P3 · AMP
Sites: Prajavani (PV, www.prajavani.net) and Deccan Herald (DH, www.deccanherald.com) · Priority: 3 of 6 · Prepared: 18 Nov 2026 by the Product, SEO and Data team · Internal refs: #26, #14, #40, #34 (AMP live-blog point only), #3
Summary
Every article on both sites has a second, AMP copy. Googlebot fetches those copies about as often as the real pages: AMP URLs are about 45–50% of Googlebot's successful HTML fetches on both sites. Our AMP health check on 17 Nov 2026 found no defect that stops Google serving AMP, and Google already serves AMP for almost every mobile article click. So the question is not "is AMP broken?" but "is it worth keeping?". That decision is made only after package P1 (server-rendered article body) ships and has been measured for about four weeks. Two things do not wait for the decision: premium stories are readable in full, free, on AMP (a narrow fix that can ship sooner), and a short list of AMP defects on DH.
Impact
Crawl cost (Search Console Crawl stats, exports of 12 and 16 Nov 2026, sample-based).
- PV: 453 of 993 sampled "OK (200)" URLs (46%) are
/amp/story/…(12 Nov 2026). PV "By file type → HTML": 441 of 991 (45%) (16 Nov 2026 export). - DH: 491 of 986 sampled HTML fetches (50%) are
/amp/story/…(16 Nov 2026 export). - Page indexing report (18 Nov 2026): "Alternate page with proper canonical tag" is the largest not-indexed reason on both sites and is almost all AMP: PV 1,328,617 pages (95% of a 1,000-URL sample are AMP), DH 1,255,065 (99%). PV's count grew by 114,586 in 85 days, about one AMP copy per new story. DH's is flat.
- Context (Crawl stats, 25 Jul – 21 Oct 2026): PV gets about 360 Googlebot requests per new URL, DH about 2,700. Crawl spent on AMP copies competes with main pages and new articles. Whether it directly slows discovery of new articles is not measured.
AMP traffic (Search Console search appearance and GA4, 16 Aug – 13 Nov 2026). Clicks Google served as AMP:
| PV | DH | |
|---|---|---|
| Web search | 946,321 of 12,904,219 (7.3%) | 1,310,275 of 4,495,898 (29%) |
| Discover | 242,252 of 28,536,333 (0.85%) | 649,435 of 24,550,565 (2.6%) |
| Google News | none reported | none reported |
| All three | about 1.19M of 41.5M (2.9%) | about 1.97M of 30.3M (6.5%) |
- September 2026 (1–30 Sep): AMP share of Google Search clicks PV 7.9%, DH 30.6%. This matches the 16 Aug – 13 Nov figures.
Current AMP traffic (last 28 days, 20 Oct – 16 Nov 2026, before P1 reached Google). This is the baseline to compare against at decision time (same 28-day length):
| PV | DH | |
|---|---|---|
| Web search | 240,559 of 3,359,126 (7.2%) | 429,995 of 1,295,268 (33.2%) |
| Discover | 17,150 of 7,954,006 (0.2%) | 139,534 of 7,485,188 (1.9%) |
| Google News | none reported | none reported |
| All three | 257,709 of 11.32M (2.3%) | 569,529 of 9.21M (6.2%) |
| GA4 sessions starting on an AMP page | 162,182 of 10.70M (1.5%) | 521,258 of 12.42M (4.2%) |
- Web share is stable since September; Google still serves AMP for nearly every mobile article click on both sites.
- Discover AMP share is falling on both sites (September to now: PV 1.8% to 0.2%, DH 5.0% to 1.9%). The cause is not known; it lowers the traffic at stake.
- PV's window includes the Deepavali traffic dip from 6 Nov 2026; shares are less affected than totals.
- GA4 sessions starting on an AMP page, 16 Aug – 13 Nov 2026: PV 752,063 (1.9% of 39.96M), DH 1,784,868 (4.5% of 39.38M).
- Correction: an earlier page-level count (AMP URLs about 2% of PV web clicks) under-stated AMP, because Search Console mostly credits AMP clicks to the canonical URL. Use the figures above.
- DH's higher share is traffic mix, not a technical difference. Only articles have AMP. Over 16 Aug – 13 Nov 2026, 10% of PV web clicks land on articles (the rest go to home, e-paper, cartoons and listings) against 53% on DH.
- Among mobile article clicks, AMP-served clicks are about 87% on PV and about 97% on DH (approximate; the two counts are measured slightly differently). Where an AMP copy exists, Google serves it for nearly every mobile article click.
Premium stories on AMP (stories published 7 Sep – 5 Nov 2026; Search Console 7 Sep – 13 Nov 2026). DH: 958 premium stories (7.8% of 12,262), with 67,780 web and 533,350 Discover clicks. PV: 658 (1.7% of 38,193), with 28,076 web and 1,442,901 Discover clicks.
Not proven:
- Retiring AMP does not delete these clicks (Google has given AMP no ranking or Top Stories advantage since 2021), but we have not measured what happens to rankings, speed or Discover after a switch-over on these sites.
- We do not know how many readers actually read premium stories on AMP.
- The effect of AMP crawl on discovery speed of new articles is unmeasured.
- Why DH's HTML response time to Googlebot rose on 11 Nov 2026 (see item 4) is not proven.
Decision: keep or retire AMP
When: not before P1 has been live about four weeks on that site. DH: P1 went live 19 Nov 2026, so the DH decision can be made from about 17 Dec 2026, staged by section. PV: about four weeks after PV's P1 release. P1 changes what Google sees on the main page. Today the AMP copy is the only complete-text copy on PV (one article checked live on 12 Nov 2026: main page 3 paragraphs / 723 characters in server HTML, AMP copy 6 paragraphs / 958 characters; one article, an illustration, not a measurement). Do not change AMP in the same window as P1, so effects can be attributed.
Data needed before deciding:
- P1 live about four weeks; body indexing re-checked on a sample of new articles.
- Mobile speed of main article pages (Core Web Vitals; see P6). Readers coming from Google will land on them instead of AMP.
- Search Console AMP search-appearance clicks, re-run for the latest 28 days, per site (web, Discover, News), compared with the 20 Oct – 16 Nov 2026 baseline above.
- Crawl stats before and after, so the effect of the P1 change is known. The low crawl per new PV article (about 360 Googlebot requests per new URL against about 2,700 on DH) is re-analysed at the same point: once Google has the body text without rendering, low crawl matters much less, and what remains is decided here.
Options
| Option | What it means | Criteria for choosing it | Main risk |
|---|---|---|---|
| A. Retire AMP | Remove StoryAmpTags from generators on each site; 301 every /amp/story/… URL (encoded and plain forms) to its article, permanently. |
Main pages are complete in server HTML (P1) and fast on mobile; body indexing improved; no AMP-specific traffic we cannot afford to move. | About 87% (PV) and 97% (DH) of mobile article clicks are AMP-served today, so nearly all mobile article search traffic moves to the main pages at once. |
| B. Middle ground (PV first) | Set ignoreAmpHtmlStoryTemplates: ["visual-story", "syndicated-print"] so print copies (79% of PV output, little traffic) lose their AMP copy; 301 their AMP URLs. |
Want to cut most PV AMP crawl with one config line and low traffic risk. | Smaller crawl relief on DH. Not tested on DH. |
| C. Keep AMP | Leave it as is; fix the defects below and add a paywall to AMP (item 1). | AMP traffic is judged too valuable, or main pages are not ready. | Keeps about half of Googlebot's HTML fetches on a second copy of every article, and keeps the code to maintain. |
Staging. PV: only after P1 is live on PV (option B is the first step then, before any full retirement). DH: 29% of web clicks are AMP-served, so retire in stages (one section first) with monitoring. If retiring: remove the amphtml link, keep the 301s permanently, and watch Discover, News and Crawl stats for four weeks.
Items
1. Premium stories are fully readable and free on AMP — Medium · PV, DH · Fix proposed
Issue: Subscription stories (access: subscription) are paywalled on the main page, but the AMP copy shows the whole story to anyone, with no paywall. Mobile readers from Google Search can read premium stories in full, free.
Evidence:
- Live check as Googlebot, 17 Nov 2026, all 8 premium stories in the sample. Words inside
<p>in server HTML, main page vs AMP: DH dh-exclusive 0 and 0 vs 823 and 765; DH interview 6–13 vs 1,109–1,178; DH syndicated-print 7 vs 534; PV pv-exclusive 15 and 15 vs 412 and 432. - No AMP page loads
amp-accessoramp-subscriptions. - AMP JSON-LD says
NewsArticlewithisAccessibleForFree: false, buthasPartis deleted by the hard-gate for premium stories (PVapp/server/seo-hard-gate.js:270-274, :291@70d7bebf; DHapp/server/seo-hard-gate.js:170-174, :194@22f4c0f0). - Cause:
@quintype/amp(PV 2.22.8, DH 2.17.13) renders its paywall only when the AMP route options includefeatureConfig.subscriptions. Neither site passes it:ampRoutesoptions at PVapp/server/app.js:364@70d7bebfand DHapp/server/app.js:485@22f4c0f0. (Library read from minifieddist/index.jsby grep.) - PV writes
"@productID"instead ofproductIDinisPartOf(PVseo-hard-gate.js:278), which Google ignores. - Exposure: see Impact. Premium is set by
access, not by template.
Fix:
- If AMP stays: configure
featureConfig.subscriptionsfor AMP (AccessType authorisation and pingback URLs, per Quintype'samp-subscriptionsdocs; ask Quintype for the supported config for each library version). AddhasPart: {"@type": "WebPageElement", "isAccessibleForFree": false, "cssSelector": "<paywalled container>"}on both page types. - Quicker: no AMP for premium stories. Omit
amphtmlwhenaccess === "subscription"(a custom check aroundStoryAmpTags), or set the story attributedisable-amp-for-single-story, which makes the AMP route 301 to the main page.ignoreAmpHtmlStoryTemplatesalone will not work, because premium is not a template. - If AMP is retired, this goes away.
- PV: change
"@productID"toproductID.
Impact: Stops free access to premium text on mobile. Makes paywall signals consistent for Google.
Verify: AMP copy of a premium story shows only the preview when not logged in (also on the AMP cache URL). Rich Results Test shows isAccessibleForFree: false with a hasPart selector that matches an element on the page. Re-fetch premium AMP pages and compare <p> word counts with the table above.
2. PV amphtml link uses %2F, creating two AMP URLs per article — Low · PV · On hold
Issue: PV's <link rel="amphtml"> points to an encoded path, e.g. /amp/story/news%2Findia-news%2F…-4387728. Both the encoded and the plain-slash AMP URL return 200 (both canonical to the article), so there are two AMP URLs per article. PV has no redirect, so there is no crawl waste like DH had.
Evidence:
- Live check on 3 PV articles, 11 Nov 2026: encoded form still emitted, returns 200 directly.
- DH fixed the same thing on 11 Nov 2026 (commit
3b847c28,decodeAmpUrl: trueingenerateSeo,app/server/app.js). - On the PV training site, 20 of 20 sampled stories already had no
%2Finamphtml(6 Nov 2026). Production was not re-checked after that. - PV's link is a relative path; Google resolves it (
StoryAmpTags, PVapp/server/app.js:592@70d7bebf).
Fix: Output the amphtml href with real slashes, as DH now does (decodeAmpUrl: true); optionally 301 the %2F form to it. On hold: changing it saves nothing on PV and becomes moot if AMP is retired.
Impact: Small. Removes a duplicate AMP URL per article.
Verify: amphtml in article HTML has no %2F; the AMP URL returns 200 with no redirect.
Re-checked 19 Nov 2026: still not live on PV. Fresh server HTML (Cloudflare MISS) links /amp/story/<section>%2F<slug> and that URL returns 200 (4 of 4 articles). The one-line fix (decodeAmpUrl: true in generateSeo, as on DH) is on branch amp-issue1 (5610ce4f, 26 Oct 2026) and not merged into master.
3. DH AMP request handling: redirect then next(); 5xx outage cause unconfirmed — High · DH · Investigate
Issue: DH had an outage on AMP story URLs, 24 Aug – 1 Sep 2026: about 40,000 5xx a day (peak 63,815 on 29 Aug), 980 of 984 sampled 5xx URLs were /amp/story/…. Search Console Host status showed about 25% of requests failing in the same days. It cut DH's Googlebot crawl, which has not recovered: non-error requests fell from 532k to 266k a day (Crawl stats, 14 Aug – 9 Nov 2026). 5xx have been about 0 since 4 Sep. Cause not confirmed.
Evidence:
- A DH AMP middleware in
tpml-dhmain,app/server/app.js(about lines 396–425, earlier read) looked up every AMP request's story through the API and compared the raw path to/amp/story/${storySlug}. The raw path contained%2F, so it never matched, giving a 301 on every AMP hit. The same block callednext()afterres.redirect(...), so the AMP handler kept running on an already-sent response. A possible source of the 5xx, unverified. The 301 side is fixed (below). - 301s: 173,864 a day (14–24 Aug 2026) fell to 87,023 a day (23 Oct – 9 Nov), still 949 of 995 sampled as AMP. The fix went live about 11 Nov 2026, after that export. "Other file type" requests fell from 161,526 (10 Nov) to 18,693 (11 Nov), consistent with the 301s stopping.
- Page indexing report (18 Nov 2026): 617 DH "Server error (5xx)" pages left over from the outage (95% AMP, falling); "Page with redirect" 1,424,424 (66% of sample AMP), falling since 24 Oct.
Fix: Confirm with Quintype and the DH tech team what failed on AMP 24 Aug – 1 Sep. Check the redirect-then-next() block and add return before the redirect, with no next() after it, if it is still there. Keep decodeAmpUrl: true.
Impact: Avoids a repeat outage on AMP. Not shown to have affected DH web search clicks yet (weekly web clicks stayed in the 300–450k band while crawl halved, 17 Aug – 8 Nov 2026); effect on discovery of new articles unmeasured.
Verify: No 5xx on /amp/story/… in Crawl stats; 301s fall to near zero; "Page with redirect" falls.
4. DH HTML response time to Googlebot rose on 11 Nov 2026 — Medium · DH · Investigating
Issue: DH HTML response time to Googlebot rose from about 140–155 ms (1–10 Nov 2026) to 245–271 ms (11–14 Nov) and had not come back at the last export (Crawl stats "By file type").
Evidence:
- 10 Nov: 577,786 requests, 164.7 KB per request, 155 ms. 11 Nov: 337,578 requests, 187.5 KB, 263 ms.
- Only code change that day: commit
3b847c28(decodeAmpUrl: true). Later deploys47a7e85e(13 Nov) and22f4c0f0(16 Nov) came after the jump. - Live timing as Googlebot, 16 Nov 2026, 6 stories: first request median 666 ms (main) and 752 ms (AMP); cached repeat about 310–330 ms; forced origin about 1.0 s for both. AMP is not slower to build.
- AMP responses carry
cache-control: public, max-age=60(nos-maxage); Cloudflare holds them for hours through an edge rule. Main pages uses-maxage=900. - Most likely cause: Googlebot now fetches full AMP pages instead of cheap redirects, so fewer cache hits and more bytes per fetch. Not proven.
Fix: Ask Quintype for Cloudflare cache-hit ratio and origin response time for Googlebot before and after 11 Nov (asked by email on 16 Nov 2026; no reply recorded). Consider s-maxage on AMP responses. Re-check with the next Crawl stats export.
Impact: Unknown until measured.
Verify: DH HTML response time back near 150 ms, or a cache explanation confirmed.
5. DH live blogs have no LiveBlogPosting markup on AMP — Medium · DH · Fix proposed
Issue: The AMP copies of DH live blogs have only NewsArticle and BreadcrumbList JSON-LD, so no LiveBlogPosting. Without it Google cannot show the "LIVE" label in Top Stories. (The non-AMP page has the same gap; that part belongs to package P5.)
Evidence:
- Live, 17 Nov 2026: 4 of 4 sampled DH live blogs, AMP JSON-LD is NewsArticle + BreadcrumbList only.
- Cause:
enableLiveBlog: false(tpml-dhapp/server/app.js:683-687@22f4c0f0). The custom builder inapp/server/amp/schema-type.jsis used only by the non-AMPrender-layout.js:93. - PV's AMP live blogs do have it: PV builds it in
seo-hard-gate.js:25-133for both page types.
Fix: Enable live-blog schema (enableLiveBlog: true, or fix schema-type.js) and check it on the AMP copy as well. The custom builder has invalid fields if switched on (eventStatus: "Live", relative card image URLs, a plain-string author, no publisher); fix these first. Skip if AMP is retired.
Impact: Eligibility for the "LIVE" label on DH live blogs (not measured).
Verify: Rich Results Test on the AMP copy of a live blog shows LiveBlogPosting.
6. DH story with a different canonical on its AMP copy — Low · DH · Investigate
Issue: For DH story 4208038, the regular article and its AMP copy had different canonicals.
Evidence: Training-site check, 6 Nov 2026: article canonical https://www.deccanherald.com/world/for-israelis-us-midterms-show-how-far-their-star-has-fallen-in-america-4208038; AMP canonical https://www.nytimes.com/2026/11/04/world/middleeast/israel-trump-midterm-elections.html. Seen in 1 of 20 DH stories; other sampled AMP canonicals matched. The 17 Nov 2026 production check (68 PV, 62 DH stories) found canonical = the article on all sampled AMP pages. Whether this story was fixed or only missed is unknown.
Fix: Confirm the intended canonical for syndicated stories (the AMP and web canonical should agree) and re-check this story.
Impact: Small. One story in the sample.
Verify: AMP and main page canonicals match for story 4208038.
7. Smaller AMP findings — Low · PV, DH · Note
- AMP pages carry no
max-image-previewmeta (the library adds it to story pages only,@quintype/seoimage-tags.js:133); the main page hasmax-image-preview:large. - URL Inspection warnings only (not errors): "Image size smaller than recommended" on visual stories (both sites), PV
cartoon(3 of 4),nammakshetra-namma-ooru(4 of 4), and a few PVtext/syndicatedstories. - DH: 3 of 62 sampled stories had no AMP pass: 2
/dhbrandpr/stories blocked by robots.txt (a separate issue, #30) and 1 with no AMP result yet. - AMP page weight, live 16 Nov 2026: PV about 127 KB, DH about 234 KB per AMP article.
- AMP health, 17 Nov 2026: every template except
visual-storyhasamphtml; all AMP URLs return 200 with no redirect, canonical = the article, nonoindex; the official AMP validator found 0 errors on all 122 AMP pages; URL Inspection AMP result PASS on PV 68 of 68, DH 59 of 62. Nothing here needs fixing for AMP to work.
Release checklist
- [ ] Item 1: decide paywall-on-AMP or no-AMP-for-premium; ship; verify with a logged-out check, including the AMP cache URL.
- [ ] Item 1: PV
@productIDchanged toproductID. - [ ] Item 2: PV
amphtmlplain slashes (only if AMP stays; otherwise skip). - [ ] Item 3: AMP middleware reviewed; cause of 24 Aug – 1 Sep outage documented.
- [ ] Item 4: Quintype cache and origin data received and reviewed.
- [ ] Item 5: DH AMP live-blog markup (if AMP stays).
- [ ] Item 6: DH story 4208038 canonical reviewed.
- [ ] Keep-or-retire decision recorded, with the four-week post-P1 data attached.
- [ ] If retiring: remove
StoryAmpTags; 301s from every/amp/story/…form to the article, kept permanently; monitor Discover, News and Crawl stats for four weeks; stage DH by section.
Dependencies and related packages
- P1 Article server HTML: gates the keep-or-retire decision. Do not change AMP in the same four-week window.
- P2 One URL per story: the PV
%2Fduplicate AMP URL (item 2) and any AMP-to-article redirect rules overlap with it. - P4 Listing pages and internal links: not directly linked. AMP pages link to normal article URLs, not to other AMP pages, and daily sitemaps list no AMP URLs (PV 0 of 590, DH 0 of 129 on 10 Nov 2026). So AMP is not a crawl trap.
- P5 Structured data and head tags: carries the non-AMP side of DH live-blog markup and the rest of the DH structured-data and social-tag findings. PV's publisher
@idwith no Organization node is also there. - P6 Page speed, ads and analytics: main pages must be fast on mobile before AMP is retired. Item 4 (DH response time) overlaps.