Create your first dataset

This example explores how companies discuss supply chains. You’ll run a query, inspect the matching sentences, and export one row per earnings call.

The queries, sentences, and numbers below are illustrative, not real dataset results.

1. Run a first query

For your own project, first write one sentence describing the concept you want to measure.

Open New search and enter:

supply chain

Choose your date range, optionally enter a Search name, and select Search corporate text. Running it uses one search from your allowance. When it completes, open the dataset.

Treat this first result as exploratory. The next step is to check what the query actually found.

2. Inspect and refine the matches

On the dataset page, open Evidence. Read the highlighted sentences and choose Speaker context when you need the surrounding text.

Suppose you want to study supply-chain disruption, rather than all supply-chain discussion. A sentence such as “We invested in supply chain software to improve reporting” matches the first query but does not describe disruption.

To require disruption language too, try this starting query:

supply chain AND (disruption OR disruptions OR delay OR delays OR shortage OR shortages)

AND requires both parts in the same sentence; OR lists alternatives. Include word forms explicitly, such as delay OR delays.

Choose Actions → Edit & rerun, revise the query, and run it again. This creates a new dataset and uses another search. Inspect the new matches for irrelevant results and missing language.

3. Understand the output

The main CSV contains one row per earnings call, identified by earningscallID. The exposure column counts sentences matching the query; nr_of_sentences is the total number of sentences in the call.

Here are three illustrative rows showing selected columns:

Alpha has 14 matching sentences out of 412. This counts discussion, not confirmed real-world disruption.

Keep zero rows. Gamma's call is in the sample with zero matches, not a missing call; see why zero rows are kept.

See measure definitions when you need Risk, sentiment, or more detail on how settings affect the counts.

4. Export and save your settings

Open Downloads → Firm panel data and choose Download CSV beside Metrics. The filename ends in _firmlevel.csv. Check its call identifiers, dates, measure columns, and zero rows.

Download Search configuration JSON where available. Keep it with the CSV and your review notes; use the research record to capture settings the file does not include.

You now have an exploratory dataset and a record of how you made it. Use the checks below before relying on it for research.

Before using this in research