**TL;DR** — Across 30 client sites through May 2026 we audited a structural choice that lives in the shape of the answer sentence on comparison queries: whether the passage that answers a "X vs Y" question is written as an **explicit comparative sentence** that names both sides and states the verdict ("PostgreSQL handles concurrent writes better than MySQL because of its MVCC implementation.") or as a **one-sided descriptive sentence** that praises only the side the page sells and leaves the reader to infer the comparison ("PostgreSQL handles concurrent writes well thanks to MVCC."), and whether the comparative shape changes how often the AI Overview lifts that sentence into the card. Across 7,540 cited-passage events on comparison queries we joined each cited sentence to whether its main clause named both compared entities and stated a direction. The headline is that the explicit comparative sentence is a real and large citation lever, and it is really a question-completeness lever wearing a "both sides named" costume. A comparative sentence that named both entities and gave a direction was cited 2.4× more often than a matched one-sided sentence describing only one entity on the same comparison query. The strongest predictor was both-sides naming — a sentence that mentioned only the page's own product was passed over even when its claim was true. The second was verdict direction — a sentence that stated which side won on the compared dimension was lifted more than one that listed both sides neutrally without resolving the comparison. The third, and the warning, was unscoped superlatives — flipping every comparison to a flat "X is better than Y" without naming the dimension was cited no more, and on 6% of pages the composer paired the bare claim with a competitor's sentence that supplied the missing dimension. One change — rewriting one-sided answer sentences on comparison queries into a comparative sentence that names both entities, states the direction, and scopes it to a dimension — lifted cited-passage rate by 23% 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 a comparison query asks a two-sided question. "Is PostgreSQL or MySQL better for concurrent writes" is a request to weigh two named things on a stated dimension, and a sentence that answers it has to carry both things and a direction, while a sentence that praises only one of them answers half the question and leaves the composer to find the other half elsewhere. A human reading a one-sided page infers the comparison from context — they know the page is about PostgreSQL, so the praise reads as an implicit win — but the composer extracting one sentence has no such context, and a sentence that names only one side is, to it, an answer to "is PostgreSQL good", not "is PostgreSQL better than MySQL". We suspected the composer was quietly preferring sentences that named both compared entities because they answer the whole comparison, and we wanted to know whether the preference was about naming both sides or about stating a direction between them.
The second motivation was a writing habit endemic to comparison pages. A page built to sell PostgreSQL describes PostgreSQL — its strengths, its features, its wins — and mentions MySQL only in passing, because the page exists to make the case for one side, not to umpire a neutral contest. A human reader on a "vs" query gets the answer anyway, because the whole page is the comparison and the omission of MySQL reads as "MySQL is the loser we are not dwelling on". The composer, hunting for one sentence that answers "X vs Y", finds a sentence that only names X and treats it as failing to address the comparison, then lifts a competitor's sentence that names both. We needed to know whether the one-sidedness cost the citation, because if it did, the fix is nearly free — name the other side in the answer sentence and state the direction — and it does not require conceding the page's position.
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 comparison queries ("X vs Y", "is X better than Y", "difference between X and Y", "X or Y for Z") where the answer is a two-sided judgment. Twice daily through May 2026 we captured every AI Overview card, and for cards citing a client page on a comparison query we identified the specific lifted sentence and classified its shape: explicit comparative (both entities named in the main clause plus a stated direction), neutral two-sided (both named, no direction), or one-sided (only one entity named). For each cited sentence we built a matched control: a comparable sentence on a similar comparison query whose shape differed but whose underlying claim was the same, so the comparison was comparative-vs-one-sided rather than good-page-vs-bad-page. The cited cohort was 7,540 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 one-sided sentence that reads as an obvious comparison inside a page titled "PostgreSQL vs MySQL" reads as a one-entity description 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 comparative shape is not just the comparative pages being better written overall. The 2.4× and 1.8× figures are from those matched comparisons, not raw averages.
The shape of the comparison pattern
The flat headline first. On comparison queries, explicit comparative sentences are cited more. A sentence that named both entities and stated a direction was lifted 2.4× more often than a matched one-sided sentence on the same comparison query. The effect held through the quality match and the citability control: among sentences our rubric scored as equally liftable, the comparative ones were lifted far more than the one-sided ones. The composer behaves as though it prefers a sentence that answers the whole two-sided question over one that answers only the side the page happens to sell.
The most decision-relevant cut was that this is about answering the whole question, not about mentioning a competitor for its own sake. We tested whether the win came from naming the second entity or from stating a direction, and both mattered but the direction was the larger half: a neutral sentence that named both sides without resolving the comparison ("PostgreSQL and MySQL both handle concurrent writes") was cited better than a one-sided sentence but worse than a directional one ("PostgreSQL handles concurrent writes better than MySQL"). Naming both sides is necessary; stating which side wins on the dimension is what the composer rewards. Write the comparison as a verdict, not a list of two contenders.
Driver one: name both compared entities in the answer sentence
The single strongest predictor was whether the answer sentence named both sides of the comparison. Holding the claim constant, a sentence that named both entities was lifted at 2.4× the rate of one that named only the page's own product. The composer extracts a sentence and reads it as the answer to "X vs Y"; a sentence that names only X is, in isolation, an answer to "is X good", and a composer matching answer to query treats it as off-question for the comparison. A human reader supplies the missing side from the page's title and topic; the composer reading the sentence cold cannot, so the second entity has to be in the sentence itself.
We ran a structural test on 28 answer sentences across 15 clients, each a one-sided sentence on a comparison query that praised the client's product and left the rival unnamed. We rewrote each to name both entities in the main clause, changing no underlying claim — only promoting the rival into the sentence. Over the 45 days that followed, 20 of the 28 sentences began being lifted on at least one comparison query where the one-sided version had been skipped. The lever was not new content; it was putting both sides of the comparison into the single sentence the composer would extract, so the sentence answered the two-sided query on its own.
Driver two: state the direction, do not list two neutral contenders
Holding both-sides naming constant, the second driver was whether the sentence resolved the comparison. A sentence that named both entities and stated which won on the dimension ("PostgreSQL handles concurrent writes better than MySQL") was cited more than one that named both and stayed neutral ("PostgreSQL and MySQL both support concurrent writes"). The reading consistent with the data is that a comparison query asks for a judgment, and the composer prefers a sentence that delivers one over a sentence that restates the contest without settling it — a neutral both-sides sentence answers "what are the two options", not "which is better".
We ran a structural test on 21 neutral two-sided sentences across 12 clients, each naming both entities but withholding a verdict. We rewrote each to state the direction on a specific dimension, changing no facts — only adding the resolution the comparison implied. Over the 60 days after the change, 14 of the 21 sentences improved their cited-passage rate. The two drivers compound: a one-sided sentence that names neither rival nor direction is the worst case, a neutral sentence that names both but resolves nothing sits in the middle, and a directional sentence that names both and states the winner on the dimension is the best — the composer wants the verdict, scoped, in the one sentence it lifts.
Driver three: unscoped superlatives, and the verdict with no dimension
The third driver was the warning. A direction is only useful when it is scoped to a dimension, and flipping every comparison to a flat "X is better than Y" with no stated dimension backfires. A sentence like "PostgreSQL is better than MySQL" — true on some dimension, false on others, and silent about which — was cited no more often than the neutral both-sides version, and on 6% of audited pages the composer paired the bare superlative with a competitor's sentence that supplied the missing dimension ("…for analytical workloads"), so the citation was shared rather than won outright. The reading consistent with the data is that the composer rewards a verdict the reader can trust, and an unscoped superlative is a verdict it cannot place against the specific comparison query, so the composer treats it as an opinion rather than an answer. A directional sentence still has to name the dimension the direction holds on.
We confirmed this on 16 sentences across 10 clients where an earlier optimisation pass had added flat superlatives without dimensions. We rewrote each to scope the verdict to the dimension the query asked about ("PostgreSQL handles concurrent writes better than MySQL", not "PostgreSQL is better than MySQL"), keeping the direction but adding the scope. Over the following 45 days the scoped sentences regained their solo citation while reading as a defensible judgment, and none drew a shared-citation pairing. The actionable rule is blunt: name both sides and state a direction, but the direction must be scoped to the dimension the comparison query asked about — a winner with no stated dimension reads as an opinion the composer will not lift alone.
What changed in our content checklist
Three changes. We added a both-sides pass for comparison queries: before publishing, we read each comparison section's lead answer sentence and check that both compared entities appear in the main clause, and a sentence that names only our own product gets the rival promoted into it — because the composer lifts a sentence whole and reads a one-entity sentence as off-question for a "vs" query. We added a verdict check to the same pass: the sentence has to state which side wins on the dimension, not list two neutral contenders, because the composer prefers a judgment over a restated contest. And we added a scope guard: every direction names the dimension it holds on, so a flat superlative gets pinned to the specific axis the query asked about.
We dropped one habit. For years our comparison pages had been built as one-sided cases — the page sells PostgreSQL, so it describes PostgreSQL and treats the rival as an afterthought, because the page's job is to win the reader, not umpire a draw. The audit removes that default for the answer sentence: the one sentence the composer would lift on a "vs" query has to carry both sides and a scoped verdict, and a one-sided sentence spends the citation to keep the page on-message. So one-sided answer sentences left our playbook for comparison queries — we now write the lead answer sentence to name both entities and state the scoped direction, and let the rest of the section make the one-sided case, accepting that the cited sentence reads more even-handed than a sales page would choose because it is built to answer the two-sided query alone.
- 01Name both compared entities in the answer sentence. A sentence naming both was cited 2.4× more than a one-sided one on the same comparison query — the composer reads a one-entity sentence as off-question for "X vs Y".
- 02State the direction, do not stay neutral. A sentence that resolved the comparison was lifted more than one that listed both sides without a verdict — a comparison query asks which is better, not what the two options are.
- 03Scope the verdict to a dimension. A flat "X is better than Y" was cited no more than a neutral sentence, and on 6% of pages the composer shared the citation with a competitor that supplied the missing dimension.
- 04Promote the rival into the sentence. 20 of 28 one-sided sentences were lifted after both entities were named in the main clause and the direction was stated.
Where this argument breaks
For single-entity queries — "what is PostgreSQL", "how does MySQL handle writes" — there is no second side to name and the comparative shape is irrelevant, so the lever is for queries that ask for a comparison. 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 rival and stating a verdict is an editorial and positioning choice, not a citation lever, and the both-sides pass is for the answer sentences on comparison queries only. For some languages the effect may differ — in our parallel Chinese-language audit (文心一言, 元宝, 通义) the both-sides-naming win was present but the direction-stating win was smaller, since Chinese comparison constructions often carry the direction in a marked structure (「比」) that the composer read reliably even when the verdict was understated. The 6% shared-citation figure is small and noisy; we are confident an unscoped superlative 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 comparative answer sentence — both entities named, a direction stated, the direction scoped to a dimension — far more readily than a one-sided sentence that praises only the side the page sells, the unit is the individual answer sentence rather than the page, and the cheapest citation win on a comparison query is to write the answer as a scoped verdict that names both contenders.