Navigation
Hub: 00 - Fact-Check MOC · Previous: 07 - Query Design Audit · Next: 09 - The Civilians Query Problem
Every count in the book set beside the claim attached to it, sorted by whether a frequency can carry that claim.
Every quantitative finding in the book, paired with the interpretive claim attached to it, and an assessment of whether the counting method can support that claim.
The recurring pattern: the measurement is a frequency, and the claim attached is frequently a universal — all, virtually all, almost always, every time. A frequency count establishes how often a string appeared. It cannot establish what every instance meant. Where a universal is asserted, hand-coding of every instance would be required, and no such coding is reported.
What the counts include
The book states its counting rule once, parenthetically, at Attachment (file not included in this release):
(Note: this count is conservative in that it does not include quotes from US and Israeli officials or outside commentators. The count is only the terms being used outside of quotes—it is editorial language used in ostensibly straight factual reporting.)
This matters more than its placement suggests, and it cuts both ways. It is a genuinely conservative choice that strengthens the emotive-terms findings: editorial voice is the thing an editor controls, so restricting to it isolates the editorial decision. It is also the reason the near-zero Palestinian counts may not mean what an unattributed replication would suggest, since a query that cannot distinguish quoted from editorial usage is measuring a different quantity. That is the unresolved question in 04 - Results Stage 1.
Note also that the rule is stated for the Chapter 2 emotive-terms analysis specifically. Whether it was applied to the later counts in this table is not stated anywhere.
The table
| Search term | Outlets | Time period | Result | Attached claim | Can the count support it? | Book |
|---|---|---|---|---|---|---|
| massacre, barbaric, savage, slaughter | AP, CNN.com, Politico, USA Today, NYT, WaPo, Axios | First 100 days | 494:1 (Israeli:Palestinian) | Atrocity language reserved for Israeli victims | Partly. Count sound; vulnerable to event-form objection | Attachment (file not included in this release) |
| massacre, barbaric, savage, slaughter | CNN, MSNBC | First 30 days | 721:30 (MSNBC 348:12, CNN 373:18) | Same | Partly. Same caveat. Corrected 2026-08-04: this row previously read 1,053:43, a figure that appears nowhere in the book | Attachment (file not included in this release) |
| “human shield(s)“ | NYT, WaPo, Politico, Axios | First year | 231:5 (5 all WaPo) | Allegation applied to one party, withheld for other despite comparable documentation | Yes. Same allegation, same vocabulary, same evidentiary basis available. Strongest finding in the book | Attachment (file not included in this release) |
| “human shield(s)“ | CNN, MSNBC | First year | 888:0 (CNN 438, MSNBC 450) | Same | Yes. Same reasoning | Attachment (file not included in this release) |
| “Hamas-run” / “Hamas-controlled” | CNN, MSNBC | First 100 days | CNN 277, MSNBC 146; zero before Oct 17 | Doubt-sowing qualifier adopted as policy after al-Ahli | Yes. Evidence is the before/after discontinuity, not the total. Corroborated by CNN standards memo reported by the Intercept | Attachment (file not included in this release) |
| “right to defend itself” | CNN, MSNBC | Survey period (unstated) | 755:8 | Typically followed by rationalization of mass civilian killing | No. Claim about what the phrase does in a sentence; function never counted | Attachment (file not included in this release) |
| “right to defend itself” | Print outlets | First year | 459:4 | Same | No. Same problem | Attachment (file not included in this release) |
| “civilization” | CNN, MSNBC | Survey period (unstated) | 150 mentions | All 150 referred to Israeli/American culture, all contrasted with alleged Palestinian barbarity | No. Double universal over 150 instances; no coding reported | Attachment (file not included in this release) |
| “barbaric”, “savage” | Print outlets | First 100 days | 14:0 and 16:0 | Terms carry colonial-racial connotation, deployed to dehumanize | No. Etymological argument, possibly correct; frequency cannot demonstrate connotation or intent | Attachment (file not included in this release) |
| “hostage” vs “prisoner” | CNN, MSNBC | One year | 46,585 vs 3,947 | Virtually all hostage uses Israeli, virtually all prisoner uses Palestinian | No. “Virtually all” over 46,585 instances is unverified and unverifiable without sampling | Attachment (file not included in this release) |
| “hostage”/“kidnapped” vs “prisoner”/“detain” | 7 print outlets | One year | 30,237 vs 6,420 | Same | No. Same problem; also non-commensurable categories | Attachment (file not included in this release) |
| “Hamas stronghold” and variants | NYT | Oct 2023 – Oct 2024 | 154 uses | Almost always deployed while reporting war crimes or pending war crimes | No. Five headline examples given; five instances do not establish a pattern across 154 | Attachment (file not included in this release) |
| “Israeli hostages” vs “Palestinian prisoners” | Sunday shows | Oct 7 2023 – Oct 7 2024 | 835 vs 52 | Every prisoner mention was in exchange context only | No. Universal over 52 instances, not reported as checked; also non-substitutable legal categories | Attachment (file not included in this release) |
| “right to defend itself” | Sunday shows | Oct 7 2023 – Oct 7 2024 | 149, all for Israel | Palestinians never granted the right | Partly. Zero is checkable; but see commensurability problem | Attachment (file not included in this release) |
| “Hamas-run” | Sunday shows | Oct 7 2023 – Oct 7 2024 | 31 uses | Palestinian deaths hedged, Israeli deaths stated as fact | Yes. Asymmetric sourcing convention is directly observable | Attachment (file not included in this release) |
| antisemitism vs Islamophobia | Sunday shows | Oct 7 2023 – Oct 7 2024 | 65 vs 4 | Asymmetric concern for bigotry | Partly. Commensurable, but no control for news events driving each | Attachment (file not included in this release) |
| massacre, barbaric, savage, slaughter | Morning Joe (MSNBC) | First 6 months | 81:0, 22:0, 38:0, 119:1 | Most aggressively anti-Palestinian program on cable news | Partly. Not Johnson’s data — cites Marcetic’s Jacobin study; inherits its method unexamined | Attachment (file not included in this release) |
| Civilian mentions, Ukraine vs Gaza | CNN, MSNBC | First 100 days of each | 29,534 vs 15,062 (CNN 15,593 vs 7,569) | Twice the coverage despite five times the death toll | Qualified. Matched windows, same networks, victim group varied — the best design in the book. But the term list is not what the claim says it is, and correcting for that takes 1.96x to 1.40x. See 09 - The Civilians Query Problem | Attachment (file not included in this release) |
| Refugee/displacement terms, Ukraine vs Gaza | CNN, MSNBC | First 100 days of each | 1,663 vs 211 | Eight times more humanizing language despite greater displacement | Yes. Same matched design | Attachment (file not included in this release) |
| Child victims, Ukraine vs Gaza | CNN, MSNBC | First 100 days of each | 4,223 vs 3,632 | Asymmetric attention to children | Partly. Gap is narrow relative to the claimed disparity | Attachment (file not included in this release) |
| “genocide”/“war crime”, Ukraine vs Gaza | CNN, MSNBC | First 30 days of each | 17.2x for Ukraine | Asymmetric willingness to name atrocity | Unresolved. Gaza query string records “genoicde” misspelled | Attachment (file not included in this release) |
The pattern, stated plainly
Sorting the rows, three groups emerge.
Findings where the number is the argument. Human shields, “Hamas-run,” and the Ukraine comparisons. In each case the count directly instantiates the claim: the same term, available to both parties, applied to one. Nothing interpretive has to be added. These survive scrutiny.
Findings where a universal has been bolted onto a frequency. “Civilization” all 150 times, “hostage” virtually all 46,585, “Hamas stronghold” almost always, “Palestinian prisoners” every time. Here the number is real and the claim is a separate assertion that the number does not test. Some may well be true — a hand-coded sample could establish them — but as published they are unsupported.
Findings where the claim is about connotation or function. Barbaric and savage as colonial vocabulary; “right to defend itself” as a rationalization device. These are literary and rhetorical arguments. They may be persuasive on their own terms. Frequency data neither supports nor refutes them, and presenting a count alongside them implies an evidentiary basis that is not there.
Why this matters more than the arithmetic
The discrepancies in 05 - Findings Register are minor and mostly run against the book’s own case. This is the more serious methodological problem, and it is systematic rather than incidental.
The rhetorical structure repeats: state a count, then state a universal, then let the count’s specificity lend authority to the universal. A reader who checks the number finds it accurate, and reasonably infers the sentence around it was checked too. It was not.
What this does and does not establish
That a claim is unsupported by the cited evidence does not make it false. Several of these universals are plausible on their face and would likely survive a coding pass. The finding here is about the gap between what was measured and what was asserted — which matters especially for a book that explicitly claims to prove its case beyond a reasonable doubt.
Checkable next
The universals are the most tractable remaining work, because most are small enough to verify exhaustively:
- “Palestinian prisoners” on the Sunday shows — 52 instances, all retrievable via GDELT clip export. Code each for whether it appears in an exchange context. Fully settles that claim. Book: Attachment (file not included in this release)
- “Civilization” — 150 instances. Code each for referent and contrast. Book: Attachment (file not included in this release)
- “Hamas stronghold” — 154 NYT uses. Code each for whether it accompanies war-crime reporting. Book: Attachment (file not included in this release)
- “Hostage”/“prisoner” — too large to code exhaustively. Draw a random sample of 200 per term and estimate the “virtually all” rate with a confidence interval.
Each of these either substantiates Johnson’s claim or does not, and each is a day’s work at most.