The failure mode for AI-assisted research is a memo that looks confident and has no idea what it's citing. This workflow builds in verification at each step — the model produces the draft, but you (or a review process) confirm every citation actually says what the brief claims it says. Result: a brief you can put your name on with a colleague, editor, or client, knowing that if someone clicks through to a citation, the source will support the claim.
Frame the question with search intent
Goal: Turn a vague topic into a specific answerable question with clear scope.
I want to research [topic]. Before I ask for a brief, help me frame the question tightly. Return: (1) the specific question a good brief on this topic would answer, (2) the scope — what's in-scope and out-of-scope, (3) three sub-questions that would need to be answered to answer the main one, (4) the type of sources that would count as authoritative for this topic (primary studies, regulatory filings, industry reports, etc.).
Gather sources with cited search
Goal: For each sub-question, find 2-3 authoritative sources and record what each actually says.
For each of the three sub-questions above, search the web with cited sources. Return, per sub-question: 2-3 authoritative sources with links, the specific passage in each source that addresses the sub-question (quoted), and a note on the source's date and any known limitations. Cite; do not paraphrase without a citation.
Verify each citation manually
Goal: Click every source link and confirm the passage says what LADLE quoted.
Before we draft the brief: I'll open each cited source above and confirm the quoted passage. Any citation that doesn't verify, I'll flag as 'unverified — replace' and you'll suggest an alternate source. Only verified citations enter the draft.
Handoff: This step is the difference between a brief that stands up and one that doesn't. Skipping it produces the AI-slop research memo everyone's been complaining about.
Draft the brief
Goal: One integrated memo with every claim traceable to a verified source.
Draft a 500-800 word brief answering the main question, integrating the three sub-questions and their verified sources. Structure: TL;DR (3 sentences), background (150 words), the three sub-answers (150 words each), practical implications for small AI startups (100 words), sources cited inline (Source 1, Source 2…) with a source list at the bottom.
Repeatability, not cleverness.
What lives where when the workflow finishes.
Verified brief → your publication surface (blog, internal memo, client email). The verified source list stays with the brief permanently — if a reader clicks through to a source, the source supports the claim. Add the brief to a LADLE Project called 'Research library' with the topic name; you'll want to reference it in six months.
The same shape, different jobs.
- Legal-brief variation — Step 3 verification includes a lawyer's read before Step 4 drafts; brief is a legal memo not a policy analysis.
- Investment-memo variation — Sub-questions map to thesis/team/market/traction; sources include public filings and cap-table data.
- Academic literature review — Step 2 gathers 20 sources not 6-9; Step 4 becomes a 3,000-word review, not an 800-word brief.
- Quick-brief variation — collapse Steps 2-3 for casual research where the stakes don't justify manual verification; skip Step 3 for anything you'd publish.
Prompts: find the original source · fact check a claim · compare two papers
Try this workflow once. Then decide if it earns a recurring slot.
The value compounds after the third run, not the first.