Your first extraction
In about ten minutes you will write a three-item extraction schema, run it against one of your own contracts, and read the answers back with their sources. By the end you will have seen a found value, a source you can check, and the difference between an answer and a gap.
You need one contract already in your team's library — any signed agreement will do: a lease, a supply agreement, an engagement letter. If you have none yet, upload one first. Extractions must also be enabled for your team; if the Extractions sub-tab is not there, see Extraction schemas and results.
1. Open the Extractions page
Open the Workflows page and select the Extractions sub-tab. On the left is your list of schemas, empty if this is your first; on the right, the editor.
2. Create a schema
Select + beside the Extraction schemas heading and choose Blank schema from the menu. A window asks for a name: type "Key terms", leave the description empty, and select Create schema. The schema appears in the list and opens in the editor.
You do not need to save. "Saving…" appears briefly above the items whenever you pause typing.
3. Add three items
The three items below have an answer in almost any contract, so whatever document you chose, the run has something to find. The schema starts with one empty item: fill it in, then select the + button on the line beneath it to add the next, and again for the third.
| Label | What to extract | Extraction format |
|---|---|---|
| Parties | The full names of every party to the agreement | Free text |
| Date of agreement | The date the agreement was made, signed, or dated | Date |
| Governing law | The law that governs the agreement, e.g. England and Wales | Free text |
Notice that Date of agreement has the Date extraction format: the agent must return a real date for it, and the result will show it as a date.
4. Run it
- Select the run control on the schema's row in the list.
- Pick your contract from the document picker.
- Dispatch the run.
A message confirms the dispatch with a View link. The run works in the background; a count appears on the Activity button in the action bar while it does.
5. Read the results
Select the View link in the confirmation, or select Activity in the action bar and pick the run with the Extraction pill named "Key terms". Wait for the green tick, typically a minute or two, then select the run.
Its results open in their own window: three rows, one per item. Look at each in turn.
- Parties shows Found, with the names in Value and at least one entry under Sources. Select a source: the passage the agent drew the names from appears, with an Open document link that opens the contract. Select the row itself to see its details in full beside the table.
- Date of agreement shows Found, with the date written as day, month, and year — the extraction format at work.
- Governing law shows either Found, with the jurisdiction, or Not found if your contract has no governing-law clause. Not found is a real answer: the agent looked and is telling you the clause is absent. Read the Notes cell, which often says what the document has instead.
Beneath the rows you may also see Notes from the extracting agent — a short observation about the contract that none of your three items asked for. A simple contract often draws none, so do not worry if the section is absent. If you want a summary of the run in prose, produce one from the Reports section above the table: select Produce and pick Standard summary.
6. Check an answer against the contract
Select the source under Parties to read the quoted passage, then select Open document to open the contract. This is the habit that makes extractions useful: every value in the grid points at the words it came from, so checking one is a click rather than a search.
What you have done
You wrote a schema, ran it, and read a source-cited result. The same schema now runs against any other contract in one step — select its row's run control and pick the document — and the results gather in Activity under its View runs control.
From here:
- Add the items you actually check on every contract, with a What to extract note for each, and fix answer shapes where they matter — see Extract information from contracts.
- Let an agent draft a longer schema from a client's list of data points or from a few example contracts — see Draft a schema with AI.
- Keep a run's results as a document in your library with Save to documents — see Track agent runs.