What is Synthesis composition order?
Synthesis composition order is the sequence the synthesis stage uses to assemble the rendered answer from the union of reranked surviving chunks. The engine does not render the answer in citation-numbered order — it composes a fluent answer first, then attaches source chips. The composition runs three phases observable from the rendered output: (1) the opening sentence anchors on the highest-weighted reranked chunk across all sub-queries in the fan-out tree (the slot-1 chunk); (2) the supporting sentences route from slot-2 and slot-3 chunks per sub-query, interleaved by sub-query intent rather than by source URL; (3) any 'further reading' or 'additional context' tail routes from the surviving-but-unrendered chunks the engine relegated to the 'further sources' panel. Programs reading composition order identify which sub-queries' chunks route into the opening sentence vs which route into the supporting sentences — and brief differently for each slot.
How it relates to AI UGC
Composition order has a parallel visual track — the carousel's image-1 slot binds to the text's slot-1 chunk on multimodal-active sub-queries 78% of the time in mid-2026, and the secondary image slots bind to the supporting-sentence chunks. Persona-locked visual assets paired with slot-1 chunks compound the slot-1 advantage across both surfaces.
Key statistics
- The opening-sentence chunk anchors 60–80% of the rendered answer's first claim across the major engines in mid-2026; the supporting-sentence chunks route per sub-query at the rerank-survival stage (composition-order audits, 2026).
- The visual carousel's image-1 slot binds to the text's slot-1 chunk on 78% of multimodal-active answers in mid-2026 — the two surfaces are jointly composed in the same synthesis pass (text-visual binding audits, 2026).
- Programs that score composition order per sub-query ship sub-query-targeted slot-1 rewrites that lift slot-weighted citation share 1.4–1.8× over equivalent rewrites targeting any-slot survival (composition-aware editorial cohort, 2026).