What is Candidate set?
The candidate set is the chunk-level shortlist an AI search engine's retrieval stage returns to the rerank stage per sub-query — typically 40–120 chunks on Google AI Mode and Microsoft Copilot, 60–90 on ChatGPT Search, 40–80 on Perplexity, and 30–60 on Claude in mid-2026. The candidate set is not published by any engine but is inferrable from two signals: competitor candidate-set membership (chunks the engine rotates through over a 4–8 week window even when not cited on a given day) and chunk-pattern competitor analysis (chunks with the same shape, entity density, and claim specificity as the cited set are very likely in the candidate set on the same sub-query). The candidate set is the upstream pool the reranker prunes against; rerank survival rate measures the fraction of the candidate set that survives into the synthesis prompt.
How it relates to AI UGC
Multimodal-active sub-queries surface a parallel image candidate set — image chunks indexed by the multimodal substrate and reranked by the visual cross-encoder pass. The visual candidate set is typically 20–60 image chunks per sub-query, pruned to 3–6 by the visual reranker. ppl.studio production cadence keeps persona-locked image chunks in the visual candidate set on the 4–12 week image freshness window the substrate scans against.
Key statistics
- Mid-2026 candidate-set size per sub-query: 80–120 on Google AI Mode, 60–90 on ChatGPT Search, 40–80 on Perplexity, 30–60 on Claude — the size gap is one of the structural reasons rerank survival rates vary by engine (candidate-set audits, 2026).
- Inferred candidate-set membership lands within ±30% of the true set on the typical chunk-pattern competitor analysis — enough precision to drive editorial priority on the rerank-edit backlog (inference-precision audits, 2026).
- Candidate-set membership without rerank survival is the largest single source of misread 'retrieval health' — programs that score candidate-set inclusion alone report 60–80% coverage while only 12–22% of those chunks survive the rerank pass (candidate-vs-survival audits, 2026).