**TL;DR** — Across 30 client sites through May 2026 we audited a structural choice that lives in whether the answer sentence states its condition: whether the passage that answers a query whose true answer depends on a condition is written as an **explicit conditional sentence** that names the condition the answer holds under ("If your pages are static HTML, prerendering cuts time-to-first-byte by about 60%; on a server-rendered app it does almost nothing.") or as an **unconditioned blanket sentence** that states the answer flat and leaves the condition out ("Prerendering cuts time-to-first-byte by about 60%."), and whether the conditional shape changes how often the AI Overview lifts that sentence into the card. Across 7,620 cited-passage events on condition-dependent queries we joined each cited sentence to whether its main clause carried an explicit condition. The headline is that the explicit conditional sentence is a real and large citation lever on these queries, and it is really an answer-completeness lever wearing an "if-clause" costume. A conditional sentence that named the condition was cited 2.3× more often than a matched blanket sentence stating the same answer without it on the same condition-dependent query. The strongest predictor was condition-in-clause — a sentence that put the deciding condition in the sentence itself was lifted far more than one that left the reader to supply it. The second was single-condition scope — a sentence gating the answer on one clear condition was lifted more than one stacking three nested "if"s. The third, and the warning, was false conditioning — bolting an "if" onto a query whose answer does not actually depend on one was cited no more, and on 5% of pages the composer paired the over-qualified sentence with a competitor's clean unconditioned answer. One change — rewriting blanket answer sentences on condition-dependent queries into a single-condition conditional that names the condition the answer holds under — lifted cited-passage rate by 22% on the affected sites over a 30-day follow-up.
Why we ran this audit
The AI Overview composer lifts a sentence and drops it into a card as the answer to a query, and some queries have no single true answer — the answer depends on a condition. "Does prerendering speed up my site" is answered by "yes, if your pages are static, and barely if they are server-rendered", and a sentence that states "prerendering speeds up your site" flat is true under one condition and false under another. A human reading the surrounding page picks up the condition from the section it sits in — they read the paragraph about static sites and apply the answer there — but the composer extracting one sentence has no such frame, and a blanket sentence with the condition stripped is, to it, an answer that is right sometimes and wrong sometimes. We had spent weeks on the shape of the answer sentence — its length, its mood, whether it named both sides of a comparison — and the condition is the natural next structural variable, because a condition-dependent query is one whose answer is incomplete without the "if".
The second motivation was a writing habit that hides the condition. A page wants a confident, quotable answer, so it states the headline result plainly and pushes the "it depends" into a later sentence, a footnote, or a caveat paragraph below — "Prerendering cuts TTFB by 60%" reads stronger than "Prerendering cuts TTFB by 60% if your pages are static". A human reader on the page reaches the caveat eventually and scopes the claim correctly. The composer, hunting for one sentence that answers "does prerendering help me", finds the confident blanket sentence, and either lifts a claim that is wrong for half the queries or skips it for a competitor whose sentence carries the condition. We needed to know whether dropping the condition cost the citation, because if it did, the fix is nearly free — fold the deciding condition back into the answer sentence — and it costs only the appearance of a cleaner headline.
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 condition-dependent queries ("does X work for Y", "is X worth it", "should I use X", "when does X help") where the honest answer turns on a condition. Twice daily through May 2026 we captured every AI Overview card, and for cards citing a client page on a condition-dependent query we identified the specific lifted sentence and classified its shape: explicit conditional (the deciding condition stated in the main clause via "if", "when", "for", "on", "unless"), partial (a condition gestured at but not named), or blanket (the answer stated flat with no condition). For each cited sentence we built a matched control: a comparable sentence on a similar condition-dependent query whose shape differed but whose underlying answer was the same, so the comparison was conditional-vs-blanket rather than good-page-vs-bad-page. The cited cohort was 7,620 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 blanket sentence that reads as obviously scoped inside a section headed "For static sites" reads as an unconditioned claim in the card. And we matched on sentence citability before comparing shape — we paired each cited sentence 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 conditional shape is not just the conditional pages being better written overall. The 2.3× and 1.6× figures are from those matched comparisons, not raw averages.
The shape of the conditional pattern
The flat headline first. On condition-dependent queries, explicit conditional sentences are cited more. A sentence that named the deciding condition was lifted 2.3× more often than a matched blanket sentence on the same query. The effect held through the quality match and the citability control: among sentences our rubric scored as equally liftable, the conditional ones were lifted far more than the blanket ones. The composer behaves as though it prefers a sentence that answers the whole question — including the "under what condition" the query implies — over one that states a result that is only sometimes true.
The most decision-relevant cut was that this is about completing the answer, not about adding a qualifier for its own sake. We tested whether the win came from the presence of an "if"-clause or from the condition actually being the one the query turned on, and the second was the whole story: a sentence with an irrelevant condition ("Prerendering cuts TTFB, if you have a modern browser") was cited no better than a blanket one, while a sentence naming the deciding condition ("…if your pages are static") was lifted far more. The conditional wins when the condition it names is the one the answer genuinely depends on. Name the condition that flips the answer, not any condition.
Driver one: put the deciding condition in the answer sentence
The single strongest predictor was whether the answer sentence carried the deciding condition in its own clause. Holding the answer constant, a sentence that stated the condition was lifted at 2.3× the rate of one that left it to the surrounding text. The composer extracts a sentence and reads it as the answer to "does X help"; a sentence that says "X helps if your pages are static" answers the question completely, while one that says "X helps" answers it only for the readers whose pages happen to be static and misleads the rest. A human reader scopes the bare claim from the section it sits under; the composer reading the sentence cold cannot, so the condition has to be inside the sentence itself.
We ran a structural test on 26 answer sentences across 14 clients, each a blanket sentence on a condition-dependent query that stated a result without its condition. We rewrote each to fold the deciding condition into the main clause, changing no underlying answer — only naming the condition the result held under. Over the 45 days that followed, 18 of the 26 sentences began being lifted on at least one condition-dependent query where the blanket version had been skipped. The lever was not new content; it was moving the condition that already lived in the paragraph into the single sentence the composer would extract, so the sentence answered the conditional query on its own.
Driver two: gate on one condition, not three nested ones
Holding condition-in-clause constant, the second driver was how many conditions the sentence carried. A sentence gating the answer on one clear condition ("Prerendering cuts TTFB if your pages are static") was cited more than one stacking several ("Prerendering cuts TTFB if your pages are static and you are not already on a CDN and your origin is the bottleneck"). The reading consistent with the data is that the composer lifts a sentence as one self-contained answer, and a single-condition sentence is a complete answer to "does X help, and when", while a triple-nested one is a decision tree the composer cannot resolve into a clean card — so it skips it for a sentence carrying the one condition that matters most.
We ran a structural test on 19 over-conditioned sentences across 11 clients, each stacking three or more "if"s into one answer. We rewrote each to lead on the single dominant condition and move the secondary ones to a following sentence or list, changing no facts. Over the 60 days after the change, 13 of the 19 lead sentences improved their cited-passage rate. The two drivers compound: a blanket sentence with no condition is one failure mode and a sentence buried under three nested conditions is the other — the sentences that won named exactly one deciding condition and left the rest to the prose around them.
Driver three: false conditioning, and the "if" the answer does not need
The third driver was the warning. A condition helps only when the answer genuinely depends on it, and bolting an "if" onto a query whose answer is unconditional backfires. A sentence like "Compression cuts page weight by about 40%, if you compress your images" — a condition wrapped around an answer that does not actually vary — was cited no more often than the clean blanket version, and on 5% of audited pages the composer paired the awkwardly-qualified sentence with a competitor's unconditioned answer that read more directly, so the citation was shared rather than won outright. The reading consistent with the data is that the composer rewards a condition that carries real information about when the answer changes, and a redundant "if" reads as noise around a fact that is simply true. A query with a single true answer wants that answer stated flat; only the query whose answer flips on a condition wants the "if".
We confirmed this on 15 sentences across 9 clients where an earlier optimisation pass had added conditions to answers that did not depend on them. We rewrote each back into a clean blanket statement for the unconditional query while keeping the conditional on the adjacent query whose answer really did vary, matching shape to whether the answer turned on a condition. Over the following 45 days the unconditioned answers regained their solo citation while reading directly, and none drew a shared-citation pairing. The actionable rule is blunt: state the condition when the answer depends on one, and state the answer flat when it does not — a needless "if" reads as a qualifier the composer will pass over for a cleaner sentence.
What changed in our content checklist
Three changes. We added a condition pass for condition-dependent queries: before publishing, we read each section's lead answer sentence and ask whether the answer actually depends on a condition, and a blanket sentence answering a query whose true answer turns on one gets the deciding condition folded into its main clause — because the composer lifts a sentence whole and reads a conditioned answer as the complete answer to "does X help, and when". We added a single-condition check to the same pass: the lead sentence names exactly one deciding condition, with secondary ones moved to a following sentence or list, because the composer prefers a self-contained answer over a nested decision tree. And we added a false-conditioning guard: we strip "if"s off answers that do not actually vary, so a needless condition never clutters a sentence whose answer is simply true.
We dropped one habit. For years our pages had stated the headline result plainly and pushed the "it depends" into a caveat below, on the belief that a confident blanket sentence read stronger than one hedged with a condition — "Prerendering cuts TTFB by 60%" felt more quotable than "Prerendering cuts TTFB by 60% if your pages are static". The audit removes that default for the answer sentence on condition-dependent queries: the one sentence the composer would lift has to carry the deciding condition, and a blanket sentence spends the citation to read like a cleaner headline. So unconditioned answer sentences left our playbook for condition-dependent queries — we now fold the deciding condition into the lead sentence and let the surrounding prose carry the cases, accepting that the cited sentence reads more qualified than a headline writer would choose because it is built to answer the conditional query alone.
- 01Put the deciding condition in the answer sentence. A sentence naming the condition was cited 2.3× more than a blanket one on the same query — the composer reads a bare answer as right-sometimes-wrong-sometimes for a condition-dependent query.
- 02Name the condition that flips the answer. An irrelevant "if" was cited no better than a blanket sentence; only the condition the answer genuinely turns on moved the citation.
- 03Gate on one condition, not three. A single-condition sentence was lifted more than a triple-nested one — the composer cannot resolve a decision tree into a clean card.
- 04Do not false-condition. Bolting an "if" onto an answer that does not vary was cited no more, and on 5% of pages the composer shared the citation with a competitor that answered flat.
Where this argument breaks
For queries with a single true answer — "what is prerendering", "how many bytes in a kilobyte" — there is no condition to state and the conditional shape is irrelevant, so the lever is for queries whose answer turns on a condition. For navigational and brand queries there is no answer sentence whose shape matters. For narrative and persuasive passages — case studies, opinion, story-driven content — naming a condition is an argument and framing choice, not a citation lever, and the condition pass is for the answer sentences on condition-dependent queries only. For some languages the effect may differ — in our parallel Chinese-language audit (文心一言, 元宝, 通义) the condition-in-clause win was present but the placement effect was smaller, since Chinese conditionals lead with the condition by default («如果……就……») and the composer read the gate reliably wherever it sat. The 5% shared-citation figure is small and noisy; we are confident a needless condition does not help and mildly confident it splits the citation, but it is the weakest finding here and we would not restructure a page on it alone. 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 conditional answer sentence — the deciding condition named, gated on one condition, used only where the answer truly varies — far more readily than a blanket sentence that states a result that is only sometimes true, the unit is the individual answer sentence rather than the page, and the cheapest citation win on a condition-dependent query is to fold the "if" that flips the answer back into the sentence the composer lifts.