Research a topic question by question
Turns a topic into the questions worth asking, searches each one, and files the findings in Notion with the source behind every line.
What it does
The integration steps this workflow runs, in the order it first runs them.
- 1AnthropicGenerate model response
- 2Jina AISearch web
- 3NotionCreate page
- 4NotionAppend block children
How it works
Everything the template sets up, and what to fill in before the first run.
Give it a topic and it works out the questions that topic actually raises, searches for each one separately, and files what it found as a page in your Notion database: the questions it asked, one finding per line, and the numbered list of pages every line came from.
Nothing reaches the page unless a page behind it was fetched first. The numbered extracts handed to the model are built from the search results themselves, and each line the model writes back has to cite one of those numbers - a line citing a number that was never fetched is discarded, and so is any line containing a web address, because the only addresses on the finished page are the ones copied out of the search results. At most one line is kept per extract, so a page cannot read as more corroborated than the handful of sources behind it. If the results carry less than 200 characters of text between them the run stops and says so rather than asking the model to fill the gap, and if every line is discarded nothing is filed at all.
Three settings decide the bill. 'maxQuestions' caps how many questions are planned, 'resultsPerQuestion' how many results each one fetches, and 'maxSources' the total across the whole run - the searching stops the moment that total is reached. Each has a ceiling of its own that a typo cannot get past: 20 questions, 10 results apiece and 30 sources, which is also what keeps the page inside the number of blocks Notion will write in one go. The page ends with a line saying how many sources it was written from and how many lines were dropped, so a thin run cannot pass for a thorough one.
Set it up by typing the topic into 'researchTopic' and pasting your database's id into 'notionDatabaseId'. The page is titled with your topic under the database's title property, which is assumed to be called Name - change 'titlePropertyName' if yours is called something else. No other column is written, so this works with a database of any shape. Everything else goes in the page body, which is written once and never read back, so each run files a new page and nothing you already have is touched.
You May Also Be Interested In
Work out the questions worth asking, then answer themName a subject and it decides which questions are worth asking, then answers each one from a page it actually read.
Turn one document into a study guide, a briefing and a timelineReads one Google Doc and writes three new documents from it: a study guide, a ten-minute briefing and a dated timeline.Research data cleaning with a decision logCleans a research dataset by fixed rules and writes down every change, exclusion and unresolved value as it makes it.
Drill your vocabulary list in TelegramSends /quiz a word from your Google Sheet and marks your /answer against what the sheet says.
Start from a workflow that already works.
Add "Research a topic question by question" to your workspace, connect its apps, and make it yours. No credit card required.
Free plan available · No credit card required