Research tools
Insight Extraction guides.insight-extraction.hero.title_accent
It doesn't summarise your source. It maps the argument inside it — the claims, the evidence, and the disagreements — so you can start thinking instead of re-reading.
When to reach for it
You've uploaded a dense source — an academic paper, an exhibition catalogue, an artist's statement, a collection of press — and you need to know what's actually arguable in it before you can position your own reading. Extraction is the step between reading and writing: it gives you the source's structure so you're arguing with it, not paraphrasing it.
Select the materials to analyse
Open a project and choose one or more materials from your library. Extraction reads what you've already uploaded and indexed — so this is a step you take after a document is in, not a fresh upload. Select a single source to read it closely, or several to have the AI reason across them at once.
Selecting research materials for extraction
Screenshot
Selecting several related sources is where this earns its keep — the AI cites across all of them and flags where they disagree.
Run the extraction
Trigger Insight Extraction. It doesn't skim the first page — it retrieves the indexed content of each source and reads across it, then returns structured insights. On a typical source this takes a few seconds. You don't set a number; the AI returns as many substantive insights as the material actually supports, not a padded list to hit a quota.
Extraction running and returning insights
Screen recording (~15s)
Read what came back — including the disagreements
Each insight is a claim, not a bullet point. Under it you'll find the arguments that support it, each with exact quotes and a link back to the source material, plus counter-positions the AI found — labelled by how they relate: contradicts, complicates, extends, or reframes. Where an insight sits inside a theoretical tradition, or opens a question rather than answering one, it says so. This is the part that saves the hours: the tensions are already surfaced.
An expanded insight with arguments, citations, and counter-positions
Screenshot
The counter-positions aren't errors to reconcile — they're the productive friction. A source that only agrees with itself isn't giving you much to write about.
Keep what you want, take it further
Save the insights worth keeping to the project. From there they feed the rest of your workflow: map how they connect, compare them against another source, or pull them into a synthesis matrix you can export. Extraction is the intake; it's meant to hand off.
Saving an insight and the next-step actions
Screenshot
What it will — and won’t — do
It handles the mechanical read: finding the claims, pulling the quotes, tracking who's cited, noticing where a source contradicts itself. What it doesn't do is decide what any of it means for your exhibition. The interpretation is still yours — extraction just gets you to the part where interpreting is the only thing left to do.
Extraction on a typical source costs roughly 3,000 tokens — a few cents. Analysing several sources at once uses proportionally more, and is usually worth it.
What comes next
Put the insights to work
Connection Mapping
5 minDiscover relationships between your materials with interactive graph visualization
Research Synthesis
5 minGenerate structured data matrices across multiple sources — exportable to CSV/JSON
Comparative Analysis
5 minSide-by-side source comparison with contradiction scoring and common ground detection
Stop re-reading.
Start arguing.
Upload a source and see the argument inside it. Curatorial is free to start.
Try it free