How to Query Earnings Call Transcripts in Snowflake SQL

By Fruit Stand Team · November 3, 2026 · Guides To query earnings call transcripts in SQL, store them one paragraph per row with the company, fiscal quarter, speaker and paragraph number, then filter and sort on those columns. In Fruit Stand’s Earnings Call Transcripts listing on Snowflake, Apple’s July 30, 2026 call comes back as 108 rows from 12 speakers, ready to read in order or split into prepared remarks and Q&A. This post is about working with transcripts once they’re in Snowflake. If you don’t have them yet, Where to Get Earnings Call Transcripts in Bulk compares the sources, from free datasets and APIs to Snowflake Marketplace listings. Below are the queries to list a company’s calls, read one in order, split the prepared remarks from the Q&A, and find which analysts asked questions. They use Apple, but work for any ticker.

List a company’s calls

select fiscal_year, fiscal_quarter, report_date,
       count(*) as paragraphs,
       count(distinct speaker) as speakers
from earning_call_transcripts_flat_vw
where symbol = 'AAPL'
group by 1, 2, 3
order by report_date desc;

For Apple this returns every call since fiscal Q4 2005, reported in October 2005. The three most recent, as of November 4, 2026:

Fiscal year	Quarter	Call date	Paragraphs	Speakers
2026	4	Nov 2, 2026	122	15
2026	3	Jul 30, 2026	108	12
2026	2	Apr 30, 2026	52	13

Fiscal quarters follow the company’s own calendar. Apple’s fiscal year ends in September, so its “Q3 2026” call covered April to June. Use report_date when you need calendar time. The examples below use the July call.

Read one call in order

select paragraph_number, speaker, content
from earning_call_transcripts_flat_vw
where symbol = 'AAPL'
  and fiscal_year = 2026
  and fiscal_quarter = 3
order by paragraph_number;

speaker holds the person’s name, such as the CEO, the CFO, the investor relations host, the operator and each analyst. To rebuild the full text as one document, aggregate the paragraphs in order:

select listagg(speaker || ': ' || content, '\n\n')
         within group (order by paragraph_number) as transcript
from earning_call_transcripts_flat_vw
where symbol = 'AAPL' and fiscal_year = 2026 and fiscal_quarter = 3;

Split prepared remarks from the Q&A

Calls open with prepared remarks from management, then the operator opens the line for questions. The first paragraph spoken by Operator marks the switch:

with call as (
    select *
    from earning_call_transcripts_flat_vw
    where symbol = 'AAPL' and fiscal_year = 2026 and fiscal_quarter = 3
),
qa as (
    select min(paragraph_number) as qa_start
    from call
    where speaker = 'Operator'
)
select
    iff(c.paragraph_number < q.qa_start, 'Prepared remarks', 'Q&A') as section,
    count(*) as paragraphs,
    sum(length(c.content)) as characters,
    count(distinct c.speaker) as speakers
from call c
cross join qa q
group by 1
order by 1;

For Apple’s July 2026 call:

Section	Paragraphs	Characters	Speakers
Prepared remarks	32	24,488	3
Q&A	76	23,908	12

The prepared remarks came from three people (the investor relations host, the CEO and the CFO). The Q&A was about as long, with far more voices. This split works for most calls: 95% of calls since January 2025 in the listing have an Operator paragraph. For the rest, look for the host’s handoff, such as a paragraph mentioning “questions”.

Find which analysts asked questions

Each analyst speaks right after the operator introduces them, so the speakers who follow an Operator paragraph are the analysts:

with ordered as (
    select paragraph_number, speaker,
           lag(speaker) over (order by paragraph_number) as previous_speaker
    from earning_call_transcripts_flat_vw
    where symbol = 'AAPL' and fiscal_year = 2026 and fiscal_quarter = 3
)
select distinct speaker
from ordered
where previous_speaker = 'Operator'
  and speaker <> 'Operator'
order by speaker;

On Apple’s July 2026 call, that returns the seven analysts who asked questions. Run it across a company’s last eight calls to see which analysts cover it most closely.

The free option in Snowflake

Snowflake’s free public data share (Snowflake Public Data, Free) also has earnings call transcripts, in company_event_transcript_attributes. Each row is a whole event, with the transcript as JSON. Its speaker_mapping gives each speaker’s name, role and company, so analysts come labeled with their firm. To get the same paragraph-per-row shape as above, flatten it:

select p.index as paragraph_number,
       s.value:speaker_data:name::string as speaker,
       s.value:speaker_data:role::string as role,
       s.value:speaker_data:company::string as company,
       p.value:text::string as content
from snowflake_public_data_free.public_data_free.company_event_transcript_attributes t,
     lateral flatten(t.transcript:paragraphs) p,
     lateral flatten(t.transcript:speaker_mapping) s
where t.primary_ticker = 'AAPL'
  and t.event_type = 'Earnings Call'
  and t.transcript_type = 'SPEAKERS_ANNOTATED'
  and t.fiscal_year = '2026' and t.fiscal_period = 'Q3'
  and s.value:speaker = p.value:speaker
qualify row_number() over (partition by p.index order by t.updated_at desc) = 1
order by p.index;

For Apple’s July call it returns the same 108 paragraphs from 12 speakers as the queries above. Checked on November 4, 2026:

It lags. The latest event was August 5, 2026, so Apple’s November 2 call isn’t there yet.
Duplicates. Each call appears as a RAW and a SPEAKERS_ANNOTATED transcript, and sometimes more than once, so pick one type and the latest row, as the qualify does.
Roles included. Speaker roles and firms are the one thing the free data gives you that our listing doesn’t.

For how it compares with other sources on coverage and price, see Where to Get Earnings Call Transcripts in Bulk.

Things to know

Speaker names, not roles. speaker is the person’s name. To label roles, map names to titles for the company’s executives, or use the operator pattern above to tell analysts from management.
Fiscal vs. calendar quarters differ by company, as above. Compare companies by report_date, not fiscal_quarter.
Upcoming calls. A call can appear shortly before or after its date as new transcripts are added weekly. Filter on report_date <= current_date() for completed calls only.

F.A.Q.

How far back do earnings call transcripts go?

The listing has transcripts since 2005 for 4,500+ companies. Apple’s start in October 2005; coverage for smaller companies often starts later.

Can I search across all companies' calls?

Yes. Drop the ticker filter and search content with ilike, or use Snowflake Cortex for semantic search. Filter by report_date to limit the time range.

How quickly are new calls added?

The listing refreshes weekly, adding new and updated transcripts.

Can I use the transcripts to train a model?

Yes, with Fruit Stand’s Earnings Call Transcripts listing you can use the transcripts to train models.

To query any company’s calls in SQL, start a free 30-day trial of the Earnings Call Transcripts listing on the Snowflake Marketplace. The schema is in the dataset docs. For other places to get transcripts, see Where to Get Earnings Call Transcripts in Bulk. Questions? Reach out at contact@fruitstand.dev.

Home

Datasets

Company

How to Query Earnings Call Transcripts in Snowflake SQL

List a company’s calls

Read one call in order

Split prepared remarks from the Q&A

Find which analysts asked questions

The free option in Snowflake

Things to know

F.A.Q.

​List a company’s calls

​Read one call in order

​Split prepared remarks from the Q&A

​Find which analysts asked questions

​The free option in Snowflake

​Things to know

​F.A.Q.

List a company’s calls

Read one call in order

Split prepared remarks from the Q&A

Find which analysts asked questions

The free option in Snowflake

Things to know

F.A.Q.