2. Articles indexed without body text — Critical (PV)
What: Google stores the server HTML without running JavaScript, so the article is indexed with headline/metadata but no body; body sentences are not searchable. Search Console still shows "URL is on Google / Page is indexed".
Evidence
-
Verified in Search Console indexed HTML (paragraph containers empty, identical to raw HTML): JDS
…-4331297, Ration card eKYC…-4251363, Mangaluru…-4331026. Control Muhurat…-4342287: indexed HTML filled, body searchable. -
Exact-phrase search test, 800 random articles (400/site, 3–120 days old): headline found but body not — PV 167/383 (44%), DH 78/357 (22%).
-
site:re-test of those "body not found" articles: PV 0/15 found (confirmed not indexed); DH 14/15 found (original search failed on ranking vs wire copies, not indexing). -
Estimated share of articles indexed without body: PV ≈ 35–44%; DH ≈ 0–7%. PV revised 16 Nov 2026 to ≈ 76% (below). -
16 Nov 2026, PV site:-restricted test (Serper, 711 paid queries;
checks/body_index_check/funnel_site_test.py, resultsfunnel_site_test_results.jsonl,funnel_site_test_verdicts.jsonl,funnel_site_test_report.md). 663 indexed PV stories (URL Inspection), one rare-word 5-word body phrase each in quotes withsite:prajavani.net; zero results = body not indexed; other PV pages only = retry with a second phrase.Group Stories Body not indexed (share of resolved) Never-shown print copies, 8 weekly batches, published 7 Sep – 1 Nov 2026 399 295 of 382 (77%; weeks 67–87%) Print copies with impressions 100 71 of 94 (76%) Web-desk textstories with impressions100 69 of 97 (71%) Older: Jul–Aug 2026 print / text 64 28 of 35 (80%) / 19 of 26 (73%) All tested (pooled; sample over-weights never-shown stories) 663 482 of 634 (76%) - Share of all PV articles: weighted to PV's output 7 Sep – 5 Nov 2026 (never-shown print ≈ 10,970 stories at 77%, shown print ≈ 19,200 at 76%,
text≈ 4,750 at 71%;checks/traffic_funnel/joined_pv_2026-11-13.json): ≈ 76% of PV print and web-desk articles, which are ~91% of output. Not tested:syndicatedwire stories (~6%) and small templates (cartoon, video, photo, listicle, horoscope, exclusive); same article code (#1), so likely similar. - Method checks: calibration against the user's 12 manual searches: 11 of 12 agree (the miss: a common phrase where other PV pages outranked the story). Positive control: for 59 "not indexed" stories a subheadline (2) or headline (57) phrase (server-rendered) found the story 57 times (97%), so exact-phrase matching on Kannada works and only the body is missing.
- Readings: (1) the share is the same for stories Google shows and never shows, so PV ranks almost entirely on headlines; (2) Jul–Aug stories are no better than recent ones: Google does not fill in the body later; (3) the earlier 35–44% (search without
site:, less distinctive phrases) under-counted. - Before print-to-web (Jan 2026; 50 credits, 41 stories resolved,
funnel_site_test_verdicts.jsonlstratum "older (Jan 2026)"): body not indexed 41% (17 of 41) vs 71–80% for Jul–Nov 2026. Holds within length (<150 words: Jan 27% vs Sep–Nov 67%; 150+ words: 58% vs 91%) and within thetexttemplate (46% vs 71–72%), so it is not a template or length effect. Same code in both periods (#1 since 8 Oct 2024). Likely explanations: (1) output rose from ~400 to ~650 stories/day when Varta Setu (automatic print-to-web) started (late Mar 2026), so a smaller share of PV pages gets rendered by Google; (2) slow re-rendering over ~10 months (Jul–Aug stories are no better than Sep–Nov, which argues against it). Small sample (±~15 points). Separating test: ~50 stories from early Mar 2026 vs ~50 from mid-Apr 2026 (same age, before/after print-to-web), ~100 credits. - After #1 ships: re-run
runon the same stories (funnel_site_test_sample.json; ~200 Serper credits kept for this) ~4 weeks after release.
- Share of all PV articles: weighted to PV's output 7 Sep – 5 Nov 2026 (never-shown print ≈ 10,970 stories at 77%, shown print ≈ 19,200 at 76%,
-
Traffic: no clear traffic gap on PV (median 2 organic sessions in both groups); DH "body not found" group median 54 vs 99 sessions (confounded by syndicated content).
-
Files:
checks/body_index_check/(body_index_check_results.csv,site_retest_results.csv,site_retest_raw.json,ga4_gsc_evidence.csv,impressions_traffic_by_group.txt),evidence/google_search_test_*(batch 2 notes:docs/google_search_test_batch2.md). -
16 Nov 2026: crawled HTML of 10 never-shown PV print copies (zero impressions, all indexed): body text missing in 9 (
checks/traffic_funnel/crawled_html_check_pv.md; see #31). Print copies (79% of PV output) are likely the largest group indexed without body. -
16 Nov 2026, DH: 8 of 8 never-shown DH editorial stories (and 2 of 2 controls) found by exact body phrase with
site:deccanherald.com(checks/traffic_funnel/crawled_html_check_dh.md): consistent with DH ≈ 0–7% indexed without body.
Fix: #1. After release: Request indexing for priority URLs (top traffic + affected list); recheck with checks/body_index_check/run.py and site_retest.py.
Rendering budget as the likely mechanism (16 Nov 2026). Every sampled PV story is crawled and indexed (URL Inspection: 599 of 600), but the text only exists after Google renders the page (#1), which is a separate, costlier step that can be delayed or skipped. Evidence that PV gets less rendering: DH has the same code, ~7× more Googlebot requests per new URL (#3) and ~0–7% of stories without text; PV indexed far more text in Jan 2026 (41% missing, ~400 stories/day) than after Varta Setu (automatic print-to-web) (71–80% missing, ~650/day; #2). Crawl stats "By file type" (exports 16 Nov 2026, 18 Aug – 14 Nov, crawl_stats/2026-11-16/) cannot measure it: JavaScript and JSON fetches are ~1% of HTML fetches on both sites (PV JS/HTML 0.007–0.009, JSON/HTML 0.009–0.011; DH 0.006–0.007 and 0.010–0.012) because Google caches resources across renders. Not proven. Fix stays #1 (text in server HTML removes the dependence on rendering); freeing crawl (#26 AMP, #21, #25, #32) only improves the odds. Clean test: ~50 stories from early Mar 2026 vs ~50 from mid-Apr 2026 (same age, before/after print-to-web), ~100 Serper credits.