grouped_webpages / search_result_groups reference/grouped-webpages

What it is

The results of a search are in metadata.search_result_groups: one group per domain with entries, each with ref_id (turn0search5), title, URL, excerpt, attribution, date. In the answer text, content_references of type grouped_webpages point to these entries via ref_id. Measured 22 Aug 2026: 14–22 URLs per searching question, 9–16 domains, 1.2–2.5 URLs per domain (search.urls_per_domain). The structure dates from the rebuild of 21 Jul 2026 (Mohanadasan), when result_source vanished and reference IDs became objects.

External evidence

• Suganthan Mohanadasan, 14 Jul 2026: rebuild of the structure on 21 Jul.
• Suganthan Mohanadasan, 10 Aug 2026: citation rate by position in the domain group.

For visibility

All candidates are here. A page in this list has cleared the first hurdle (fan-out) and not yet the third (footnote). Grouping by domain means several pages of the same domain compete with each other and usually only one makes the footnote. Excerpt length in the entries reveals the pipeline.

External sources

mohanadasan-2026-07-14 · mohanadasan-2026-08-10

Change history

No event has touched this object yet.

Open

• Why do hits from the in-house index often carry no snippet (K1-01 instant 22 Aug: 14 of 14 without)?
• Mohanadasan: position 1 in the domain group 5.2 % citation rate, position 6+ 0.3 %. Not computed for our data.

Related

sources_footnote url (inline reference) Retrieval pipelines (Labrador, scrape) open (opened results)