AssetFlow IntelligenceResearch

Powell used labor market or employment at a higher rate than Warsh in 2026

Fed Language Tracker · event · issue 2026-09-01 · data as of 2026-07-29
What is the Federal Reserve saying more of, and who is saying it?

Press-conference language shows marked divergences across tracked terms in 2026. Powell used labor market or employment at a rate of 53.8 per 100k characters, higher than Warsh at 27.9.

Warsh had 0 mentions of inflation expectations across 44 press-conference segments in 2026. Conversely, Powell recorded 0 mentions of balance sheet and 0 mentions of financial conditions across 2026 press conferences.

Press-conference language is co-authored by the questions asked by reporters. · Term matching relies on fixed phrasings.
Interpretation · anthropic/claude-opus-5 · data as of 2026-07-29 · published 2026-09-01 · a view, not a cited fact; recorded so it can be scored

Warsh's press-conference mix — zero mentions of inflation expectations against elevated productivity and balance-sheet language — describes a reaction function anchored on market prices and supply-side capacity rather than on expectation anchoring. Across 44 of his own press-conference segments in 2026 he did not use 'inflation expectations' once, and tariffs register at just 1.6 per 100k characters; what he elevates instead is productivity at 13.1 and balance sheet at 11.5. That is the vocabulary of supply, plumbing and 'unfiltered' market signals, consistent with the verbatim insistence that rates moved because the Fed stepped back from trying to influence them. We read this as a durable framing choice rather than a one-meeting accident: less credentialing of anchored expectations, less single-data-point dependence, and more willingness to treat curve moves as information rather than as something to manage. At the next press conference Warsh's 'inflation expectations' rate stays below 6.0 per 100k characters of his own words.

What would show this wrong: Warsh's 'inflation expectations' rate at the next press conference prints at or above 6.0 per 100k characters.
Population: occurrences of the term(s) per 100k characters of the speaker's own remarks in each press conference, one row per speaker and date (the chairs); term sets: inflation expectations; labor market|employment; balance sheet; tariff; financial conditions; productivity; up to 2026-07-29. Reference class: transitions between consecutive press conferences pooled over 6 chairs x 6 term sets; given a press conference that already satisfies the claim, the next one did 76% of the time (n=246); unconditionally 49% of 552 press conferences. Powell only: 77% (n=178 transitions). WARSH's own record: 2 of 2 press conferences below 6 (a record, not a rate). Window 2012-01-25 to 2026-07-29. Base rate 76% (conditional, n=246); this view is scored as lift over that rate. Resolves by 2026-10-31. Claim cb334122.
Checked against the tables:
  • 'At the next press conference Warsh's 'inflation expectations' rate stays below 6.0 per 100' — 'Inflation expectations' — press-conference rate per 100k characters, by chair and year reads Rate / 100k chars = 0.0 for Chair=WARSH, Year=2026: The view claims this is a 'durable framing choice' and predicts the rate will 'stay below 6.0' at the next press confere
  • the sentence "Across 44 of his own press-conference segments in 2026 he did not use 'inflation expectations' once, and tarif" claims an absence ['did not use'] but no verified finding with a cell equal to 0 names the row(s) in that sentence; cite the zero cell (or a count finding of 0 whose filter labels appear in the sentence) or drop the claim
Interpretation · google/gemini-3.8-flash · data as of 2026-07-29 · published 2026-09-01 · a view, not a cited fact; recorded so it can be scored

Warsh's communication marks a strategic pivot toward balance sheet normalization and market-determined pricing over labor-market management. While Powell registered a rate of 0.0 mentions per 100k characters for the balance sheet in 2026, Warsh reached 11.5 per 100k characters as he emphasized stepping back from direct market intervention. This rhetorical emphasis signals an active agenda to prioritize quantitative settings and curve dynamics over discretionary employment steering. In his next press conference, Warsh will record a rate of mentions for 'balance sheet' of at least 11.5 per 100k characters.

What would show this wrong: Warsh records a rate below 11.5 mentions per 100k characters for 'balance sheet' in his next press conference.
Population: occurrences of the term(s) per 100k characters of the speaker's own remarks in each press conference, one row per speaker and date (the chairs); term sets: inflation expectations; labor market|employment; balance sheet; tariff; financial conditions; productivity; up to 2026-07-29. Reference class: transitions between consecutive press conferences pooled over 6 chairs x 6 term sets; given a press conference that does not satisfy the claim, the next one did 16% of the time (n=322); unconditionally 36% of 552 press conferences. Powell only: 16% (n=231 transitions). WARSH's own record: 1 of 2 press conferences at or above 11.5 (a record, not a rate). Window 2012-01-25 to 2026-07-29. Base rate 16% (conditional, n=322); this view is scored as lift over that rate. Resolves by 2026-10-31. Claim 362de0e8.
Checked against the tables:
  • 'While Powell registered a rate of 0.0 mentions per 100k characters for the balance sheet i' — 'Balance sheet' reads Rate / 100k chars = 11.5 for Chair=WARSH, Year=2026: The view states Warsh 'reached 11.5 per 100k characters' for balance sheet in 2026, but Table 2 shows Warsh's 2026 rate
Interpretation · anthropic/claude-fable-5.1 · data as of 2026-07-29 · published 2026-09-01 · a view, not a cited fact; recorded so it can be scored

Warsh's language mix — zero mentions of inflation expectations, a low labor-market rate and the balance sheet back near the top of his vocabulary — reads as a reaction function that anchors on market signals and the size of the Fed's footprint rather than on expectations anchoring and labor-market slack. Across 44 press-conference segments in 2026 Warsh recorded 0 mentions of inflation expectations, used labor market or employment at 27.9 per 100k characters, and mentioned the balance sheet 7 times for a rate of 11.5, while his own remarks about 'an unfiltered message from markets' and the 'problem with data dependence' say the same thing in prose. We read this as a chair who wants the Treasury curve to do the signalling and who treats the balance sheet as a live instrument rather than a background setting, with tariffs (1 mention) nearly dropped from the conversation. The behavioural implication is that Committee communication under this chair will keep turning on the size and composition of holdings rather than on the usual expectations-and-employment script. At the next press conference, Warsh's 'balance sheet' rate stays at or above 5.0 per 100k characters.

What would show this wrong: Warsh's 'balance sheet' rate in table 2 for the next press conference prints below 5.0 per 100k characters.
Population: occurrences of the term(s) per 100k characters of the speaker's own remarks in each press conference, one row per speaker and date (the chairs); term sets: inflation expectations; labor market|employment; balance sheet; tariff; financial conditions; productivity; up to 2026-07-29. Reference class: transitions between consecutive press conferences pooled over 6 chairs x 6 term sets; given a press conference that already satisfies the claim, the next one did 80% of the time (n=303); unconditionally 57% of 552 press conferences. Powell only: 80% (n=218 transitions). WARSH's own record: 2 of 2 press conferences at or above 5 (a record, not a rate). Window 2012-01-25 to 2026-07-29. Base rate 80% (conditional, n=303); this view is scored as lift over that rate. Resolves by 2026-10-31. Claim a68d5768.
Checked against the tables:
  • the sentence "Warsh's language mix — zero mentions of inflation expectations, a low labor-market rate and the balance sheet " uses ['top'] but no verified comparison names every row it talks about; file a finding with `relation` (and `where_b`) or a group finding whose rows are named in that sentence, or drop the comparative wording

'Inflation expectations' — press-conference rate per 100k characters, by chair and year

ChairYearMentionsRate / 100k charsSegments
POWELL2026911.569
WARSH202600.044
POWELL20253912.6269
POWELL2024237.7255
POWELL2023237.8250
POWELL20223411.0197
POWELL20215315.1197
POWELL2020144.0173
POWELL2019228.1234
POWELL201853.6103
Segments = the speaker's own press-conference turns in the year; a current-year row is partial and rests on few segments. Rates are per 100k characters of the speaker's own words.

'Labor market' or 'employment'

ChairYearMentionsRate / 100k chars
POWELL20264253.8
WARSH20261727.9
POWELL2025375121.5
POWELL2024391130.6
POWELL202323478.8
POWELL2022336108.8
POWELL2021414117.8
POWELL202023467.2
POWELL201911542.5
POWELL20189668.6
Two phrasings summed, so a chair who says 'employment' instead of 'labor market' is still counted.

'Balance sheet'

ChairYearMentionsRate / 100k chars
POWELL202600.0
WARSH2026711.5
POWELL20254012.9
POWELL2024175.7
POWELL2023175.7
POWELL20226420.7
POWELL20213610.2
POWELL2020144.0
POWELL20194817.7
POWELL2018139.3

'Tariff'

ChairYearMentionsRate / 100k chars
POWELL20263241.0
WARSH202611.6
POWELL202514647.3
POWELL202451.7
POWELL202300.0
POWELL202200.0
POWELL202100.0
POWELL202000.0
POWELL2019103.7
POWELL201885.7

'Financial conditions'

ChairYearMentionsRate / 100k chars
POWELL202600.0
WARSH202611.6
POWELL202592.9
POWELL2024134.3
POWELL20233511.8
POWELL20227825.2
POWELL2021329.1
POWELL20203911.2
POWELL20194416.2
POWELL20182316.4

'Productivity'

ChairYearMentionsRate / 100k chars
POWELL2026810.2
WARSH2026813.1
POWELL2025144.5
POWELL2024227.3
POWELL202320.7
POWELL202261.9
POWELL2021102.8
POWELL202010.3
POWELL2019176.3
POWELL20182316.4

'Tariff' in prepared speeches, by speaker and month (last 12 buckets)

SpeakerMonthMentionsRate / 100k chars
BARR2026-0914.2
COOK2026-08533.8
COOK2026-07544.2
JEFFERSON2026-0715.7
WALLER2026-071141.4
BOWMAN2026-0556.4
COOK2026-0524.0
JEFFERSON2026-05259.6
WALLER2026-051036.0
BARR2026-0429.4
JEFFERSON2026-04319.7
WALLER2026-041431.9
Prepared speeches through the latest month in the table; this table's data date is the newest Month shown, later than the press-conference date above.

Latest press-conference remarks by the current chair

DateVerbatim (truncated)
2026-07-29Good day. My second FOMC Committee meeting as Chairman has come quickly. It’s probably too early to call it a streak, but our discussions again were collegial and constructive. I’m truly lucky to work with colleagues so capable and mission focused, and so determined, like I am, to sharpen the performance of the Federal Reserve. Today, as you know, our Committee decided to vote by a 9-to-3 vote to…
2026-07-29Yeah, so I think officially it’s eight weeks and four days. But, I’m not counting. [Laughter] The message from markets is the message from markets. What I’ve really been trying to do, Steve, as I think you appreciate—and your colleagues appreciate— is getting an unfiltered message from markets. Getting a direct message. Letting buyers and sellers meet at prices for Treasuries—for the foreign exch…
2026-07-29I’m sorry—it’s your question. So interpreting markets is an imperfect business. We central bankers, like market pros, can think these things are overdetermined. But let me offer some speculation. First, as we said in the FOMC statement that you got at two o’clock, the economy—output is solid. Capex and productivity are strong— labor markets—solid, steady. The bond market—the Treasury market—it se…
2026-07-29So I guess I shouldn’t give you their best arguments. I’ll give you some others. [Laughter] So you’re right. I asked for a good family fight, and I got one. That’s the purpose; that’s the design feature. I come into this meeting, even this press conference, heartened by what I’ve experienced the last two days. Most of our discussion were on the big questions that matter to the conduct of monetary…
2026-07-29So, in two words, not much—not much—I’d like to believe that the Committee shares my views, which is the historic problem with data dependence is the data and the dependence. We are not relying on any one individual piece of data as cover or as an excuse or as validation. What I care about, and what I think the Committee cares about, is trends on the data. Sure, we got some encouraging inflation …
2026-07-29There’s a lot in there, Neil. So rates are higher today than they were 42 days ago. Markets have made decisions because we stepped back, in part, from trying to influence those. Market judgments have moved up on what nominal rates are across the Treasury curve. That doesn’t mean we take them as—by dictation, but we’re observing them. So I think it’s a mischaracterization to say that markets haven…
Public-record transcripts. Press conferences are Chair-era speech; prepared speeches are inter-meeting and span all governors and presidents, so a rate change across the two mixes role and era. Role: chair (all rows).

Ask deeper

Quote Powell vs Warsh on tariffs →When did 'inflation expectations' peak? →Speeches vs pressers on the labor market →

Method

The tracker counts how often a fixed set of concepts appears in Federal Reserve speech, normalized per 100,000 characters of each speaker's own words, so a chair who talks more is not credited with saying more. Six concepts are tracked in press conferences (Chair-era speech, 2012 to date) for the current and previous chair, one concept is tracked across prepared inter-meeting speeches by speaker and month, and the current chair's most recent press-conference remarks are reproduced verbatim from the public record. The source is the public transcript archive; nothing here is inferred stance.

Terms are matched as plain substrings inside each speaker's own segments; a concept with several phrasings is tracked as one term per table so rates are never summed across concepts. Counts describe language frequency and nothing else. A rate that rests on fewer than five mentions is shown with its mention count and should be read as an anecdote, not a rate. The issue's data date is the date of the current chair's most recent press conference. The publication is event-driven rather than weekly: a new issue is prepared when a new press conference is on the record or a new month of prepared speeches enters the speeches table, and is otherwise held as unchanged.

Two limits are part of the method. Press-conference language is co-authored by the press corps: a chair's rate on a concept is conditional on what reporters ask that year. And matching is on fixed phrasings, one or two per concept, so a chair who uses different words for the same idea registers as a change in emphasis that may be vocabulary. Counts describe language frequency and nothing else.

Published by AssetFlow.ai LLC. Descriptive research on attention and public-record data; not investment advice and not an offer or solicitation. No return claim attaches to anything above. Issue date 2026-09-01.