A word can reach a conversation long before the facts behind it do, and the American use of this one has been measured: the Quinnipiac University Poll recorded 50% of 1,220 self-identified registered voters saying, in August 2025, that they think Israel is committing genocide in Gaza, on a question that names Israel as the actor and asks the respondent what they happen to know. Nineteen months earlier a randomised split sample that named no actor at all moved the yes share from 32% to 40% by exchanging a single noun, and by 18 percentage points among respondents 65 and older (YouGov, 2024). Inside an earlier grid on the same stem the word attached more readily to Jewish people in Israel than to Arabs in Gaza, and reached 72% only for Europe between 1939 and 1945 (YouGov, 2023). Four readings of the word exist in the published record, they are four instruments rather than four points on a line, and that is the finding rather than the caveat.
Key Findings
- The Quinnipiac University Poll recorded 50% of self-identified registered voters saying Israel is committing genocide in Gaza, against 35% saying it is not and 15% offering no opinion (n = 1,220, fielded 21 to 25 August 2025, margin of error plus or minus 3.4 percentage points including the design effect). It publishes no first-asked date and no trend line for the item, so the reading is a level rather than a movement.
- The two party columns on that item stand 57 points apart: 77% of Democrats answered yes against 20% of Republicans, with independents at 51%. The Quinnipiac University Poll publishes no subgroup base sizes for the item, so every column runs with its label and no interval.
- Changing one noun moved the yes share 8 points inside a single survey. In a randomised split sample fielded from 12 to 17 January 2024, the row reading “Arabs in the Gaza Strip (2023 - present)” drew 32% yes and the row reading “Palestinians in the Gaza Strip (2023 - present)” drew 40%, on an unweighted 500 respondents in each arm (YouGov, 2024). The topline reports no significance test for the gap and the two arms overlap.
- On the identical stem, inside one sample of 1,000 US adult citizens fielded from 21 to 29 November 2023, the yes share read 39% for the row naming Jewish people in Israel against 34% for the row naming Arabs in the Gaza Strip, and 72% for the row naming Jewish people in Europe, 1939 to 1945 (YouGov, 2023).
- Neither of the two organisations with a standing American series on the conflict asks the word. Pew Research Center reports a two-year study fielded from 22 to 28 September 2025 among 3,445 adults whose full text contains “genocide” zero times and which measures the subject as whether Israel’s military action “has gone too far”; Gallup’s Middle East series records approve-or-disapprove items and a forced-choice sympathies question, and no item using the word.
How many registered voters say Israel is committing genocide in Gaza?
The Quinnipiac University Poll recorded 50% of self-identified registered voters saying they think Israel is committing genocide in Gaza, against 35% saying they do not (n = 1,220, fielded 21 to 25 August 2025, margin of error plus or minus 3.4 percentage points including the design effect). It published the reading on 27 August 2025.
A further 15% offered no opinion, the third figure the topline carries and the one most often dropped when the reading is quoted. Quinnipiac University states the row in full: “Fifty percent of voters think Israel is committing genocide based on what they know about the situation in Gaza, while 35 percent do not think Israel is committing genocide and 15 percent did not offer an opinion.”
The population is narrower than the public. These are self-identified registered voters, not all adults and not a general-population sample, so the reading describes the people the instrument reached across those five days, and the margin of error it reports, plus or minus 3.4 percentage points, expressly includes the design effect and applies to the whole sample (Quinnipiac University Poll, 2025).
One feature of the release settles what kind of statement the 50% can carry. The Quinnipiac University Poll dates its own trends, giving a first-asked date of 2 November 2023 for the military-aid item that precedes this one and December 2001 for the sympathies item that opens the block. For the genocide item it published no first-asked date, no earlier reading and no trend line, so nothing in the document supports a claim that the share has risen or fallen.
The word also sits in dated institutional circulation that has nothing to do with polling. The International Court of Justice issued provisional-measures orders on 26 January, 28 March and 24 May 2024, a United Nations commission of inquiry published findings in September 2025, and Amnesty International, B’Tselem and Physicians for Human Rights-Israel have each published positions under their own names. Those are dated positions held by named bodies, and what the Convention requires of a finding is the subject of a separate report on the Convention standard.
What exactly did the question say?
Question 24 read, in full: “In thinking about what you know about the situation in Gaza, do you think Israel is committing genocide, or not?” The Quinnipiac University Poll published no definition beside it, no reference to any legal standard and no follow-up prompt.
Two features of that stem do most of the work. The first is the prefatory clause, which asks the respondent to reason from what they happen to know. That makes the item a report of an impression formed against whatever the respondent has encountered, not a judgment measured against a stated threshold. The second is the actor: Israel is named and the verb stands in the present tense, so an answer of yes is an attribution to a named party rather than a characterisation of an event.
The standard a court applies is a different object, and the legal literature says so in its own terms. A peer-reviewed forum contribution to the Journal of Genocide Research describes the inference rule courts work to as a high legal requirement (Sultany, 2024). A named-author commentary published on the EJIL:Talk! blog on 18 September 2025, editorially curated and not peer-reviewed, states that fact-finders who are not courts, “governments, NGOs, academics, or ordinary people”, are “not bound to apply a high standard of proof” (Milanovic, 2025). A survey respondent sits in that second category by construction, and this stem puts them there explicitly.
So the item records an utterance. The Quinnipiac University Poll documents what a sample said when asked, in one field period, and that is evidence about the sample and about the question rather than about events in Gaza.
A second construction exists in the published record and is built the other way round. It names no perpetrator, asks about the treatment of a named group in a named time and place, and leaves the actor out of the sentence entirely. The two constructions return different numbers, and the size of that difference is measurable because one survey organisation ran both forms in a controlled comparison.
Where in the questionnaire is the word reached?
The word is reached at question 24 of the 27 printed questions, and at the end of a four-item block on Israel. The Quinnipiac University Poll recorded 60% opposing more United States military aid on the item immediately before it, the highest opposition it reports since first putting that question to registered voters on 2 November 2023.
The block runs in a fixed order. Question 21 asked where the respondent’s sympathies lie and returned 37% more with the Palestinians, 36% more with the Israelis and 27% offering no opinion, which the release calls an all-time high for the Palestinians and an all-time low for the Israelis since it began asking registered voters in December 2001 (Quinnipiac University Poll, 2025). Question 22 asked about United States support for Israel. Question 23 asked about military aid. Question 24 asked about genocide.
The release measures no order effect, and none is claimed here. What is established is the position: a respondent reaches the word only after three items that have already placed the conflict, American support for it and American military aid in front of them, and only after supplying an answer to each.
That position matters because placement is known to move shares on questions of this kind. A methodology article on the forced-choice Middle East sympathies question documents that where the item sits inside a survey changes the shares it produces (Gallup, 2018). That is a finding about one house’s instrument rather than a measurement of this one, and it is the reason the sequence is worth setting out.
The sequence is reconstructable at all because the Quinnipiac University Poll publishes every question it asked. All 27 items appear with their toplines, so a reader can see what preceded the word and in what order. That is not true of every instrument in this record, and the difference becomes material below.
How far apart are the partisan columns?
The two party columns stand 57 points apart: the Quinnipiac University Poll recorded 77% of Democrats answering yes against 20% of Republicans (n = 1,220, fielded 21 to 25 August 2025, margin of error plus or minus 3.4 points on the whole sample). Independents sat between them.
The full columns the Quinnipiac University Poll reports read 77% yes, 11% no and 13% no opinion among Democrats; 20% yes, 64% no and 16% no opinion among Republicans; and 51% yes, 34% no and 14% no opinion among independents. The independent column sits closer to the Democratic one than to the Republican one on the yes share.
Every one of those figures runs here with its column label and no interval, and the reason is in the document rather than in an editorial choice. The plus or minus 3.4 percentage points is the whole-sample figure and expressly includes the design effect; no base sizes are published for the party columns, so no interval can be computed for any of them.
Quinnipiac University published a second cut alongside it. Among men the item read 45% yes, 44% no and 12% no opinion; among women it read 55% yes, 27% no and 18% no opinion, a 10-point gap on the yes share and a 17-point gap on the no share. Both columns carry the same limitation as the party columns.
A spread of this size is a fact about the sample rather than an explanation of it. Nothing in the release measures what the columns are tracking, and one cross-sectional reading cannot separate party identification from prior attitude toward the conflict or anything else. What it establishes is that a national figure of 50% averages columns that disagree with each other by more than the figure itself.
Changing one noun moved the yes share 8 points
One noun moved the yes share 8 points. In a randomised split sample fielded in January 2024, one half of the survey met a row reading “Arabs in the Gaza Strip (2023 - present)” and returned 32% yes, while the other half met “Palestinians in the Gaza Strip (2023 - present)” and returned 40% (YouGov, 2024). Same event, same date range, same stem, same field period, one noun changed.
The YouGov Wording Test topline states the design in its own words: “A random half of respondents was asked one of the two Gaza Strip questions and the other half was asked the other one.” Each arm holds an unweighted 500 respondents. The arm naming Arabs read 32% yes, 32% no and 37% not sure; the arm naming Palestinians read 40% yes, 30% no and 30% not sure, and the first of those crosstabs carries a stated total of 101% from rounding.
What did not change is a longer list than what did. The stem was identical: “Do you consider the way the following people were treated in the specified time and place to be a genocide?” The time and place were identical, the response options were identical, and the panel, the weighting and the fielding window of 12 to 17 January 2024 were identical. The only difference between 32% and 40% was which noun named the people.
The gap has to be reported as the size of a difference measured rather than an effect established. The margin of error published on that survey, plus or minus 3.9 percentage points adjusted for weighting, is the whole-sample figure and covers neither arm; each arm rests on an unweighted 500 and therefore carries a wider interval than the published one. The YouGov Wording Test topline reports no significance test, and the two arms overlap.
That caveat narrows the claim without dissolving it. Two halves of one survey, matched by design, returned different numbers about the same event, and the variable between them was a word.
Which age group moved most when the noun changed?
18 percentage points is the size of the wording effect among respondents 65 and older, the age contrast printed for the January 2024 split sample: 24% answered yes for the row naming Arabs in the Gaza Strip, and 42% for the row naming Palestinians (YouGov, 2024).
The YouGov Wording Test write-up states it directly. The wording change, it reports, “has no effect among adults under 30, while increasing the share of people 65 and older who characterize the events as genocide by 18 percentage points.”
The first half of that sentence rests on cells the topline does not carry. The YouGov Wording Test crosstabs record the 18-to-29 cells with a suppression asterisk in both Gaza arms, on an unweighted 54 respondents in the arm naming Arabs, so the reported absence of an effect among the youngest adults is not a result a reader can check against the published table. Only the 65-and-older contrast is printed, and only that contrast is carried here.
An age contrast inside a wording experiment is a statement about the wording, not about how an age group views Gaza. It establishes that the two nouns are not interchangeable for these respondents, and says nothing about why. The same topline reports the gap running in the same direction across party groups, which is evidence that the effect is not a partisan artefact, and it publishes no test of the age difference any more than of the headline gap.
The reading also has to be held against the base it rests on. A crosstab cell for respondents 65 and older inside a 500-respondent arm is a small cell, and no interval is published for it. The contrast of 24% against 42% is the largest printed movement the experiment produced, and it is printed without a bound, which is the honest statement of what the topline contains.
Did the same survey ask the question in both directions?
A November 2023 grid put the word in both directions on one stem, in one sample and in one field period: 39% yes for the row naming Jewish people in Israel, against 34% for the row naming Arabs in the Gaza Strip (YouGov, 2023).
The sample was 1,000 US adult citizens, interviewed online between 21 and 29 November 2023, with a margin of error of plus or minus 4.3 percentage points adjusted for weighting. YouGov’s November grid recorded 39% yes, 32% no and 29% not sure for Jewish people in Israel, and 34% yes, 32% no and 35% not sure for Arabs in the Gaza Strip. Both were rows in one grid, put to one set of respondents on one screen.
In the January 2024 battery the Israel row read 35% yes, 37% no and 28% not sure (YouGov, 2024). The battery itself changed between the two surveys, from thirteen rows to eight, so the two Israel readings were reached after different run-ups and are not two points on one line either.
Neither reading is a judgment about either case, and the stem is built so that it cannot be. It asks whether the respondent considers the way a named group was treated, in a named time and place, to be a genocide. It names no perpetrator anywhere, for any row, which is precisely why the word can be applied in both directions inside a single battery without the instrument contradicting itself.
What the pair establishes is narrow and worth stating plainly. On one stem, in one room, on one day, the word attached slightly more readily to one of the two groups in the current conflict than to the other, and the five points between them sit well inside the interval published for the whole sample. A reader who takes only the Gaza row out of that grid has taken half of what the instrument recorded.
How does the public apply the word to cases already settled?
72% said yes for the row naming Jewish people in Europe, 1939 to 1945, against 47% for the row naming the Tutsi, moderate Hutu and Twa in Rwanda in 1994 (YouGov, 2023). Both rows sat in the same battery, on the same stem, in the same field period as the Gaza row.
YouGov’s November grid records 56% for the row naming Native Americans in North America across the 15th to the 19th centuries. The not-sure column is where the grid is most revealing: 22% were not sure about the Europe row, and 46% were not sure about the Rwanda row, a share within a point of that row’s own yes share.
Those readings set the scale the Gaza figure sits on. The word did not run anywhere near unanimity in this grid, including on the row with the longest documented record: 72% is the top of the range these respondents produced, and 7% answered no to that same row. A 34% for Gaza is therefore not a reading measured against a ceiling of 100%, and the distance between the two is smaller than the raw framing of either figure suggests.
A caution about repeat readings belongs beside all of this. Follow-up surveys, the publisher reports, “show that these estimates can change over time, in part because of the margin of error inherent to any survey estimate”, and the example offered is its own: the Europe row moved from 72% to 79% across five weeks (YouGov, 2023). The same row, the same stem, the same organisation, seven points apart.
Read together, the grid describes a public applying a legal word unevenly across cases with very different evidentiary records, and answering not sure at rates between 22% and 46% depending on the case.
Which polling organisations ask about the word at all?
Pew Research Center reports a two-year study of American views of the conflict, fielded in September 2025 among 3,445 adults, whose full text contains the string “genocide” zero times and which measures the subject as whether Israel’s military action “has gone too far”. Gallup’s Middle East series records approve-or-disapprove items and a forced-choice sympathies question, and no item using the word.
The Pew study was fielded from 22 to 28 September 2025 and takes the same subject as the two instruments above (Pew Research Center, 2025). Its chosen measure is a proportionality item rather than a word item, and the absence of the word was established by a full-text search of the report rather than inferred from its summary.
Gallup carries the deepest American trend on this conflict. Gallup’s Middle East series recorded 63% of US adults sympathizing more with the Israelis in February 2010, described there as the highest reading since 1991, and 46% by February 2025 (n = 1,004, 3 to 16 February 2025, plus or minus 4 points). It measures relative sympathy between two named parties, and a search of its Israel releases and its country topic index surfaced no item asking whether a genocide is occurring (Gallup, 2025).
Neither does the Quinnipiac University Poll publish a tracker on the word, and neither does YouGov. Each of the four readings in this record came from a survey that asked the question once, inside a battery or a block that was not repeated in the same form.
The absence is therefore the finding, and it was searched for rather than assumed. No organisation in this record publishes a standing, identically worded series on whether Americans call Gaza a genocide, so no sentence describing that share as rising, falling or holding has a series behind it. Where a trend is quoted, the quoter has assembled it out of readings never built to be compared.
Do the four published readings measure one quantity?
Four published readings of the word exist, and they are four instruments rather than four points on a line: the November 2023 grid row, the two arms of the January 2024 split sample, and the Quinnipiac University Poll item of August 2025. They differ in question construction, in population, in mode, in sampling frame and in date.
Question construction is the largest difference. The YouGov stem is passive, asks about the treatment of a named group in a named time and place, and names no actor at all; the Quinnipiac stem names Israel, puts the verb in the present tense, and asks the respondent to reason from what they happen to know. Population is next: the YouGov surveys reached US adult citizens, and the Quinnipiac University Poll reached self-identified registered voters. Mode is third: YouGov interviewed online through an opt-in panel matched against the 2019 American Community Survey, while Quinnipiac describes its surveys generally as probability-based samples using random digit dialling with live interviewers.
The temptation this record creates is to read the readings as a rise. The two January arms alone refuse that reading: they were produced in the same survey, on the same days, from the same panel, and they differ by 8 points because of one word (YouGov, 2024). Whatever separates the November row from the August item includes at least a change of stem, of population, of mode and of house, and no published design isolates any one of them.
The Quinnipiac University Poll makes the point without setting out to. It published a first-asked date for the military-aid item, 2 November 2023, and for the sympathies item, December 2001, and neither for the genocide item, nor any earlier reading of it.
So the honest statement of this record is a level and a construction, never a trajectory. On the named-actor question, in one field period, 50% of a sample of self-identified registered voters said yes. On the passive question that names no actor, in different field periods and a different population, the yes share was 32% or 40% depending on a noun.
Methodology and limitations
This report measures utterances recorded by named survey instruments, and nothing else. Two kinds of source carry it: named survey organisations that publish their own toplines with the question stem, the base, the weighting and the margin of error, which is the standard applied here for admitting a poll at all; and two named-author items in the legal literature, cited only for what a standard of proof is and whom it binds.
The Quinnipiac University Poll recorded 1,220 self-identified registered voters nationwide from 21 to 25 August 2025, margin of error plus or minus 3.4 percentage points including the design effect, and published 27 questions with toplines on 27 August 2025; its random-digit-dialling description is a standing account of its surveys generally, not an itemised methodology for this field period. YouGov surveyed 1,000 US adult citizens from 21 to 29 November 2023, margin of error plus or minus 4.3 percentage points adjusted for weighting, in a thirteen-row grid with 80 further questions not reported (YouGov, 2023); and 1,000 US adult citizens from 12 to 17 January 2024, margin of error plus or minus 3.9 percentage points adjusted for weighting, in an eight-row grid with 91 further questions not reported and two Gaza arms of an unweighted 500 each (YouGov, 2024). Respondents there were drawn from an opt-in internet panel by sample matching against the 2019 American Community Survey and weighted afterwards, which is not a probability sample.
Six limits sit on the figures. Every published margin of error is a whole-sample figure and covers no subgroup, no party column and neither split-sample arm. No subgroup base sizes are published for the genocide item, so its columns run without intervals. No significance test is published for the 8-point wording gap, and the two arms overlap. The under-30 cells in both January crosstabs are suppressed on an unweighted 54 respondents in one arm, so the reported absence of an effect among the youngest adults is not verifiable from the table. Rounding is visible in the published rows, one of which totals 101%. And the surrounding questionnaire is undisclosed for both YouGov batteries, so what preceded each grid cannot be assessed.
Two rules governed the presentation. No chart crosses an instrument or a field period, and the four readings are never plotted, averaged or summed together. And no figure speaks to events in Gaza: each records what a sample said when asked, in a stated window, by a stated house.
Conclusion
Why do 50% of registered voters call Gaza a genocide? On the polling record the question splits in two, and the halves do not have the same answer. Ask what share of Americans use the word and the record returns a level, dated and bounded. Ask why they use it and the record returns the question they were asked.
The measurements hold together into one portrait. On a question that names Israel as the actor and asks the respondent to reason from what they happen to know, the Quinnipiac University Poll recorded 50% saying yes, 35% saying no and 15% offering no opinion, and the party columns behind that average stand 57 points apart. On a question that names no actor at all, one organisation moved the yes share from 32% to 40% inside a single survey by exchanging a noun, and moved it 18 points among respondents 65 and older. On the identical stem the word attached to Jewish people in Israel at 39% and to Arabs in Gaza at 34%, and it reached 72% only for Europe between 1939 and 1945, where 7% still answered no and 22% were not sure. And the two houses with the deepest American trend lines on this conflict do not ask the word at all, so nothing in the record can say whether any of these numbers is moving.
The instrument is not a detail beneath the finding. On this record it is most of the finding. A survey question is a container, and what comes out of it carries the shape of the container as much as the shape of the world outside it. Eight points moved on a noun. Eighteen moved on a noun among the oldest respondents asked. So when the next figure arrives and is reported as what Americans believe about Gaza, which of the two things is actually being counted: what people have concluded, or which word they were handed to conclude it with?
Frequently Asked Questions
Why does the Quinnipiac reading carry no trend line?
Because the poll publishes one only where it has asked the question before. The Quinnipiac University Poll records a first-asked date of 2 November 2023 for its military-aid item and December 2001 for its sympathies item, and neither for the genocide item, so the August 2025 reading stands alone in the release.
How were YouGov’s respondents selected?
From an opt-in internet panel, by sample matching against the 2019 American Community Survey, with weighting applied afterwards, a method stated on the toplines themselves (YouGov, 2023). It is not a probability sample of American adults, and the published margins of error are described as adjusted for weighting.
Why are the under-30 cells missing from the age comparison?
They are suppressed. The YouGov Wording Test topline records an asterisk in place of the 18-to-29 cells in both Gaza arms, on an unweighted 54 respondents in the arm naming Arabs, so the reported absence of a wording effect among the youngest adults is not a figure a reader can check.
What does Pew measure if it does not use the word?
Whether Israel’s military action in Gaza “has gone too far”. Pew Research Center reports that item as its instrument for American views of the conflict in a study of September 2025 fielded among 3,445 adults, and the string “genocide” appears in its full text zero times.
Which American question about this conflict does carry a long trend?
The forced-choice sympathies item. Gallup’s Middle East series recorded 46% of US adults sympathizing more with the Israelis in February 2025, the lowest in 25 years of that trend, against 63% in February 2010 (n = 1,004, 3 to 16 February 2025, plus or minus 4 points). It measures relative sympathy between two named parties and carries no item on the word.
Does a poll reading establish anything about events in Gaza?
No, and neither instrument is built to. The Quinnipiac stem opens “In thinking about what you know about the situation in Gaza”, and the YouGov stem asks whether the respondent considers a treatment to be a genocide. Each records a characterisation offered by a respondent in a stated field period.
Why do the two YouGov surveys not compare cleanly with each other?
Because the battery changed between them. The November 2023 grid ran thirteen rows and the January 2024 grid ran eight, with several historical cases dropped, so the Gaza row was reached after a different run-up in each survey. The publisher also cautions that repeat readings “can change over time, in part because of the margin of error inherent to any survey estimate”.
Sources
- Gallup, 2010 to 2025. Middle East sympathies trend series, including the February 2010 and February 2025 readings and the 2018 methodology article on question placement. news.gallup.com/poll/1639. Named survey organisation.
- Milanovic, M., 2025. Proving Genocide. EJIL:Talk!, the blog of the European Journal of International Law, 18 September 2025. ejiltalk.org. Named-author academic commentary, editorially curated and not peer-reviewed.
- Pew Research Center, 2025. Two-year report on American views of the Israel-Hamas conflict, fielded 22 to 28 September 2025 among 3,445 adults. pewresearch.org. Named survey organisation.
- Quinnipiac University Poll, 2025. National poll of 1,220 self-identified registered voters, fielded 21 to 25 August 2025, released 27 August 2025, with toplines and methodology. poll.qu.edu. Named survey organisation (university polling institute).
- Sultany, N., 2024. A Threshold Crossed: On Genocidal Intent and the Duty to Prevent Genocide in Palestine. Journal of Genocide Research. DOI 10.1080/14623528.2024.2351261. Peer-reviewed journal (forum contribution).
- YouGov, 2023. YouGov Survey: Genocide, topline results, fielded 21 to 29 November 2023 among 1,000 US adult citizens. yougov.com. Named survey organisation, opt-in online panel.
- YouGov, 2024. YouGov Survey: Genocide: Wording Test, topline results and accompanying write-up, fielded 12 to 17 January 2024 among 1,000 US adult citizens. yougov.com. Named survey organisation, opt-in online panel.