P6 · Page speed, ads and analytics
Sites: Prajavani (PV, www.prajavani.net), Deccan Herald (DH, www.deccanherald.com) · Priority: 6 of 6 · Prepared: 18 Nov 2026 by the Product, SEO and Data team · Internal refs: #39, #7, #5, #6, #8, #10, #15, #16, #29
Summary
This package covers front-end weight, layout shift, ad code and analytics timing. Most items are PV only. Start with two cheap, low-risk changes: port DH's footer loader to PV, and stop sending the raw CMS menu to the browser on both sites. Together they remove roughly 670 KB of embedded data from every PV article page (about 300 KB from the footer, 371 KB of raw menu; uncompressed). DH saves 93 KB from the menu. The Taboola space fixes (items 4 and 5) are small and can ship early. The remaining items are a hero and header layout flip, ad-slot clean-ups, a broken font build, and a GA4 page_view that fires too late. One item (field CLS on some pages) still needs investigation.
Core Web Vitals are only a light ranking signal. The benefit for readers (less jumping, faster first paint) is better established than any ranking gain. Do not expect a ranking change from this package alone.
Impact
What the measurements show:
- PV article HTML is 1,864 KB uncompressed (344 KB gzip); DH is 1,289 KB. Of PV's total, 954 KB is embedded page data. (Live measurement, one recent article per site, Googlebot user agent, 16 Nov 2026.)
- PV real-user (CrUX p75) for one article URL: LCP 3.6 s, CLS 0.33 (poor), INP 306 ms. DH lab on a comparable test: LCP 0.73 s, CLS 0.01. (Chrome DevTools traces and CrUX, 16 Nov 2026; mobile 412×915, Fast 4G, 4× CPU, cold cache. Different articles, so treat the comparison as indicative.)
- PV lab LCP is 17.0 s, and the LCP element is a Google ad image, not the article image (16 Nov 2026 trace, one article).
- Search Console Core Web Vitals report: 67.7K PV mobile URLs fail on "CLS issue: more than 0.1"; the main group has 67,258 URLs and group CLS 0.19 (Search Console, as recorded in Nov 2026).
- GA4 landing page "(not set)", 14 Oct to 10 Nov 2026 (GA4 Data API): 828K PV sessions (7.3%) and 1.07M DH sessions (9.1%). The web part, not counting app sessions, is PV 391K and DH 866K. This is about 3% of each site's organic sessions.
What is NOT proven:
- No ranking gain from fixing CLS or LCP has been shown. Google uses page experience as a light tie-breaker. Top Stories has not required good Core Web Vitals since 2021, and Google does not list them as a Discover requirement.
- The lab and field numbers come from one or a few articles, not a site-wide sample.
- The cause of high field CLS on some pages (item 8) is unknown.
- The cause of the GA4 gap is inferred from code, timing and event mix. It was not reproduced directly.
Items
1. PV: load the footer collection like DH does — Medium · PV · Fix proposed (code already exists in DH)
Issue: PV embeds a footer data collection of about 311 KB in every article page, but phone users cannot see most of it.
Evidence:
qt.data.footerCollectionis 311 to 328 KB per PV page. It holds 4 sub-collections of 40 stories each, 117 distinct stories (DevTools check, article…-4376516, 16 Nov 2026).- The rendered PV footer has 17 story links, 16 of them from this collection. On mobile (375×667 emulation) 1 of 17 links is visible, and none of the 16 collection links is visible. Desktop shows all 17 (16 Nov 2026).
- DH fixed this on 13 Nov 2026 (
tpml-dhcommit47a7e85e, "Cwv issue (#2094)",app/server/data-loaders/menu-data.js:5-37). It requests"story-fields": "headline,slug,url"and only 4 stories per sub-collection. DH's footer collection is 5 KB and all 12 footer links are visible on mobile (412×915, 16 Nov 2026). - PV's
loadFooterCollection(tpmlapp/server/data-loaders/menu-data.js:5-7@70d7bebf) still requests full stories with only{ limit: 4 }.
Fix:
- PV: port DH's
menu-data.jsfooter loader totpmlunchanged. PV's footer shows 4 groups × 4 links, so the same limits fit. - Keep the footer links in the HTML. Google can still follow them. Do not remove the footer.
Impact: about 300 KB less embedded data per PV article page. Smaller HTML helps every Googlebot fetch and mobile first render. Not a proven ranking factor.
Verify: view-source on a PV article: footerCollection is far smaller (DH's is 5 KB). Mobile footer still shows its links on staging. Compare HTML size and Search Console Crawl stats KB/request before and after.
2. Both sites: stop sending the raw CMS menu to the browser — Medium · PV, DH · Fix proposed
Issue: qt.config.layout.menu is the raw CMS menu export. It is on every page but the browser never reads it.
Evidence:
- Size: PV 371 KB, DH 93 KB (live, 16 Nov 2026).
- In both repos it is read only on the server, to build
navigationMenu(app/server/load-data.js:56, :134intpml@70d7bebf;:56, :145intpml-dh@47a7e85e). - Every header and footer component reads
qt.data.navigationMenu(PV 133 KB, DH 34 KB) instead. The only browser-side reader in the installed libraries is Quintype'sMenucomponent (@quintype/components/dist/es/components/menu.js:26), which neither site imports. - Other repeated data: PV
qt.config.sections83 KB; DHconfig.sections50 KB.
Fix:
- Leave
layout.menuout of the client page data on both sites. The server still fetches/api/v1/config; that API is untouched. - Then review
qt.config.sectionsand trimnavigationMenuto what the page uses. The frameworkloadDataandapp/server/data-loaders/are the places to look.
Impact: up to 371 KB less per PV page and 93 KB less per DH page, with no visible change expected.
Verify: on staging, click through the header, hamburger, footer and a client-side page change. View-source shows layout.menu is gone. PV article HTML should end well below 1 MB (embedded data close to story size plus a small config).
3. Both sites: move inline CSS and repeated SVG out of the HTML — Medium · PV, DH · Fix proposed
Issue: Article HTML carries large inline CSS and inline SVG icons.
Evidence (16 Nov 2026, same live measurement as above):
- Inline CSS: PV 461 KB, DH 637 KB.
- Inline SVG icons: PV 373 KB, DH 367 KB.
- After items 1 and 2, this is DH's main remaining weight.
Fix: keep only critical CSS inline and serve the rest as cached external stylesheets. Replace repeated inline SVG with a sprite or external file.
Impact: smaller, more cacheable HTML. Magnitude not measured.
Verify: record HTML size, origin TTFB and Crawl stats KB/request before and after. Re-run Lighthouse (mobile) on one article per site.
4. PV: reserve space for the Taboola feed below the article — Medium (lowered from High on 12 Nov 2026) · PV · Fix proposed. Can ship early.
Issue: The Taboola feed below the article has no reserved height. The page jumps as the reader scrolls, and the text moves by about 240 px.
Evidence:
- Search Console: 67.7K mobile URLs fail on CLS above 0.1; main group 67,258 URLs, group CLS 0.19.
- Lab test, 64 URLs (Oct 2026): on 59 of them CLS was above 0.1 only after scrolling; load-only, 1 of 59. The largest shift on 55 of 59 URLs was
#taboola-below-article-…(tbl-feed-container). - With Taboola blocked, CLS after scrolling fell from 0.28–0.67 to 0.01–0.11 (6 URLs).
- Cause:
ads/taboola-ad/index.jsandtaboola.m.css. The container has no reserved height and the feed mounts immediately (rendered fromstory-templates/story-elements/index.js).
Fix:
- Reserve a
min-heightfor the feed container, matching the typical rendered mobile feed height. - Mount it lazily near the viewport (
WithLazywith a margin). - Confirm with the Taboola account owner that blank reserved space is acceptable when Taboola does not fill.
Impact: removes the biggest measured source of CLS. Ranking effect is expected to be small (see Impact section).
Verify: Chrome DevTools, Performance panel, Live metrics. Mobile emulation, 4× CPU, Fast 4G. Load an article, wait 5 s, scroll slowly to the bottom. Pass: CLS below 0.1, and taboola-below-article / tbl-feed-container or footer is no longer the main shifting element. Test at least two text articles and two cartoon pages (cartoons were about 0.67 after scrolling before the fix). Field CLS in Search Console updates on a rolling 28-day window, so expect results about a month after release.
5. PV: reserve space for the Taboola mid-article block — Low (lowered from Medium on 12 Nov 2026) · PV · Fix proposed. Ship with item 4.
Issue: #taboola-mid-article-personalisation (.taboola-4u in story-elements.m.css) is inserted after the 4th paragraph with no height.
Evidence: measured shift 0 to 242 px (lab, Oct 2026).
Fix: give .taboola-4u a reserved height. Same blank-space check with the Taboola owner as item 4.
Impact: reading comfort. Small effect on rankings.
Verify: same test as item 4; no jump at the 4th paragraph.
6. PV: header and hero layout flip on phones — Medium · PV · Fix proposed
Issue: helper/custom-hook/useDevice.js returns null until after mount, so every phone first renders the desktop layout, then switches.
Evidence:
story-templates/story-header/index.js:31:heroImageMinHeight = isMobile ? "184px" : "456px", withtransition: min-height 0.3s. The hero area goes 456, then 184, then about 271 px on mobile. The server HTML containsmin-height:456px.- Measured load-time shift 0.03 to 0.11 (lab). A 16 Nov 2026 trace shows CLS 0.03 on one PV article.
- Other
!isMobile/isMobile ?branches that also render desktop first on phones: sidebar (text/index.js:46),cartoon/index.js:54, andOutOfPageAdinads/index.js.
Fix:
- Size the hero with CSS
aspect-ratioand media queries (for example 16/9 on the wrapper), not JavaScript. - Remove the
min-heighttransition. - Review the other
useDevicebranches listed above. - Server-rendering the hero image itself, with
fetchpriority="high", is covered in package P1 (Article server HTML). It is the main LCP fix, because the PV lab LCP element is currently a Google ad image (17.0 s).
Impact: removes a 100 to 270 px jump on load. Load CLS is already small, so the gain is mainly smoothness and LCP (with P1).
Verify: mobile emulation, 4× CPU, reload an article with a hero image. Headline, byline and hero should not move in the first 3 s. Load-only CLS below 0.05.
7. PV: ad slot and ad container clean-ups — Medium (double requests), Low (hydration) · PV · Fix proposed
Issue: Ad slots are defined twice, and ad containers differ between server and client.
Evidence:
- Double requests: Google's renderer console shows
[GPT] Error in googletag.defineSlot: Cannot create slot 1013527/PV_Mweb_AP_Sticky. Div element "div-gpt-ad-1675837571080-0" is already associated with another slot. Every slot is requested twice, about 2 s and 22 s after load. This may cause late layout shifts. - Hydration:
AdContainerinads/index.jspicks the mobile or desktop slot fromglobal.width, which is set only in the browser (app/client/app.js:29). The server renders an empty.ad-background; React inserts the slot during hydration (Expected server HTML to contain a matching <div>on desktop). Ads still load correctly. - Ad slots for 336×280 and 300×250 already reserve height in
ads/ads.m.css(min-height50/90/250/600 px). The measured CLS did not come from them.
Fix:
- Guard
defineSlotso it runs once per div id (ads/dfp-component/AdSlotinads/index.js). Confirm whether the 22 s request is an intentional refresh. - Render both slot wrappers and hide one with CSS (
hide-mobile/hide-desktop), or choose the slot after mount without changing markup. - Reserve fixed space for above-the-fold ad slots.
Impact: fewer duplicate ad requests, no hydration warnings, possibly less late CLS. Ad ops should check fill, revenue and viewability around the change.
Verify: console shows no defineSlot error. Network panel shows one request per slot until a refresh is expected. No hydration warning on desktop. Ad ops confirm slots still fill.
8. PV: unexplained high field CLS on some pages — Medium · PV · Needs investigation
Update 19 Nov 2026: folded into item 4. The lab figures below were measured on load only. The item 4 lab test shows PV CLS above 0.1 only after scrolling, mostly from the Taboola feed, and both field values fall in that after-scroll range. No separate fix is needed; re-check these two URLs after item 4 ships, and add web-vitals attribution logging only if field CLS stays above 0.1.
Issue (original): Field CLS is far above lab CLS on some pages, and the cause is not known.
Evidence:
- Weather page
…-4323207: field CLS 0.54, lab 0.095 to 0.100, with or without Taboola. Candidates: ad fills, scroll behaviour, premium or login states. Not reproduced. - One article (
…-4376516, feature): CrUX p75 CLS 0.33 against lab CLS 0.03 (16 Nov 2026). Taboola or unsized slots are the likelier cause by code reading, but this is not measured.
Fix (no code fix yet): add web-vitals attribution logging on live pages so real shifting elements are captured, then act on the results.
Impact: unknown until measured.
Verify: after a week of logging, list the top shifting elements by page type. Re-check field CLS after items 4 to 7 ship.
9. PV: fix broken font files — Low (Googlebot view), but readers are affected · PV · Fix proposed
Issue: Some "font" files served by PV are not fonts. They fail for every reader, not only Googlebot.
Evidence:
- Search Console page resources:
PrajavaniTextRegular.woff2,PrajavaniTextSemibold.woff2,PvHeadlineSemibold3.woff2and Roboto Slab show "Other error" for Googlebot. Google often skips fonts, so no indexing impact is expected. - Chrome console on a PV article, 16 Nov 2026:
Failed to decode downloaded font: https://www.prajavani.net/tpml/assets/ebd8d869f9071312fbd0.woffand…/3f9cbec7131a3db6b491.ttf, each withOTS parsing error: invalid sfntVersion: 1702391919. That value is hex6578706F, the text "expo". The files start withexport …: a JavaScript module saved under a font name. - Likely cause:
webpack.config.js:27-31uses the oldfile-loader(name: "[name].[ext]"), while the webpack 5 build in@quintype/buildalso processes the same files as assets. Together they writefile-loader's JS output (export default __webpack_public_path__ + "…") into a hashed asset. The named.woff2entries (quintype-build.config.js:38-40, used bylayout.ejs:288-307) are separate. Which stylesheet references the hashed.woff/.ttfstill needs confirming. robots.txtDisallow: /*.woff$affects.woffonly, not.woff2.- Related lab finding:
font-displaycosts about 270 ms of FCP on PV and about 220 ms on DH (16 Nov 2026 traces).
Fix: in the font rule add type: "javascript/auto" (keeps file-loader working under webpack 5), or use type: "asset/resource" and drop file-loader.
Impact: fonts load for readers; fewer font-swap shifts are possible (not measured).
Verify: curl -s <font url> | head -c 4 returns wOFF or wOF2 (or blank bytes for TTF). The console shows no font decode errors.
10. PV: third-party scripts and GTM load — Medium · PV · Fix direction only
Issue: Third-party code dominates PV's main thread in the lab.
Evidence (one PV article, lab, 16 Nov 2026): Google Tag Manager 1.7 MB and 3.0 s main thread; Google/DoubleClick ads 1.2 MB and 2.7 s; Amazon ads 104 KB and 1.8 s; Clarity 28 KB and 1.4 s; Taboola 1.1 MB and 1.3 s; Facebook 271 KB and 0.8 s.
- GTM
GTM-MGBFQ9Rloads at page start (views/pages/partials/gtm.ejs, included atlayout.ejs:145). The public container (532 KB, fetched 16 Nov 2026) loads Clarity and the Facebook pixel, plus 5 custom HTML tags, Google Ads conversion tags and GA4. Amazon, vdo.ai, Taboola, iZooto and Criteo are not in the container. - Amazon ads and vdo.ai are not in site code (the Amazon loader is commented out at
ads/utils.js:157-159). They most likely come from Google Ad Manager line items or creatives. Only the Ad Manager setup can confirm that.
Fix: audit GTM tags and defer non-essential ones (Clarity, Facebook) until after load. Ad ops should check where Amazon and vdo.ai are injected.
Impact: unmeasured. INP of 306 ms (CrUX p75, 16 Nov 2026) is a candidate to improve.
Verify: repeat the DevTools trace and compare third-party main-thread time and LCP.
11. Both sites: send the GA4 page_view earlier — Medium · PV, DH · Fix proposed, not built. Needs GA4 report-owner input.
Issue: The page_view event is sent after other events, so some visits have no landing page in GA4 ("(not set)"). This is not a ranking issue. It makes GA4 page-level reports undercount real visits, including organic landings.
Evidence:
- Numbers: see the Impact section (GA4 Data API, 14 Oct to 10 Nov 2026). Share of web sessions affected: DH articles 8.9%, DH home 3.5%, DH tags 1.5%; PV articles 4.2%, PV home 3.6%, PV cartoon hubs 1.0 to 1.5%; AMP 0% on both. App sessions are part of the total and are expected (PV 439K, DH 214K).
- Affected sessions have
session_start,first_visit,user_engagementandad_impressionoruser_properties_update, but almost nopage_view(DH: 1,752 across about 528K sessions). DH peaked on 21 to 22 Oct 2026 on viral stories, mostly fromgooglequicksearchbox. - Headless Chrome runs on live articles, 15 Nov 2026. DH: the first GA4 hit was
ad_impression;page_viewfollowed 0.1 to 0.8 s later (about 3.3 s desktop, about 20.8 s throttled mobile). PV: first hituser_properties_updateat 2,025 ms;page_viewat 2,940 ms (desktop). - Cause: both GTM containers (
GTM-T46PRBZDH,GTM-MGBFQ9RPV) fire the Google tag withsend_page_view: false. GA4page_viewfires only on the site's owndataLayerpush. That push waits for the React app, then a headeruseEffectdelay, then 1,500 ms on story pages for the Coral comment count.- DH (
tpml-dh,feature/nav-v2-home-top@c459d2bd):app/isomorphic/components/layouts/header/index.js:746-757(100 ms);app/isomorphic/components/helper/datalayer-utils.js:348-358(1,500 ms). - PV (
tpml,fix/ssr-story-paragraphs@bcd5d909):app/isomorphic/components/layouts/header/index.js:169-183(150 ms);app/isomorphic/components/helper/datalayer-utils.js:705-710(1,500 ms);app/isomorphic/components/helper/GlobalSubscriptionSyncer.js:249pushesuser_properties_updatefor every anonymous visitor as soon as the app runs.
- DH (
- Not reproduced directly: no session without
page_viewwas captured in the lab. The cause is inferred from setup, code, timing and event mix. - Also found: DH's
views/pages/layout.ejsloadsgpt.js(line 88) andadsbygoogle.js(line 92) before the GTM partial (line 111). DH's Google tagG-8ZF3X8Y1X2differs from its web streamG-F45W975H2J(events are sent by override). vdo.ai configures two extra GA4 IDs and a dead Universal Analytics ID (UA-113932176-46) through the sites'dataLayer.
Fix:
- Render the
page_viewdataLayerpush on the server from the story metadata (app/server/handlers/render-layout.js:41already has the story), inviews/pages/layout.ejsjust before the GTM partial. Remove the app's ownpage_viewpush. comment_numbermoves to a later event or is dropped. Agree this with the GA4 report owners first.- Smaller first step: remove the 1,500 ms comment-count wait.
- Housekeeping: remove
UA-113932176-46.
Impact: more accurate GA4 landing-page reports. This changes event order and any report that uses comment_number, so tell report owners before release.
Verify: fresh browser profile: the first GA4 hit is page_view with _ss=1. After release, the web "(not set)" share on DH articles should fall from 8.9% toward 1% or less. Track daily for 2 to 4 weeks. App sessions, 30-minute session timeouts on open tabs and bots will still show "(not set)".
Release checklist
- Items 1 and 2 first (smaller HTML, low risk). Staging click-through: header, hamburger, footer, client-side navigation.
- Items 4 and 5 together (Taboola space). Confirm blank-space behaviour with the Taboola owner.
- Item 9 (font build). Check font URLs with
curl. - Items 6 and 7, ideally alongside P1's hero-image work, so LCP and layout change once.
- Item 11, with GA4 report owners informed before release.
- Start item 8 (web-vitals logging) at once; it needs time to collect data.
- After each release: Lighthouse mobile on one article per site, CrUX p75, Search Console Core Web Vitals report (allow about 28 days), GA4 "(not set)" share, and Crawl stats KB/request.
- Ad ops sign-off that ad fill and viewability did not drop.
Dependencies and related packages
- P1 Article server HTML: server-rendering the hero image with
fetchpriority="high"and explicit width and height is the main LCP fix. Item 6 here removes the layout flip around it. - P2 One URL per story, P3 AMP, P4 Listing pages and internal links: no direct dependency. Footer links (item 1) stay in the HTML, so internal linking is unchanged.
- P5 Structured data and head tags: none. JSON-LD is only 11 KB (PV) and 20 KB (DH) of article HTML.
- Item 11 depends on GA4 report owners. Items 4, 5 and 7 depend on the Taboola and ad ops owners.