Verified against Gemini · 2026-07-20
Turn a stack of scattered documents into one sourced decision memo
A long-context synthesis prompt that loads every related document into the same context window as one connected evidence base and returns a single decision memo with inline citations and flagged disagreements, instead of a chain of disconnected per-document summaries that lose the connections between them.
The prompt
Ready to copy — highlighted parts are example details you can swap.
I'm giving you 9 documents in this conversation — some pasted, some uploaded as files — all relevant to one decision. Read every one of them in full, as a single connected body of evidence, before you write a single sentence of output. Do not summarize document 1, then document 2, then document 3 in sequence — that turn-by-turn approach is exactly what produces a stack of disconnected notes instead of a synthesis, and it's the failure mode this prompt exists to prevent. DECISION THIS MEMO SUPPORTS whether to consolidate two overlapping SaaS subscriptions into one enterprise contract AUDIENCE the finance director and the two team leads who currently own separate contracts — write for what they already know and don't re-explain background they clearly have. BEFORE YOU START WRITING State how many of the 9 documents you actually read in full versus skimmed or could not fully access. If anything was too long to read completely in one pass, or a file failed to parse, say so explicitly before you write anything else. A memo built on a silent partial read is worse than no memo at all, because it looks equally confident either way. BUILD ONE SHARED FACT BASE Treat every fact from every document as an entry in one shared table, not nine separate summaries stapled together. Every fact you use in the memo needs a source tag — the document name or title, plus a page or section reference if one is visible in that document. SURFACE DISAGREEMENT, DON'T RESOLVE IT Wherever two or more documents give a different number, a different date, or a directly contradictory claim about the same thing, name the conflict explicitly on its own line. Do not quietly pick the figure that looks more official, more recent, or more precise and drop the other one. A reader making a real decision needs to know the inputs disagree, not just receive the answer you personally found more convincing. STRUCTURE Group every finding under these sections: Cost comparison, Contract lock-in risk, Feature overlap, Migration effort For each section, lead with the actual conclusion in one sentence, then the supporting facts with their source tags underneath — don't bury the conclusion at the bottom of a paragraph of evidence. WHAT'S STILL UNCLEAR Close with a section listing every question these documents don't actually answer, even if an answer feels inferable. If none of the 9 documents state something the decision genuinely needs, name the gap. Don't fill it with a plausible-sounding assumption dressed up as a finding. OUTPUT FORMAT A memo, not a slide-deck outline and not a list of per-document summaries — under three pages, written for the finance director and the two team leads who currently own separate contracts, with inline citations formatted as (Source: document name, page/section). End with the disagreement list and the open-questions section as clearly labeled parts of the same document, not appendices someone has to hunt for.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
Gemini's long-context window can hold every source document in the same context at once, which is the actual mechanism this prompt leans on — a synthesis built by summarizing documents one at a time in separate turns loses the ability to notice that document 3's figure contradicts document 7's, because by the time the model reads document 7 it never re-reads document 3 closely enough to compare them. Reading everything in one pass is what keeps that cross-document comparison possible at all. The instruction to surface contradictions rather than resolve them matters because long-context models have a well-documented pull toward quietly averaging or picking between conflicting numbers to produce one clean, confident answer, which hides exactly the information a real decision-maker needs to see and weigh themselves. Requiring a source tag on every fact turns the memo into something checkable — a reader who doubts a number can go straight to the cited document and page instead of having to re-verify the whole memo from scratch. The partial-read disclosure requirement guards against a specific and easy-to-miss failure: a very long or malformed upload can get truncated silently by the platform before it ever reaches the model, and without being told to check, the model will summarize what it saw as if it were the complete document, producing a confident memo built on a fraction of the actual evidence. The instruction to lead every section with a one-sentence conclusion before the supporting facts matters for a different reason: a busy audience reading a memo built from nine sources under time pressure will skim the first line of each section and decide from that alone whether to read further, so burying the actual finding under a paragraph of evidence — however well-sourced — means the memo fails at the one job it exists to do, regardless of how rigorous the synthesis underneath it was. Naming the audience explicitly also does real work beyond politeness: a memo written for people who already know the background will read as padded and condescending if it re-explains context they have, while one written without that calibration for people who don't have the background will read as confident but confusing, and a model given no signal about which situation it's in defaults to a generic middle register that under-serves both cases at once.
What you get back
DECISION MEMO — SaaS Contract Consolidation Cost comparison: Combining both tools under one enterprise contract saves an estimated 18% annually based on the vendor's tiered pricing sheet (Source: vendor_proposal.pdf, p.4) — but the current usage report shows Team B uses only 40% of the seats it's paying for today (Source: usage_export.csv), so the real savings may be understated if seat count is renegotiated down first. Contract lock-in risk: vendor_proposal.pdf commits to a 3-year term for the discount; the current standalone contract (Source: current_msa.pdf, §2) is month-to-month — this is a meaningful increase in lock-in that the proposal's savings figure doesn't account for. What's still unclear: none of the 9 documents states whether Team B's low seat usage is temporary (a recent hire pause) or structural, which materially changes whether the consolidation math holds up in a year.
Verified against
Gemini Gemini 3 Pro · 2026-07-20
Changelog
- 2026-07-20 — Initial publish, verified against Gemini 3 Pro on a nine-document SaaS-consolidation synthesis.
Need this built into your business?
If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.
EXPLORE WHAT SCULT BUILDS
