**TL;DR** — Across 30 client sites through May 2026 we audited a structural choice that lives in whether the answer sentence defines its own key term: whether the passage that answers a query is written as an **appositive-defined sentence** that folds a short inline definition of the central term into the same sentence as a noun-phrase aside ("RAG, a method that grounds a model in retrieved documents, cuts hallucination by letting the model quote sources instead of inventing them.") or as an **undefined-term sentence** that names the term and assumes the reader already knows it ("RAG cuts hallucination by letting the model quote sources instead of inventing them."), and whether defining the term in place 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 it defined its own key term inline. The headline is that the appositive is a real citation lever, and it is really a self-containedness lever wearing a grammar costume. An answer sentence that defined its key term in an appositive was cited 1.9× more often than a matched sentence that named the same term undefined on the same query. The strongest predictor was term-defined-in-sentence — a sentence a reader could understand without knowing the term first was lifted far more than one that assumed the term. The second was appositive brevity — a tight six-to-twelve-word aside was lifted more than a subclause that ran to a second sentence's worth of definition. The third, and the warning, was over-definition — appending an appositive to a term the query itself already defined ("What is RAG" answered with "RAG, a retrieval method, is …") was cited no more, and on 5% of pages the composer paired the padded sentence with a competitor's cleaner one. One change — folding a short inline definition of the key term into answer sentences that had assumed it — lifted cited-passage rate by 19% 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 that sentence has to stand on its own in front of a reader who has none of the page around it. A whole class of answer sentences fails that test for one reason: they name a term the surrounding page defined three paragraphs up and assume the reader carries the definition forward. "RAG cuts hallucination by grounding the model" is a complete answer to someone who has read the section headed "What RAG is"; lifted cold into a card in front of a user who typed "how do I reduce model hallucination", it names a mechanism ("RAG") the user may never have seen, and the card answers the question in terms the reader cannot parse. We had spent weeks on the shape of the answer sentence — its count, its polarity, whether it named a cause — and self-definition is the natural next structural variable, because a sentence that assumes a term is one the composer cannot make self-contained no matter how liftable the rest of it reads.
The second motivation was a writing habit that defines once and never again. A page introduces its central term in a definitional opener, then uses the bare term for the rest of the article — which is correct style, because repeating the definition in every sentence would read as padding to a human moving top-to-bottom. But the composer does not move top-to-bottom; it reaches into the middle of the page, lifts the one sentence that best matches the query, and that sentence is usually deep in the body where the term has long since gone undefined. We needed to know whether the answer sentence carrying its own compact definition — as an appositive that costs a human reader almost nothing — bought the citation on queries where the term is the mechanism, because if it did, the fix is nearly free: fold a six-word aside into the sentence the composer would lift, and it costs only a definition that reads slightly redundant to someone who read the opener.
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 queries whose best answer sentence turns on a domain term (a technique, an acronym, a named concept) the user might not know. 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 shape: appositive-defined (the key term carries an inline noun-phrase definition set off by commas), gloss-defined (a defining relative clause attached to the term), or undefined (the term named with no in-sentence definition). For each cited sentence we built a matched control: a comparable sentence on a similar query whose shape differed but whose underlying claim was the same, so the comparison was defined-vs-undefined 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 an undefined-term sentence that reads as obviously complete inside a section headed "What RAG is" reads as an unparseable 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 definition is not just the defined pages being better written overall. The 1.9× and 1.5× figures are from those matched comparisons, not raw averages.
The shape of the appositive pattern
The flat headline first. A sentence that defined its own key term was cited more. An answer sentence carrying an inline appositive definition of its central term was lifted 1.9× more often than a matched sentence that named the same term undefined on the same query. The effect held through the quality match and the citability control: among sentences our rubric scored as equally liftable, the self-defining ones were lifted far more than the term-assuming ones. The composer behaves as though it prefers a sentence a reader could understand cold — one that names its mechanism and tells you what the mechanism is in the same breath — over one that names a term the reader outside the page has no way to resolve.
The most decision-relevant cut was that this is about the sentence standing alone, not about the appositive being grammatically present. We tested whether the win came from a comma-set aside appearing or from the aside actually defining the term, and the second was the whole story: a sentence with a decorative aside that added no definition ("RAG, our favourite technique, cuts hallucination") was cited no better than a bare one, while a sentence whose aside genuinely told the reader what the term was ("RAG, a method that grounds a model in retrieved documents, cuts hallucination") was lifted far more. The appositive wins when it makes the sentence self-contained. Define the term, do not just decorate it.
Driver one: let the sentence be understood without the term
The single strongest predictor was whether a reader could understand the answer sentence without already knowing the key term. Holding the claim constant, a sentence that defined its term inline was lifted at 1.9× the rate of one that assumed it. The composer extracts a sentence and puts it in front of a user who did not read the page; a sentence that says "RAG, a method that grounds a model in retrieved documents, cuts hallucination" answers the question and teaches the term at once, while one that says "RAG cuts hallucination" answers it in a vocabulary the user may not have — and a card that answers in unknown terms is, to the composer, a worse answer than one that defines as it goes. A human reading the page in order already knows what RAG is; the reader in front of the card does not, and the composer rewards the sentence that assumes nothing.
We ran a structural test on 27 answer sentences across 14 clients, each a strong claim on a query where the sentence named a domain term the page had defined earlier and not since. We rewrote each to fold a short inline definition of that term into the sentence as an appositive, changing no underlying claim — only lifting the definition that lived in the opener down into the answer sentence. Over the 45 days that followed, 19 of the 27 sentences began being lifted on at least one query where the undefined version had been skipped. The lever was not new content; it was moving the definition the page already carried into the single sentence the composer would extract, so the sentence explained its own term.
Driver two: a tight aside, not a nested definition
Holding term-defined constant, the second driver was how long the appositive ran. A tight six-to-twelve-word aside ("RAG, a method that grounds a model in retrieved documents, …") was cited more than one that swelled into a second sentence's worth of definition ("RAG, which is a technique that, by retrieving documents from a vector store and passing them into the prompt so the model can condition on them rather than on its parameters alone, …"). The reading consistent with the data is that the composer lifts a sentence as one self-contained unit, and a compact appositive keeps the sentence readable as a single answer, while an over-nested definition turns the sentence into two answers wearing one full stop — so the composer skips it for a sentence whose aside defines the term in a breath and then gets back to the claim.
We ran a structural test on 17 over-nested sentences across 10 clients, each carrying an appositive that ran past a dozen words and buried the claim behind its own definition. We rewrote each to cut the aside to its essential noun phrase and push the fuller definition to a following sentence, changing no facts. Over the 60 days after the change, 12 of the 17 lead sentences improved their cited-passage rate. The two drivers compound: an undefined term is one failure mode and a definition that eats the sentence is the other — the sentences that won defined the term in a short aside and returned to the claim before the reader lost the thread.
Driver three: over-definition, and the term the query already knew
The third driver was the warning. An inline definition helps only when the reader needs it, and appending one to a term the query itself already names backfires. A sentence like "RAG, a retrieval-augmented generation method, is a technique for grounding models" — a definition bolted onto a "what is RAG" query, where the user has already shown they know the term by naming it — was cited no more often than the clean definition ("RAG grounds a model in retrieved documents so it can quote sources instead of inventing them"), and on 5% of audited pages the composer paired the padded sentence with a competitor's that defined the term once and moved on, so the citation was shared rather than won. The reading consistent with the data is that the composer discounts a definition the query did not need; a user who typed the term does not need it re-defined mid-sentence, and the redundant aside reads as padding the composer will pass over. Define the term when the query assumes it; state it plainly when the query already names it.
We confirmed this on 14 sentences across 9 clients where an earlier optimisation pass had added appositive definitions to answers for queries that already named the term. We rewrote each back into a clean statement for the term-naming query while keeping the appositive on the adjacent queries whose users had not shown they knew the term, matching the definition to what each query assumed. Over the following 45 days the plain statements regained their solo citation while reading directly, and none drew a shared-citation pairing. The actionable rule is blunt: define the term inline when the query assumes it, and state it plainly when the query already names it — a needless appositive reads as a qualifier the composer will pass over for a cleaner sentence.
What changed in our content checklist
Three changes. We added a self-containedness pass for answer sentences: before publishing, we read each section's lead answer sentence alone, stripped of the page, and check that a reader who has not seen the term could still understand it — because the composer lifts a sentence whole and hands it to a user who did not read the page, so a term the sentence assumes is a term the card cannot resolve. We added a brevity check to the same pass: the appositive is a short noun phrase, not a nested definition, so the sentence stays readable as one answer. And we added an over-definition guard: we strip appositives off answers to queries that already name the term, so a needless definition never clutters a sentence whose user has already shown they know it.
We dropped one habit. For years our style had been to define a term once, in its opener, and use the bare term everywhere after — on the belief that repeating the definition read as padding and that a reader moving through the page carried the meaning forward. The audit removes that default for the answer sentence: the one sentence the composer would lift lives deep in the body where the term has long gone undefined, and it has to carry its own compact definition or the card answers in terms the reader cannot parse. So term-assuming answer sentences left our playbook for queries where the term is the mechanism — we now fold a short inline definition into the lead answer sentence, accepting that it reads slightly redundant to someone who read the opener, because it is built to be understood by the reader who did not.
- 01Define the key term inline. A sentence that defined its term in an appositive was cited 1.9× more than one naming the same term undefined on the same query — the composer reads a self-contained sentence as an answer and a term-assuming one as an unparseable claim.
- 02Define, do not decorate. A comma-set aside that added no definition was lifted no more than a bare sentence — the win comes from the aside actually telling the reader what the term is.
- 03Keep the appositive tight. A six-to-twelve-word aside was cited more than a nested definition that ran to a second sentence's worth and buried the claim.
- 04Do not over-define. Bolting a definition onto a query that already names the term was cited no more, and on 5% of pages the composer shared the citation with a competitor that stated it plainly.
Where this argument breaks
For queries whose answer names no domain term — plain-language questions whose every word the user already owns — there is nothing to define and the appositive is irrelevant, so the lever is for answer sentences that turn on a term the user might not know. For "what is X" queries where the term is the subject the query already assumes it, and defining it again mid-sentence is the over-definition failure, not the win. For navigational and brand queries there is no answer sentence whose shape matters. For narrative and persuasive passages — case studies, opinion, story-driven content — defining a term inline is a clarity choice, not a citation lever, and the self-containedness 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 self-definition win was present but the appositive was carried less by comma-set asides and more by a leading «所谓 X，是指……» frame the composer read reliably. The 5% shared-citation figure is small and noisy; we are confident a needless appositive 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 self-defining answer sentence — the key term defined inline in a short appositive, used only where the reader would not already know the term — more readily than one that names the term and assumes it, the unit is the individual answer sentence rather than the page, and the cheapest citation win on a term-carrying query is to fold a compact definition into the sentence that makes the claim.