Blog · 2026-09-16 · Matterfact Research

What a million podcast transcripts reveal

Podcast chatter follows prices rather than leading them, bearish pile-ons were followed by recoveries, and who is in the room changes the verdict. What a million transcripts show, and how strong the evidence is for each finding.

Podcasts talk about a stock after it moves, not before. Across 142 companies, discussion volume lines up with the same week’s price move and says almost nothing about the next one. If you are listening for a tip, you are listening too late.

That is an awkward finding for a company that sells podcast data, and we put it first on purpose. The useful signal is elsewhere: in who is talking, how one-sided the room gets, and which part of the argument moves first.

We analyzed a million podcast transcripts, and five findings stood out:

1. By the time a stock is trending on podcasts, the move has mostly happened. (Part 01)
2. After a burst of negative discussion, stocks beat the S&P 500 by an average of 5.6% over the next eight weeks. After a burst of positive discussion they went roughly nowhere. (Part 02)
3. In December 2024 the debate over MicroStrategy’s premium to net asset value fell to 26% bullish, while overall MicroStrategy tone still read 66%. The headline number followed later. (Part 03)
4. Tesla is 78% bullish on All-In and 9% on Pivot. And at Walmart and Netflix, the covering analysts sound more bullish than management does. (Part 04)
5. Salesforce’s insider episodes run 35 points more bullish than its analysts’ episodes, while its statements are among the least specific of any company we scored. (Part 05)

Not all five are equally proven. The first rests on every eligible company-week in the sample; the third rests on one company, found after the fact. We say which is which as we go.

The findings describe our September 2026 research sample. Most company-insider counts use the full study history; the closer participant comparisons begin in January 2024. Monthly charts stop in August 2026, the last complete month.

01: The talk follows the price

We compared weekly podcast discussion with weekly price moves for 142 companies with at least 300 discussion episodes, from January 2024 through September 11, 2026.

Discussion lines up most closely with the same week’s price move. The correlation between discussion volume and the size of that week’s move is +0.156. One week later it is +0.016, and two weeks later it is 0.000.

Even the same-week link is modest. A correlation of 0.156 means the two series share about 2% of their variation. The measure uses the size of the move, so discussion rises with a sharp move in either direction.

Bullish attention spikes show the timing most clearly. In the two weeks before the spike, those stocks were already 4.4% ahead of SPY, and in the spike week itself they gained another 5.7%. Over the following eight weeks they added 0.1%.

Spikes can be large. Warner Bros. Discovery’s discussion volume reached 5.9× its preceding six-month average in December 2025, Oracle 5.4× in September 2025, Circle Internet 5.3× in June 2025, and SpaceX 4.9× in June 2026.

A surge like that is a pointer to something that has already happened. It tells you where to look and who is explaining the move, not where the stock goes next.

Of everything in this study, this is the finding we are most confident in. The correlation uses every eligible company-week (not only the unusual ones) and measures attention against each company’s own previous eight-week average.

02: Bearish pile-ons were followed by recoveries

We measured average excess returns around two kinds of event: a spike in attention with a strongly bullish tone, and a spike with a strongly bearish one. Excess return is the company’s weekly price return minus SPY’s, which tracks the S&P 500.

Attention spikes require at least 15 episodes in the week, at least three times the prior eight-week average, and a baseline averaging at least three episodes. Bullish weeks are 60% or more bullish, bearish weeks 40% or less, with at least 15 directional statements. Each window includes only events with complete returns, so sample sizes differ between rows.

Both groups move sharply in the event week, in the direction the tone suggests. After that they split. Bullish spikes are followed by roughly flat returns. Bearish spikes are followed by a recovery that, over eight weeks, nearly offsets the event-week drop. That is consistent with a long-studied pattern in markets: investors tend to overreact to bad news more than to good.

Individual stocks show the same shape. Oracle’s discussion peaked at 88 episodes and 84% bullish in the week to September 12, 2025, the week before its price high. By December 2025 a comparable week of 72 episodes ran 24% bullish. Intel went the other way: its heaviest discussion week, 107 episodes in late August 2025, was just 18% bullish, and its price roughly doubled over the study period. In this sample, the most one-sided bullish weeks fell near local highs for Oracle, MicroStrategy and CoreWeave.

The page charts price alongside weekly discussion volume and tone for Oracle, Intel, UnitedHealth, CoreWeave, MicroStrategy and Disney. Prices are indexed to the first displayed week and are not market-adjusted.

A room that agrees completely is a reason to recheck a position.

It is also the result we are least sure of. Events repeat within a company and overlap across companies, so they aren’t independent observations, and we haven’t yet published event counts or confidence intervals. Returns are measured against SPY without adjusting for risk: beaten-down, volatile stocks tend to bounce harder in a rising market regardless of what podcasts said. Treat it as a hypothesis worth testing, not a trading rule. None of it establishes a causal or predictive relationship, and nothing here is investment advice.

03: The argument turns before the average

MicroStrategy is the most insider-tagged company in the study, with 682 insider-tagged episodes across 113 shows. That makes it the best place to see what the headline number hides.

Start with who is talking. Separating episodes by group changes the reading from near-unanimous to split:

Then look at what they are arguing about. Bitcoin purchases and holdings, the largest theme by volume, run 66% bullish. Preferred stock runs 66%, dilution and convertibles 71%, leverage and liquidation risk 54%, and the premium-to-NAV debate 40%.

The premium-to-NAV thread moved first. In December 2024 it fell to 26% bullish while overall MicroStrategy tone was still 66%. The overall reading declined later. By June 2026, MicroStrategy’s bullish share had dropped to 25%, from 83% over the preceding three months.

One explanation is that the weakest point in a bull case gets argued first, and the rest of the conversation catches up.

The same thing happens elsewhere at company level, and the moves can be far larger than the market-wide conversation suggests. Oracle’s bullish share fell from 61% over the preceding three months to 19% in December 2025. UnitedHealth went from 15% to 64% in August 2025, and Intel from 18% to 54% in September 2025. Each of those months has at least 150 bullish or bearish statements.

By comparison, the market-wide conversation barely moves. Its lowest reading from January 2024 through August 2026 was 38.9% bullish, in April 2025.

Themes drift too. AI and large language models went from 58.5% bullish in January 2024 to 36.8% in August 2026. Real estate moved from 39.1% to 49.0% over the same period.

If that holds more widely, the practical step is to follow the most contested sub-theme and each company on its own, rather than the average across them.

It may not hold. The MicroStrategy sequence is one company and one sub-theme, identified after the fact, and we don’t yet know how often a sub-theme turns before the headline across all companies, or how often it turns and the headline doesn’t follow. The company swings are well sampled. The theme series have no published monthly sample sizes, so treat the AI and real estate endpoints as indicative. Keyword themes can overlap, and monthly theme sentiment appears only where there are at least 15 directional statements. The MicroStrategy relative price index compares its share price with Bitcoin’s, rebased to January 2024; it does not measure net asset value.

04: Who is in the room changes the verdict

When a company insider is on the episode, statements about the company run about eight bullish for every bearish one. When the strongest voice is a journalist or commentator, bearish statements outnumber bullish ones.

We group company statements by the strongest participant relationship tagged on the episode. The group doesn’t tell us who said each sentence. An episode associated with an insider can also contain statements from a host or another guest.

Insiders being upbeat is expected. What matters more is where they are upbeat. Mad Money alone has 946 insider-tagged episodes, and company statements on the show are 84.4% bullish. Squawk on the Street (898) and Closing Bell (760) come next. Part of what looks like insider optimism is the venue: executives choose friendly shows. We haven’t yet measured insider tone with those shows removed.

Where management is more cautious than its analysts

Given that baseline, the more surprising cases are companies where the analysts sound more bullish than management does.

Each percentage uses all statements in that company’s participant group, including neutral and mixed ones. Each company has at least 200 statements in the insider group and 100 in the analyst group. At those minimum sizes, a gap needs to be roughly 12 points before chance becomes an unlikely explanation. Statements from the same episode aren’t independent, so the true bar is somewhat higher. Starbucks, Salesforce and Berkshire clear it easily, and Walmart, Netflix and Tesla clear it narrowly. Amazon and Eli Lilly don’t. The chart compares all 18 companies.

Walmart’s and Netflix’s management sounding more cautious than their own analysts is the kind of gap worth reading episode by episode. One reading is a management team deliberately keeping expectations low; another is analysts ahead of the business.

Other groups diverge as well. Meta’s former-insider episodes are 17% bullish and 38% bearish. Microsoft’s peer-executive episodes are 45% bullish and 8% bearish. The regulator group has only 36 to 86 statements per company, too few to read much into.

The show matters as much as the guest

Tesla’s bullish share is 78% on All-In and 9% on Pivot. This Week in Startups runs 81%, Bloomberg Surveillance 68%, Mad Money 57% and Decoder 23%. Readings with fewer than 40 bullish or bearish statements are left out, and statements can come from hosts or guests.

Show-level tone varies just as widely across all companies. Among shows with at least 3,000 bullish or bearish company statements since January 2024, Mad Money is 84.4% bullish, Schwab Network 71.1% and Bloomberg Intelligence 55.4%.

The same company can sound like two different stocks, depending on the show and the guest.

The gaps between participant groups rest on tens of thousands of statements, so their direction is well established; the company gaps hold for the largest ones, as above. The main limit is attribution. Groups are assigned by episode, not by speaker, which means a bullish host on an insider episode counts toward the insider group.

05: Bullish, but vague

Numbers, named metrics, dates and sources give you something to check. We score each statement from 0 to 3 for specificity. GameStop and Berkshire Hathaway average 2.06, compared with Salesforce at 1.69 and Adobe at 1.73. A higher score means more detail, not greater accuracy.

Put that next to the tone gaps in Part 04 and two companies with similar bullish gaps look very different. Berkshire’s insider episodes run 27 points more bullish than its analysts’ and come with some of the most specific statements in the study. Salesforce’s run 35 points more bullish and come with some of the least.

So a wide bullish gap isn’t a warning by itself. Enthusiasm with checkable numbers attached is one thing; enthusiasm without them is the kind to test against what the company has actually reported.

This is two companies, though, not a tested pattern. Each company in the specificity chart has at least 2,000 scored statements, from January 2024 onward, so the scores themselves are well sampled. The next step is to plot tone against specificity for every company, not just the 16 in the chart.

06: Where the insiders are, and aren’t

Apple is discussed in 32,577 episodes but has an insider tagged in just 239: about 0.7 insider episodes for every 100 discussion episodes. Goldman Sachs is discussed in fewer than a fifth as many episodes and has 487 insider-tagged ones, about 8.1 per 100.

For Apple, almost everything you hear is from outsiders. For Goldman Sachs, the company itself is often on the episode. Among the 60 most-discussed companies, 16.8% of Cisco’s discussion episodes fall in the current-insider, current-board-member or covering-analyst groups, compared with 1.2% for Nasdaq, 1.6% for Costco and 1.9% for Zillow. A low share doesn’t mean the rest of the commentary is uninformed.

Reach differs from frequency. MicroStrategy’s 682 insider-tagged episodes come from 113 shows, while Alphabet’s 600 come from 284 and Microsoft’s 648 from 272.

Between earnings calls, there is a steady supply. From January 2025 through August 2026, the sample averages 588 episodes tagged with a current company insider per month, covering 323 companies per month.

Changes in transcript availability affect these counts, so the increase doesn’t tell us how much the podcast market grew. Tags are automated and refer to roles at the episode’s air date. Our follow-up review found that some include employees below officer level, clips or mentions, so each result needs checking before it is treated as an executive interview. Repeated or syndicated audio means episode counts won’t equal unique interviews.

Technology dominates discussion, but insiders are spread more widely. Financial Services has 6,606 insider-tagged episodes against technology’s 8,371, and covers more companies, 322 against 285.

An episode discussing companies from several sectors counts in each, so shares don’t sum to 100%. Within financial services, capital markets accounts for 1,576 insider-tagged episodes and asset management for 1,309; auto manufacturers account for 1,219 and aerospace and defense for 952. In Healthcare, Eli Lilly has 111 company-insider episodes, Pfizer 109, Moderna 91 and CVS Health 61. Energy’s leaders are Chevron (80), Shell (60), Exxon Mobil (56) and Energy Fuels (33).

For some names, podcasts give you the company’s own voice between earnings calls. For others, Apple most of all, you are hearing people talk about it.

07: Most of the talk isn’t about companies

Only 33.3% of statements are about a named company. Industry subjects account for 30.3%, economic subjects 28.4%, and commodities or crypto 8.0%.

The tone split between the economy and individual assets is stark:

Some of that gap may be built into the subjects. A statement that inflation is rising or tariffs are hurting is naturally negative, so macro tone is best read as a measure of how worried the conversation is. Gold and Bitcoin are different: a bullish statement there is a view on the asset.

Volume and tone also diverge. Real estate has about 1.5× the statement volume of AI among the industry themes, with bullish shares of 44.7% and 47.4% respectively. The chart covers 34 themes, 12 economic, 12 industry and 10 commodity or crypto, and can be sorted by volume or bullish share. Themes are grouped by keyword matching over the full study period.

Two-thirds of the archive, in other words, is about something other than a single company. A question about bank funding, industrial demand or drug development can lead you to shows you don’t normally follow.

08: Building a listening list

Networks have distinct profiles. 55% of CNBC statements name a company, while 42% of Bloomberg statements concern economic themes.

Topic percentages count statements; participant percentages count sampled episodes. One episode can have both insider and analyst tags.

Shows with similar insider rates can cover very different ranges. Decoder has current-insider tags in 22% of episodes, spanning 61 companies; Masters of Scale has 20% across 77. BG2Pod and The Twenty Minute VC both run at 14%, but span five and 67 companies respectively.

Rates use the episodes sampled for each show and can include republished audio.

Co-discussion can point to related companies. Lyft and Uber appear together 66× as often as their individual frequencies would predict, and Eli Lilly and Novo Nordisk 47.7×. Other pairs include Paramount Skydance and Warner Bros. Discovery (79.9×), Chevron and Exxon Mobil (59.7×), and Ford and Rivian (23.5×). Each pair has at least 300 shared episodes; platform companies were excluded. Co-discussion alone doesn’t establish a business relationship.

For broad coverage, look at how many companies a show includes. For a specific business, search across shows, keeping in mind from Part 04 that the show you pick shapes the tone you hear.

What we would test next

Several findings above are strong enough to use and not yet strong enough to rely on. These are the tests that would settle them:

• Event counts, confidence intervals and risk-adjusted returns for the attention spikes in Part 02.
• Insider tone with the CNBC programs removed, to separate the insider effect from the venue effect.
• Speaker-level attribution, so statements are assigned to the person who made them rather than the episode.
• How often a sub-theme turns before the headline, across every company rather than one.
• Tone plotted against specificity for every company with enough statements.
• How much of the variation in tone comes from the show, and how much from who is on it.

Start with a company you know

Take a question from your research, find relevant episodes, and compare the evidence behind different views. A large archive lets you do that without choosing in advance which show will have the answer.

Explore podcast intelligence or book a demo.

Methodology and scope

We computed the analysis on September 15, 2026, and reviewed it in a follow-up audit. The sample covers December 4, 2006–September 15, 2026. Transcript availability determines what enters the sample, so it doesn’t represent the entire podcast market.

Periods. All-time tables retain the full supplied period. The separate annual chart starts in 2021, and 2026 is labeled year to date. Monthly exhibits start in January 2024 and exclude incomplete September 2026. The recent monthly averages use the 20 complete months from January 2025 through August 2026. Historical counts reflect transcript availability.

Company-insider episodes. Automated tags refer to roles at the episode’s air date. Board members and former officers have separate groups. Counts can include employees, clips or mentions and need checking before being treated as executive interviews. Some companies in the sample are private.

Company discussion. A company is a subject when an episode contains sustained discussion of it, rather than a passing mention. Sector episode counts overlap when several sectors are discussed. Executive-appearance and discussion counts are separate measures; their ratio does not establish an exact overlap between the two episode sets.

Statements and attribution. We group company statements by the strongest participant relationship tagged on an episode. The tag doesn’t tell us who said each sentence; quotations need checking against the transcript. Ambiguous company names and sponsor mentions can affect discussion counts.

Stance and specificity. Role percentages include neutral and mixed statements in their denominator. Monthly and sector bullish share excludes them, using bullish / (bullish + bearish). Specific statements contain a quantity or a named metric, period, or source; this is separate from stance.

Statistical significance. The noise band in Part 04 is an approximation at the minimum sample sizes, assuming independent statements. The event-study averages in Part 02 are descriptive and have not been tested for significance.

Excluded comparisons. Company-universe totals disagree across the source tables, so we leave out percentages that rely on those totals. Sector statement counts also differ from the overall company count and aren’t combined here. A follow-up analysis found same-month co-movement with returns, but hasn’t established a predictive relationship.

Source: Matterfact research, Sept 2026.