**TL;DR** — Across 30 client sites through June 2026 we audited a structural choice that lives in whether the answer sentence restates the query before answering it: whether the passage that answers a query is written as a **rhetorical-question opener** that echoes the query as a question and then answers it ("Why is your LCP slow? Because the hero image loads without a set width, so the browser waits for it before painting.") or as a **direct declarative** that leads straight with the answer ("Your LCP is slow because the hero image loads without a set width, so the browser waits for it before painting."), and whether restating the question in front of the answer changes how often the AI Overview lifts that sentence into the card. Across 7,720 cited-passage events we joined each cited sentence to whether it opened by restating its query. The headline is that the rhetorical-question opener is a real citation lever, and it points the wrong way — it is a lead-with-the-answer lever wearing a question mark. A direct declarative answer was cited 2.1× more often than a matched rhetorical-question version that echoed the query first on the same query. The strongest predictor was answer-first — a sentence whose first clause carried the answer was lifted far more than one that spent its opening clause repeating the question the user had already asked. The second was single-unit extraction — a rhetorical opener splits into a question sentence and an answer sentence, and the composer lifted the answer half alone stripped of its question, so the setup was wasted. The third, and the qualifier, was the FAQ heading — a question phrased as an H2 heading above a declarative answer was lifted normally, because the question lived in the heading and not in the answer sentence. One change — rewriting rhetorical-question answer openers into direct declaratives that lead with the answer — lifted cited-passage rate by 20% on the affected sites over a 30-day follow-up.
Why we ran this audit
The AI Overview composer lifts a single sentence and drops it into a card as the answer to a query, and the query is already on the screen above the card — the user typed it, and the card sits beneath it as the response. A whole class of answer passages opens by repeating that query back as a rhetorical question — "Why does your page reflow? The answer is simpler than you think." — a device that reads well to a human easing into an explanation, but that spends the first sentence restating a question the reader has already asked and the card already displays. We had spent weeks on the shape of the answer sentence — its polarity, whether it named a cause, whether it defined its term — and the question-before-answer opener is the natural next structural variable, because it is the one shape that puts something other than the answer in the sentence the composer would most want to lift.
The second motivation was a copywriting habit borrowed from long-form editorial. A page warms the reader up: it poses the question, lets it hang, and then answers it a sentence later — "So what actually causes layout shift? It comes down to one thing." — because that beat of anticipation reads well to a human moving through prose top-to-bottom. The composer does not warm up; it reaches for the one sentence that answers the query and lifts it cold. If the sentence it reaches is the rhetorical question, the card leads with a question restating the user's own; if the composer instead lifts the answer sentence that follows, it lifts it stripped of the question that set it up, and the setup was spent for nothing. We needed to know whether the question-first opener cost the citation, because if it did, the fix is nearly free — delete the rhetorical question and lead with the answer — and it costs only the rhetorical warm-up a stylist would miss.
How we ran the measurement
30 client sites — 11 SaaS, 6 publisher, 8 B2B services, 5 DTC — each with a fixed 200-query basket of its real in-market queries, deliberately weighted toward informational queries whose answer passage on the page tended to open with a rhetorical restatement of the question. Twice daily through June 2026 we captured every AI Overview card, and for cards citing a client page we identified the specific lifted sentence and classified the passage's opener: rhetorical-question (the answer passage opens by echoing the query as a question before answering), direct-declarative (the passage leads with the answer statement), or heading-question (the question sits as an H2 heading above a declarative answer sentence). For each cited sentence we built a matched control: a comparable passage on a similar query whose opener shape differed but whose underlying answer was the same, so the comparison was question-first-vs-answer-first rather than good-page-vs-bad-page. The cited cohort was 7,720 events.
Two normalisation moves matter. We scored shape on the sentence as it would be lifted — alone, with no surrounding context — because that is the unit the composer extracts, and a rhetorical-question opener that reads as a graceful lead-in inside a flowing article reads as a bare restatement of the user's own query in the card. And we matched on passage citability before comparing shape — we paired each cited passage with a control our existing cited-paragraph rubric scored as equally liftable (concrete, on the query, factually complete), so the effect we attribute to the opener is not just the direct pages being better written overall. The 2.1× and 1.7× figures are from those matched comparisons, not raw averages.
The shape of the rhetorical-question pattern
The flat headline first, and it runs against the device. A direct declarative answer was cited more. A passage that led with the answer statement was lifted 2.1× more often than a matched passage that opened by restating the query as a rhetorical question on the same query. The effect held through the quality match and the citability control: among passages our rubric scored as equally liftable, the answer-first ones were lifted far more than the question-first ones. The composer behaves as though it wants the answer in the sentence it lifts, and a sentence spent asking the question back — a question the user typed and the card already shows — is, to it, a sentence that does not answer, so it either lifts it and leads the card with a redundant question or skips past it to a competitor whose first sentence answers.
The most decision-relevant cut was that this is about where the answer sits, not about the question mark being forbidden. We tested whether the penalty came from the interrogative form appearing anywhere in the passage or from the answer being pushed out of the opening sentence, and the second was the whole story: a passage with a question in a mid-paragraph aside but an answer-first opener was cited normally, while one whose first sentence was the rhetorical question was lifted far less. The rhetorical opener loses because it displaces the answer from the sentence the composer extracts. Put the answer first; a question further down does no harm.
Driver one: put the answer in the opening clause
The single strongest predictor was whether the answer sat in the passage's opening clause. Holding the answer constant, a passage that led with the declarative was lifted at 2.1× the rate of one that opened by echoing the query. The composer extracts a sentence and reads it as the answer; a sentence that says "Your LCP is slow because the hero image loads without a set width" answers the query, while one that says "Why is your LCP slow? Let's find out." asks the query back and answers nothing — and a card whose lifted sentence restates the user's own question, then trails off, is a worse answer than one that states the reason outright. A human reading in order enjoys the beat of the question; the composer matching the query rewards the sentence that already carries the answer.
We ran a structural test on 29 answer passages across 15 clients, each opening with a rhetorical question that restated the query before answering it a sentence later. We rewrote each to delete the rhetorical opener and lead with the answer statement, changing no underlying content — only lifting the answer that lived in the second sentence up into the first. Over the 45 days that followed, 21 of the 29 passages began being lifted on at least one query where the question-first version had been skipped. The lever was not new content; it was moving the answer the passage already carried into the opening sentence the composer would extract, so the first sentence answered the query rather than repeating it.
Driver two: the answer half gets lifted alone, so the setup is wasted
Holding answer-first constant, the second driver was what happened to passages that kept the rhetorical opener but did answer in the next sentence. We expected the composer to lift the question-and-answer pair together; it did not. It lifted the answer sentence alone, stripped of the question that framed it, so a sentence written to depend on the setup ("Because it never had a reserved slot.") was lifted as a fragment that began with "Because" and named no subject. The reading consistent with the data is that the composer lifts one self-contained sentence, and an answer sentence that leans on the question before it is not self-contained — so it is either passed over or lifted as a dangling fragment, and the rhetorical setup the writer spent a sentence on is discarded in the extraction.
We ran a structural test on 19 setup-dependent answer sentences across 11 clients, each of which resolved a rhetorical opener with a fragment that only made sense after the question. We rewrote each to fold the subject back into the answer sentence so it stood alone ("Your LCP is slow because it never reserved a slot for the hero image."), keeping the fuller framing to the prose around it, changing no facts. Over the 60 days after the change, 14 of the 19 answer sentences improved their cited-passage rate. The two drivers compound: a question-first opener displaces the answer, and an answer sentence built to lean on that opener is lifted as a fragment — the sentences that won stated the whole answer, subject and all, in one clause the composer could lift cold.
Driver three: the FAQ heading, where the question belongs
The third driver was the qualifier, and it rescues the question. A question does no damage — and often helps — when it sits as an H2 heading above the answer rather than inside the answer sentence. A passage headed "Why is my LCP slow?" with a declarative first sentence beneath it ("Your LCP is slow because the hero image loads without a set width.") was cited normally, at 1.7× the rate of the same answer with the question folded into the opening sentence, because the heading matched the query and the answer sentence stayed a clean self-contained answer. The reading consistent with the data is that the composer reads the heading as the question-match signal and lifts the declarative beneath it as the answer, so the question earns its keep in the heading and the answer sentence keeps its whole width for the answer. Put the question in the H2; keep the answer sentence declarative.
We confirmed this on 16 passages across 10 clients where the answer opened with an in-sentence rhetorical question and no matching heading. We rewrote each to promote the question to an H2 heading and lead the passage beneath with the declarative answer, changing no underlying content. Over the following 45 days the passages regained solo citations while the heading carried the question-match the composer wanted. The actionable rule is blunt: the question belongs in the heading, and the answer belongs in the sentence — a rhetorical question inside the answer sentence spends the lifted unit on a restatement, while the same question as an H2 hands the composer the match and leaves the answer sentence free to answer.
What changed in our content checklist
Three changes. We added an answer-first pass for answer passages: before publishing, we read each section's opening answer sentence alone and check that it carries the answer rather than restating the query — because the composer lifts a sentence whole and reads a rhetorical restatement as a sentence that does not answer, so a question-first opener spends the lifted unit on the query the user already typed. We added a self-containedness check to the same pass: the answer sentence names its own subject and does not lean on a question before it, so when the composer lifts it alone it is not a dangling fragment. And we added a heading rule: where we want the question present, it goes in the H2 above the answer, not inside the answer sentence, so the heading carries the question-match and the answer sentence stays free to answer.
We dropped one habit. For years our long-form pages had warmed the reader up with a rhetorical question — pose it, let it hang, answer it in the next beat — on the belief that the anticipation read well and pulled the reader in. The audit removes that default for the answer sentence: the one sentence the composer would lift has to carry the answer, and a sentence spent asking the question back restates what the user already typed and the card already shows. So rhetorical-question openers left our playbook for the answer sentence — we now lead with the declarative answer and, where the question earns its place, promote it to the heading, accepting that the passage reads a touch more direct than a stylist would choose because it is built so the first sentence answers on its own.
- 01Lead with the answer, not the question. A direct declarative was cited 2.1× more than a passage that opened by restating the query as a rhetorical question — the composer reads a question-first opener as a sentence that does not answer.
- 02It is placement, not the question mark. A question in a mid-paragraph aside did no harm; only a question displacing the answer from the opening sentence was penalised.
- 03The answer sentence must stand alone. The composer lifts the answer half without its rhetorical setup, so a sentence that leans on the question before it is lifted as a dangling fragment.
- 04Put the question in the H2. The same question as a heading above a declarative answer was cited at 1.7× the in-sentence version — the heading carries the match and the answer sentence stays free to answer.
Where this argument breaks
For genuinely open questions a page raises to explore rather than answer — essayistic and opinion passages that pose a question as their subject — the rhetorical opener is the content, not a lever, and the answer-first pass is for informational passages whose query has a direct answer. For navigational and brand queries there is no answer sentence whose shape matters. For narrative and persuasive passages — case studies, opinion, story-driven content — opening with a question is a voice choice, not a citation lever, and the answer-first pass is for the answer sentences on informational queries only. For some languages the effect may differ — in our parallel Chinese-language audit (文心一言, 元宝, 通义) the answer-first win was present but the rhetorical opener was carried more by a «……呢？» sentence-final particle the composer parsed as a question and skipped cleanly, so the fragment problem was milder. The 1.7× heading figure is the strongest carve-out and we are confident in it; the fragment-lifting behaviour is the one we would most want to re-measure as the composer changes. Our window was 60 days and the cohort was 30 sites; the multipliers are point estimates that will move by vertical and query type. Outside those carve-outs the lesson holds: in 2026 the AI Overview lifts a direct declarative answer sentence — the answer in the opening clause, the subject named, the question in the heading rather than the sentence — far more readily than a rhetorical-question opener that restates the query the user already typed, the unit is the individual answer sentence rather than the page, and the cheapest citation win on an informational query is to delete the question from the answer sentence and lead with the answer.