5.4 — Synthesizing Across Sources (incl. Long-Form Papers)

A research paper isn't a stack of summaries. It's a conversation.

A common mistake on research papers is the "summary stack" structure. Paragraph 1: what Smith says. Paragraph 2: what Lopez says. Paragraph 3: what Chen says. Paragraph 4: what I think. The paper reads like a book report on three books.

What's missing is the move where you put the sources in conversation: where they agree, where they disagree, what gap one source leaves that the next one fills, what tension between them powers your own thesis. That move is the whole point of research. AI can accelerate it, IF you've already read the sources yourself. Do this lesson on top of 5.3, in that order.

Prerequisite: you've actually read the papers

Synthesis built on top of Claude's summaries of papers you didn't read is the version that produces vague, surface-level comparisons that fall apart under questioning. Synthesis built on top of your reading of the relevant 10% (Lesson 5.3) is the version that produces real arguments. Always do the reading first.

The Synthesis Matrix prompt.

Notice the constraint at the bottom: "from my notes only." That's the integrity guardrail. Without it, Claude will fill in plausible-sounding details about each paper from its general knowledge, including details that may not be in the actual paper. By limiting the matrix to your notes from your reading, the synthesis stays anchored to what's actually in the sources.

Prompt 1: The Synthesis Matrix prompt

I'm writing a paper on [specific question or thesis you're working with]. I've read [3–5] sources that are all relevant. Here are my notes on each one. These notes are mine, from my own reading of the actual papers: SOURCE 1: [Author, year, title]: - Main argument: [1 sentence in my words] - Method / approach: [1 sentence] - Key claim that's relevant to my thesis: [1–2 sentences with page reference if I have one] - What this source is NOT arguing or doesn't address: [1 sentence] SOURCE 2: [...] SOURCE 3: [...] [etc.] Build me a Synthesis Matrix: 1. **Where do these sources agree?** Quote whose claim aligns with whose. What's the consensus, if any? 2. **Where do they disagree?** Pinpoint the specific points of tension. Whose claim contradicts whose, and on what? 3. **What gaps exist between them?** What questions does Source 1 raise that Source 2 doesn't answer? What does Source 3 do that the others don't try to do? 4. **Which source is strongest on which dimension?** (Method? Evidence? Scope? Recency?) Be honest about weaknesses too. 5. **Three thesis angles** my paper could plausibly take that ARE supported by the tension or the gap between these sources: angles that don't just summarize the field but actually contribute something. Important: only build the matrix from the notes I gave you. Don't fill in details about the papers I didn't note. If you don't have what you need to compare on a dimension, say so and tell me what to go re-check.

What "agreement," "disagreement," and "gap" actually look like.

The matrix produces four flavors of useful comparison. Each one points to a different kind of paper.

🤝

Agreement

"Smith and Lopez both argue X, on different evidence." Useful as the establishing move of a paper. You're not contributing here, but you're showing the reader where consensus is. Thesis use: "While most scholars agree on X, the question of Y remains open." Agreement frames the launch pad.

⚔️

Disagreement

"Smith argues X; Chen argues NOT-X. The disagreement is over [specific point]." Thesis use: "I argue Smith is closer to right than Chen, because [the evidence Chen overlooks]." Pick a side, defend the side, address the strongest version of the other.

🕳️

Gap

"None of these sources address [specific question]. Smith focuses on the US case; Lopez focuses on the European case; nobody compares them directly." Thesis use: the most original kind of paper. "The question of Z hasn't been addressed; here's what we'd see if we asked it." Gap-driven theses are harder but more impressive.

🔬

Method tension

"Smith uses qualitative interviews; Chen uses regression. Their disagreement may be about method, not substance." Often the actual driver of contradictory findings. Thesis use: "Smith and Chen reach different conclusions because they're measuring different things. Once we [specific reframing], the apparent disagreement dissolves into..."

The "gap" is often the most original move: and the most fragile

Gap-driven theses ("nobody has asked this specific question") are the most impressive but require the most care. Sometimes the "gap" Claude surfaces is actually a well-trodden topic Claude just doesn't know about, or a question that's been answered elsewhere. Before you commit to a gap-driven thesis, do one extra search of the academic literature on the specific gap. If still nothing, you might really have a fresh angle.

Four follow-ups that turn a matrix into a paper.

The first matrix is the rough draft. These four follow-ups sharpen it, pick the angle, build the outline, and audit the result before you write a sentence of prose. Run them in sequence, in the same chat.

Prompt 2: Stress-test the matrix

Looking at the Synthesis Matrix you just built, push back on it. 1. Where might you be overstating an agreement that's actually surface-level? (Two sources can use the same word but mean different things.) 2. Where might you be overstating a disagreement that's actually about scope or method? (Two sources can disagree on the surface but actually be answering different questions.) 3. For each thesis angle you suggested, name the strongest counterargument a skeptical professor would raise. Don't soften your earlier matrix. Just stress-test it. I'd rather know the weaknesses now than after I've written 8 pages.

Prompt 3: Pick the most defensible angle

Of the three thesis angles you proposed, which one is most defensible given my actual sources and notes? Walk me through: 1. Which sources directly support this angle (and on what specific points). 2. Which sources I'd have to address as counterarguments (and what they'd say). 3. What additional evidence (if any) I'd want to find before committing. Don't pick the most ambitious angle. Pick the most defensible one given what I've actually read.

Prompt 4: Outline from the matrix

Based on the chosen thesis, draft a paper outline of [target word count]. For each section: 1. What argument the section makes. 2. Which of my sources I'd lean on (with a 1-line "this is what they let me say"). 3. What I still need to write or look up. Don't draft any prose. Just the outline, as bullets. I'll write the paper myself; this is the skeleton.

Prompt 5: Audit my synthesis before I start writing

Here's the thesis I'm going to argue, based on our matrix and our outline: [paste your draft thesis statement] And here's the outline: [paste outline] Run a final audit: 1. For each section, does my outline rely on a claim that's actually IN my source notes (which means I personally read it), or did I drift into a claim from your synthesis output that I haven't double-checked against the paper? 2. Flag any sentence in my thesis or outline where I should reopen the original PDF and confirm the claim is actually there before I write. 3. Flag any place where my argument might have flattened a real disagreement between sources for the sake of cleaner narrative. I'd rather catch this now than rewrite a section later.

The whole stack is one workflow, not a menu

These four follow-ups belong in sequence, in the same chat. Stress-test the matrix → pick the most defensible thesis → outline from that thesis → audit before writing. Each step depends on the last. The result is a paper outline grounded in the tension between sources you actually read, with the counterarguments named in advance. That outline is what you sit down to write from in Module 4.

Scaling up: long-form research papers and theses.

Everything above works for a 6–10 page research paper with 3–5 sources. What changes when the assignment is a 20-page seminar paper, a 40-page senior thesis, or a literature review chapter that needs to put 15+ sources in conversation? Three things change. The matrix scales differently, the integrity stakes rise, and the work spreads across weeks instead of days.

How the matrix scales: work in tiers, not in one giant chat

Tier 1: per-section matrices. Group your sources by the section of the paper they belong to (background / methodology / case A / case B / counterarguments). Build a small matrix for each cluster. 3–5 sources per matrix is the sweet spot. Beyond that, Claude starts producing surface-level comparisons that miss the specific tensions.
Tier 2: the master matrix. Once the per-section matrices are built, run a higher-level pass that asks Claude to compare the section matrices to each other. What does the methodology tier disagree with the case-A tier about? That's where the through-line of a long paper lives.
Tier 3: the living doc. A long paper's source list grows over weeks. Keep your matrix as a Google Doc or Markdown file you update. When you find a new source, take notes from your reading, and add it to the relevant tier. The matrix isn't a one-shot output; it's the spine you update through the whole project.

The thesis-proposal stage IS a synthesis matrix exercise

If you're in the proposal stage of a senior thesis or capstone (the "find an angle, defend why it's worth doing" stage), the Synthesis Matrix prompt above is exactly the move you need, run on the 4–6 sources that already shaped your thinking. Tier 1 is the matrix; the "three thesis angles" output IS your proposal section. Bring the strongest of the three to your advisor with the matrix attached. They'll know quickly whether the gap is real, and the conversation will be more productive than any version where you walk in without it.

Why long-form integrity is harder

On a 6-page paper, if synthesis drifted away from your sources, you'd notice; there are only so many places it could drift. On a 25-page paper, drift compounds. A claim from Claude's synthesis output gets carried into your outline, then into a section draft, then quoted in your conclusion, and four weeks later you can't trace whether the original source actually said it. On long papers, flag any claim you can't trace to specific notes from your specific reading.

Living source-tracker

For long papers, keep a one-row-per-source spreadsheet alongside the matrix: author, title, where you accessed it, the 1-sentence main argument, and the section of your paper it lives in. Update it every time you read a new source. At submission, that tracker is your evidence trail if anyone ever asks where a citation came from.

The advisor checkpoint

If you have an advisor or thesis committee, share the master matrix with them at the outline stage. It's a much faster way for them to push back on your synthesis than reading your draft prose. Their objections become the counterarguments you address in the paper itself, exactly the move Prompt 2 was rehearsing.

The Coach Project home

If you have a Coach Project (Module 7) loaded with your syllabus and writing samples, that's the right home for a long-form synthesis chat. The Coach already knows your voice and your prof's framing; your matrix prompts get sharper there than in a fresh chat.

The line between synthesis-help and synthesis-replacement.

🤖 What Claude is doing (legit)

Comparing your notes against each other faster than you could in your head. Surfacing tensions you might miss reading sequentially. Suggesting thesis angles given the source landscape. Pressure-testing your matrix for false agreements / false disagreements. Translating raw notes into outline structure. Scaling the same logic across 15 sources for a long paper. None of this is "Claude wrote my paper." It's "Claude compared my notes."

✍️ What you're doing (the load-bearing parts)

Picking the sources. Reading the sources. Taking honest notes from your reading. Verifying every claim Claude makes against your actual notes. Picking which thesis angle to commit to. Writing the actual paper. Owning every quote and citation as one you've personally seen on the page. The synthesis Claude produces is only as good as the notes you fed it. The notes are the work.

Synthesis is yours; AI just speeds the comparison

Honest Work Code · Rule 2: your work survives scrutiny. The deepest research-paper integrity question isn't "did I cite this correctly?" It's "did I actually do the synthesis my paper claims to do?" A paper whose argument depends on a tension between three sources should be written by someone who actually saw that tension on the page, not someone who got handed a tidy comparison from an AI. The Synthesis Matrix is a powerful tool because it externalizes a comparison you could in principle do in your head, and it's an honest tool because the input to the matrix is your reading, in your words. Run it that way and the paper that comes out the other end is recognizably yours, even though Claude helped you see what you were looking at.

Up next: once you have a thesis, sharpen it before you commit.

The matrix gave you a defensible angle. Lesson 5.5 is the sparring partner: argue-the-opposite, the skeptical-professor prompt, and the two traps that turn good sparring into capitulation.

Continue to 5.5 → AI as a Sparring Partner