Where to Get Earnings Call Transcripts in Bulk

​

The options at a glance

​

Which source fits which job

• Research across thousands of companies in SQL: a Snowflake Marketplace listing. The transcripts are already parsed into a table, so the first query runs the day you subscribe. Fruit Stand’s listing has one row per speaker paragraph, so filtering to what the CFO said, or to the Q&A, is a WHERE clause.
• Global coverage or point-in-time history for a fund: S&P Global or LSEG. Both cover 10,000+ companies worldwide back to the early 2000s, and both can deliver into Snowflake. Expect an enterprise sales process and pricing to match.
• Other company events, not just earnings calls: Snowflake Curated Data. Its transcripts include investor days, annual meetings, conference presentations and M&A calls, usually within three hours of the event. The free tier is delayed three months, and history averages about three years per company.
• A personal project or prototype on files: ROIC.ai or Financial Modeling Prep. ROIC.ai’s Professional plan includes every available quarter and bulk CSV downloads, one gzipped file per fiscal quarter. Read the terms first: ROIC.ai lists commercial use only on its Enterprise plan, and FMP requires a separate license to display or redistribute its data.
• Training or testing an NLP model: a free Hugging Face dataset can be enough to start, within its license limits.

​

Files vs. a table: what “bulk” means in practice

1. Download and keep downloading. Pull every quarter’s file, then check for new and corrected transcripts each week. ROIC.ai’s bulk files carry an updated_at and a checksum for this, and their download links expire after an hour.
2. Parse the text. Split each transcript into speakers and paragraphs, and separate prepared remarks from the Q&A.
3. Line up fiscal periods. Companies’ fiscal quarters don’t match the calendar, so “Q1 2025” means different months for different companies. Keep the call date next to the fiscal period.
4. Join to your other data. Map each company to the tickers or identifiers your prices and fundamentals use.

select
    fiscal_year,
    fiscal_quarter,
    count(distinct symbol) as companies_mentioning_tariffs
from earning_call_transcripts_flat
where content ilike '%tariff%'
  and fiscal_year >= 2018
group by fiscal_year, fiscal_quarter
order by fiscal_year, fiscal_quarter;

​

Free datasets: useful, with limits

• Where the text came from. That dataset doesn’t name its source, and its company IDs are S&P Capital IQ identifiers. Transcripts are usually licensed content, and a free upload doesn’t transfer the original license.
• What the license allows. It’s marked MIT, but its card also says it’s for research and educational use only. That’s fine for a class project or a model experiment, not for a product or client work.

​

What to check before you buy

1. Coverage. Large caps only, or small caps too? US only, or global?
2. History. Does it start at a fixed year, or vary by company? Ten years of history for the S&P 500 isn’t ten years for small caps.
3. Structure. Are speakers labeled? Are prepared remarks and the Q&A separated? Plain text means building a parser first.
4. Freshness. How soon after a call does the transcript arrive, and are corrected transcripts re-delivered?
5. License. Can you use the text commercially, store it in a shared warehouse, and show excerpts to clients?
6. Delivery. Files, an API, or a table in the warehouse you already use?

​

F.A.Q.