upnext

Intelligence

Private beta

What did they saylast month?

Name a person. Get every podcast where they actually spoke, with the claims that matter pulled out and timestamped. Earnings calls are scraped by everyone. This is the primary source nobody is reading.

Request accessDemos run on the names you choose.

Weekly brief

1 tracked person · last 30 days

3 appearances

Bloomberg Talks

20 May

00:06:11Joint appearance with Dell's CEO on agentic deployments and what customers are actually buying.

Training Data

10 Jun

00:31:07On capacity constraints and the buildout timeline, unscripted and at length.

Squawk on the Street

02 Jun

00:04:52Named the next trillion-dollar company and explained the reasoning behind the pick.

every line links to the second it was said

103

episodes surfaced for one name in 120 days

96%

were about the person, not the person speaking

<$20

of coverage per analyst per month at 25 names

Measured on a public figure across the open catalog. The filtering is the product; the search is a commodity.

The problem

An executive on a podcast is unscripted for ninety minutes.

They describe strategy, pricing pressure, hiring plans and what they think of a competitor, in a format nobody indexes and nobody has time to listen to. It is the least contested source in your stack.

How it works

Four steps. The second one is the whole business.

01

Sweep the full catalog

Every public podcast, not a curated list. Executives turn up on industry shows nobody on your team subscribes to, which is exactly where the unguarded version lives.

02

Separate spoken from spoken-about

A search for any public figure returns mostly commentary, hot takes and auto-generated feeds. Our classifier decides whether the person is actually in the room. This is the hard part and the reason the output is usable.

03

Transcribe what survives

Only episodes that clear the filter get transcribed, which is what makes broad coverage affordable rather than theoretical.

04

Deliver something checkable

Show, date, timestamp on every line. An analyst can confirm any claim against the audio in one click, which is the only standard that holds up in a memo.

Worked example

One executive, 120 days, run end to end

We pointed the pipeline at a single well-covered CEO and looked at everything it returned. These are the actual results, including what it threw away.

42episodes matched the name
18were the executive actually speaking

The eighteen ran across Bloomberg Talks, Squawk on the Street, Bloomberg Tech, a Sequoia interview and a long-form policy show — several on feeds no research team subscribes to. Three separate CNBC segments in a single month each carried material the others did not.

Run the same pipeline on a more controversial name and the ratio inverts: 103 matches, 4 of them the person speaking. Ninety-six percent of what gets published under a public figure's name is other people talking about them.

Rejected, with the reason

Nvidia Gets Into the PC Market With New Chip

Hosts discuss the keynote. He never speaks.

Jensen on Dwarkesh, Cursor + xAI

A show discussing his appearance on another show.

20VC: Lessons from Jensen Huang on Founder Mode

Cited as an example. Not a guest.

Biography Flash: Jensen Huang on Suffering

Auto-generated feed. No human, no interview.

China Decode: Challenging Nvidia's Dominance

About the company and a competitor, not the person.

The run also caught the same interview syndicated across three network feeds on one day. Deduplicating it is the difference between a brief with three entries and one that looks padded.

The output

A brief an analyst can put in front of an IC

No summaries of summaries. Direct quotes, attributed, with the show, the date and the second they were said, because a misattributed quote is the one failure this product cannot have.

Weekly brief

1 tracked person · last 30 days

3 appearances

Bloomberg Talks

20 May

00:06:11Joint appearance with Dell's CEO on agentic deployments and what customers are actually buying.

Training Data

10 Jun

00:31:07On capacity constraints and the buildout timeline, unscripted and at length.

Squawk on the Street

02 Jun

00:04:52Named the next trillion-dollar company and explained the reasoning behind the pick.

every line links to the second it was said

Track people or companies

A watchlist of names, monitored continuously across the catalog.

Said-by and said-about

Choose whether you want their words or the market talking about them. Both run on the same sweep.

Weekly or on demand

A standing brief, or a one-off before a call or a diligence session.

Verifiable by design

Every line carries its source, so nothing has to be taken on trust.

Access

We are onboarding design partners.

Early briefs are produced for a small group of teams on the names they care about. Tell us who you would track and the first brief you see will be about them.

Also available as a consumer app for your own listening. See upnext

Request access

Three questions, so your demo runs on names you care about.