where sources converge, isolating where they genuinely disagree and why, identifying what each uniquely contributes, and — most importantly — recognizing what the entire body of literature still leaves unanswered. This prompt is designed to force exactly that analysis.
Why Each Section of the Prompt Exists
The output format has five sections, and each one is deliberate.
Points of Strong Agreement requires the AI to find claims supported by multiple sources — not just "most sources discuss X" but "these specific sources make this specific claim." This prevents false balance, where two outlier papers get equal weight to a consensus of six.
Productive Tensions is the most important section. The prompt asks the AI to classify disagreements by type: empirical (different findings from different data), methodological (different measurement approaches reaching different conclusions), or definitional (sources using the same word to mean different things). This distinction matters enormously. An empirical conflict suggests a real-world phenomenon worth investigating. A methodological conflict often means the studies are not actually measuring the same thing. A definitional conflict means you are comparing apples and staplers.
Unique Contributions prevents the common error of treating all sources as interchangeable. Every source in a good literature review was chosen for a reason — this section forces the AI to articulate that reason explicitly.
Unanswered Questions is what separates useful synthesis from mere summary. The gaps in a literature are often more actionable than its findings: they tell you where your own judgment must fill in, where your decision carries unquantified risk, and where more research is warranted before committing.
Synthesis Statement produces a single dense paragraph suitable for quoting or adapting directly into your own writing — a ready-to-use research-landscape characterization.
How to Get the Most Out of It
The prompt works best with 3 to 8 sources. Fewer than 3 and there is nothing to synthesize; more than 8 and even the best LLMs start losing track of which claim came from which source.
Paste actual abstracts or summaries — not just titles. The more specific the input, the more specific the synthesis. If you only have article titles and bullet-point notes, you will get bullet-point synthesis back.
The instruction to distinguish empirical, methodological, and definitional conflicts does real work. Leave it in. Without it, most LLMs will note that sources "disagree" without explaining why, which leaves you no better informed than before.
Compatible Models
This prompt works well with Claude Sonnet 4.6, GPT-5, and Gemini 3.5 Flash. For very long source texts (full papers rather than abstracts), Claude and GPT-5 handle the context more reliably. For quick synthesis of short summaries, Gemini 3.5 Flash is faster and sufficient.
When Not to Use It
This prompt is for synthesis, not research discovery. It works on sources you have already found and selected. It will not tell you what literature exists on a topic, identify seminal papers you should have included, or assess whether your source selection is representative. Use a search tool or a human subject-matter expert for that — then use this prompt to make sense of what you found.