**TL;DR** — Across 30 client sites through May 2026 we audited a structural choice that lives in the polarity of the answer sentence: whether the passage that answers a query is written as a **positive sentence** that states what the thing is ("Lazy loading defers off-screen images until the user scrolls near them, cutting initial page weight by about 40%.") or as a **negated sentence** that states what it is not ("Lazy loading does not load off-screen images on first paint, so it does not add to initial page weight."), and whether the negation changes how often the AI Overview lifts that sentence into the card. Across 7,540 cited-passage events we joined each cited sentence to whether its main clause was affirmative or negated. The headline is that the negated sentence is a real and large citation drag on most queries, and it is really an answer-directness lever wearing a "not" costume. A positive answer sentence was cited 2.2× more often than a matched negated sentence stating the same fact in reverse on the same query. The strongest predictor was claim-polarity — a sentence that asserted what is true was lifted far more than one that ruled out what is false and left the reader to infer the rest. The second was double-negative scope — a sentence carrying one negation was lifted more than one stacking "not" and "unless" into a clause the composer had to unwind. The third, and the exception, was the corrective query — on misconception and "is it true that X" queries the negated sentence won, because there the negation is the answer. One change — rewriting negated answer sentences on direct queries into the positive assertion of the same fact — lifted cited-passage rate by 21% 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 most queries ask what something is, not what it is not. "What does lazy loading do" wants the affirmative — the thing it does — and a sentence that answers by ruling out what it does not do hands the composer a true statement that still leaves the actual answer to be inferred. A human reading the surrounding page assembles the positive picture from a paragraph of negations easily; the composer extracting one sentence cannot, and a sentence that says only "it does not add page weight" is, to it, an answer that dodges the question it was asked. We had spent weeks on the shape of the answer sentence — its mood, its voice, whether it named a condition — and polarity is the natural next structural variable, because a negated sentence is the most common way a confident, true statement still fails to answer the question directly.
The second motivation was a writing habit that reaches for negation. Technical writing loves the corrective frame — it answers "does X slow my site" with "X does not slow your site", and it qualifies claims by what they exclude rather than what they assert, because a negation feels precise and safe. A human reader converts "does not add page weight" into "keeps page weight flat" without noticing the work. The composer, hunting for one sentence that answers "what does X do", finds the negated sentence and has to decide whether a statement of what does not happen answers a request for what does. We needed to know whether the negated mood cost the citation, because if it did, the fix is nearly free — flip the sentence to its positive assertion — and it costs only a habit.
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, weighted toward direct queries ("what is X", "what does X do", "how does X work") where the answer is an affirmative fact, with a deliberate minority of corrective queries ("is it true that X", "does X really Y", "X myth") where the answer is a negation. Twice daily through May 2026 we captured every AI Overview card, and for cards citing a client page we identified the specific lifted sentence and classified its polarity: positive (the main clause asserts what is true), negated (the main clause is built on "not", "no", "never", "does not", "cannot"), or mixed. For each cited sentence we built a matched control: a comparable sentence on a similar query whose polarity differed but whose underlying fact was the same, so the comparison was positive-vs-negated rather than good-page-vs-bad-page. The cited cohort was 7,540 events.
Two normalisation moves matter. We scored polarity on the sentence as it would be lifted — alone, with no surrounding context — because that is the unit the composer extracts, and a negated sentence that reads as an obvious correction inside a "common myths" section reads as a bare contradiction in the card. And we matched on sentence citability before comparing polarity — 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 polarity is not just the positive pages being better written overall. The 2.2× and 1.7× figures are from those matched comparisons, not raw averages.
The shape of the polarity pattern
The flat headline first. On direct queries, positive sentences are cited more. A sentence that asserted what is true was lifted 2.2× more often than a matched negated sentence stating the same fact in reverse on the same query. The effect held through the quality match and the citability control: among sentences our rubric scored as equally liftable, the positive ones were lifted far more than the negated ones. The composer behaves as though it prefers a sentence that states the answer over one that states the answer's complement and leaves the reader to take it from there.
The most decision-relevant cut was that this is about answering directly, not about avoiding the word "not". We tested whether the drag came from negation itself or from the answer being one inference away, and the second was the story: a negated sentence whose negation was the answer ("No, gzip does not encrypt your traffic") was cited fine on the query it answered, while a negated sentence used to state a positive fact indirectly ("Caching does not refetch unchanged assets") was passed over for a positive one ("Caching reuses unchanged assets from local storage") that said the same thing forward. Negation drags when the positive fact is what the query wants. State the answer, do not state its opposite and trust the reader to flip it.
Driver one: assert what is true, not what is false
The single strongest predictor was the claim's polarity. Holding the underlying fact constant, a sentence that asserted the positive was lifted at 2.2× the rate of one that ruled out the negative. The composer extracts a sentence and reads it as the answer to "what does X do"; a sentence that says "lazy loading keeps initial page weight flat" answers the question, while one that says "lazy loading does not add to initial page weight" requires the composer to compute the positive from the negative — and on a direct query it would rather lift the page that already did that computation. A human reader flips the polarity for free; the composer matching a direct query rewards the sentence that hands over the affirmative.
We ran a structural test on 28 answer sentences across 15 clients, each a negated statement on a direct query that named what the thing did not do. We rewrote each into the positive assertion of the same fact, changing no underlying answer — only flipping the polarity to state what is true. Over the 45 days that followed, 20 of the 28 sentences began being lifted on at least one direct query where the negated version had been skipped. The lever was not new content; it was rewriting the single sentence the composer would extract so it asserted the answer instead of its complement.
Driver two: one negation, not a double negative
Where a negation was genuinely needed, the second driver was how many the sentence carried. A sentence with one negation ("Free plans do not include SSO") was cited more than one stacking two ("It is not true that no free plan lacks SSO") or pairing "not" with "unless" ("This does not work unless you are not on the legacy tier"). The reading consistent with the data is that the composer lifts a sentence as one self-contained answer, and a single clean negation resolves to a clear claim, while a double negative is a clause the composer has to unwind before it can tell what is being asserted — so it skips it for a sentence whose meaning is legible on one pass.
We ran a structural test on 17 double-negative sentences across 10 clients, each forcing the reader to cancel two negations to reach the meaning. We rewrote each into a single positive or single-negation statement, changing no facts — only removing the second negation. Over the 60 days after the change, 12 of the 17 sentences improved their cited-passage rate. The two drivers compound: a double-negative statement of a positive fact is the worst case, a clean positive assertion is the best, and even a necessary negation should be the only one in the sentence so the composer can read the claim in a single pass.
Driver three: the corrective query, where the negation is the answer
The third driver was the exception. Negation wins when the negation is the answer the query asks for. On corrective queries — "is it true that gzip encrypts traffic", "does HTTP/2 require HTTPS", "X myth" — the user is asking whether something false is true, and the honest answer is a negation. There the negated sentence ("No, gzip does not encrypt your traffic; it only compresses it") was cited more than a positive sentence that stated the true fact without addressing the misconception ("gzip compresses your traffic"), because only the negation answers the question that was asked. The reading consistent with the data is that the composer matches polarity to intent: a direct query wants the affirmative, and a corrective query wants the contradiction stated plainly, ideally with the positive fact attached.
We confirmed this on 14 sentences across 9 clients answering corrective queries, where an earlier optimisation pass had flipped the corrections into positive statements and lost the question. We rewrote each back into a clear negation that named the misconception and corrected it ("No — X does not Y; it Z"), matching polarity to the query's corrective intent. Over the following 45 days the corrections regained their citation on the "is it true" queries while reading directly. The actionable rule is blunt: assert the positive for a direct query and state the negation for a corrective one — the polarity has to match what the question is asking, and a correction flipped into a bare positive stops answering the misconception the query raised.
What changed in our content checklist
Three changes. We added a polarity pass for direct queries: before publishing, we read each section's lead answer sentence and check that it asserts what is true, and a sentence that answers a direct query by stating what the thing is not gets flipped to the positive assertion — because the composer lifts a sentence whole and reads a positive sentence as the direct answer to "what does X do". We added a single-negation check to the same pass: where a negation is needed, it is the only one in the sentence, with double negatives rewritten, because the composer cannot read a claim it has to unwind twice. And we added a corrective-query guard: on "is it true that X" queries we keep the negation, because there the contradiction is the answer and flipping it to a positive drops the misconception the query raised.
We dropped one habit. For years our pages had reached for the corrective frame on direct queries, on the belief that ruling out the wrong answer was a precise and safe way to state the right one — "X does not add page weight" felt more careful than "X keeps page weight flat". The audit removes that default for the answer sentence on direct queries: the one sentence the composer would lift has to assert the answer, and a negated sentence spends the citation to sound careful. So negated answer sentences left our playbook for direct queries — we now write the lead answer sentence as the positive assertion and reserve negation for the corrective queries where it is the answer, accepting that the cited sentence reads more plainly declarative than a careful editor would default to because it is built to answer the question forward.
- 01Assert what is true. A positive answer sentence was cited 2.2× more than a negated one stating the same fact in reverse on the same direct query — the composer reads a positive sentence as the answer and a negated one as one inference short of it.
- 02One negation, never two. A single clean negation was lifted more than a double negative — the composer skips a clause it has to unwind before it can tell what is asserted.
- 03Keep the negation for corrective queries. On "is it true that X" and myth queries the negation is the answer, and flipping it to a positive drops the misconception the query raised.
- 04Flip the negated statement. 20 of 28 negated answer sentences were lifted after being rewritten as the positive assertion of the same fact.
Where this argument breaks
For corrective and misconception queries the negation is the answer and the lever inverts, so the polarity pass is for direct queries where the answer is an affirmative fact. For navigational and brand queries there is no answer sentence whose polarity matters. For narrative and persuasive passages — case studies, opinion, story-driven content — negation is a rhetorical choice serving the prose, not a citation lever, and the polarity pass is for the answer sentences on direct queries only. For some languages the effect may differ — in our parallel Chinese-language audit (文心一言, 元宝, 通义) the positive-over-negated win was present but smaller, since Chinese marks negation with a single preverbal particle («不», «没») that the composer parsed cleanly even in stacked form, so the double-negative penalty barely appeared. The corrective-query inversion is the strongest carve-out and we are confident in it; the double-negative figure is the weakest finding 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 positive answer sentence — what is true, asserted forward, with at most one negation — far more readily than a negated sentence that states the same fact by ruling out its opposite, the unit is the individual answer sentence rather than the page, and the cheapest citation win on a direct query is to flip the answer from what the thing is not to what it is.