What is Paraphrase citation?
A paraphrase citation is the synthesis-stage outcome where the engine renders the chunk's substance in its own voice — no quoted span — but still attaches a numbered source chip to the paraphrased sentence. Paraphrase citations are the middle disposition: better than being dropped from the rendered answer entirely (source relegated to 'further sources'), but materially weaker than verbatim citation on the metrics that matter (click-through, brand recall, perceived authority). Paraphrase happens when the chunk supplies the claim but the synthesis prompt prefers a smoother answer voice over the chunk's exact phrasing — common when the chunk is wordy, when the claim shape doesn't match the answer's shape, when the chunk needs surrounding context the prompt does not have, or when the chunk's lexical distinctiveness is low (the engine can compress without loss). Programs that score citation disposition per chunk identify which chunks are paraphrased rather than verbatim and brief targeted leading-sentence rewrites to lift the verbatim rate.
How it relates to AI UGC
Paraphrased citations weakly bind the inline-carousel image selection — the engine selects a visual asset for the paraphrased sentence at materially lower rates than for the verbatim quote. Programs that lift the verbatim rate also lift the multimodal-answer share on the same sub-queries; paraphrase citations leave the visual carousel partially unbound.
Key statistics
- Paraphrase citation rate sits at 28–40% of rerank-surviving chunks on mid-market programs; category-leading programs compress it to 22–32% as verbatim citation rate rises (citation-disposition cohort, 2026).
- Click-through on paraphrase citations runs 35–48% of verbatim citation click-through at equivalent answer-position because no quoted span anchors the user's eye to the cited source (paraphrase-vs-verbatim CTR audits, 2026).
- Roughly 60% of paraphrased chunks can be lifted to verbatim with a single leading-sentence rewrite that names the claim's numeric and the entity in the sentence shape the synthesis prompt prefers (paraphrase-to-verbatim conversion audits, 2026).