Synthesis — bupa.com.hk

Where sources corroborate each other

Ahrefs + Brand Radar tell the same story from opposite directions. Ahrefs shows 73.1% of organic traffic concentrated in the top 10 pages, with the critical-illness page (1,939 traffic), pregnancy, baby insurance and VHIS pages all sitting on 0–1 referring domains while carrying heavy AI citation counts. Brand Radar independently confirms the dependency: 90.3% of Bupa's 12,420 monthly AI impressions come from Google AI Overviews + AI Mode, and the top 10 questions alone account for 7,100. Two sources, one conclusion: Bupa's visibility is borrowed from a single feature of a single platform.
Live checks + Brand Radar disprove the bot-access hypothesis. Every AI crawler tested (GPTBot, ClaudeBot, OAI-SearchBot, PerplexityBot, Bytespider) received identical treatment to the browser — 0% block rate, zero bot/browser differential. Yet Brand Radar shows Perplexity (130 impressions), Gemini (420) and Copilot (660) are negligible. The non-Google citation gap is therefore a content/citation problem, not an access problem. Any remediation aimed at "unblocking AI crawlers" would be wasted effort (with the caveat that single-IP probes can't speak for production crawler IPs).
Ahrefs internal-link data + live-check 404s point at the same architectural neglect. Ahrefs shows 752 dofollow internal links each to two PDFs and ~490 to legal pages, with an average of 155.9 internal links per page. Live checks show slow, application-layer 404s (412–448ms vs ~94ms for robots.txt), consistent with no page caching. Both are symptoms of a site nobody is technically maintaining.
Ahrefs broken backlinks + live-check 404s corroborate a URL-lifecycle problem. Real publications (am730 DR 75, howdengroup DR 78, lihkg DR 59) link to defunct pages like /bupa-dental/ and /transfer-care/ with no redirects in place. That independently confirms Bupa lets URLs die — which lends credibility to treating the probed /contact/ and /services/ 404s as a real pattern, not only a probing artefact.

Where sources contradict or complicate each other

The 404s are probably partly an artefact — but not wholly. Ahrefs shows /tc/ as the homepage variant with 612 internal links, meaning the site is locale-prefixed. The probed /contact/ and /services/ likely live under /tc/ or /en/ paths. So the 100% 404 rate may reflect wrong probe URLs rather than missing content. But the broken-backlink data proves the underlying disease (dead pages, no 301s) exists regardless. Do not conflate the two.
The traffic decline is demand-shaped, not rankings-shaped. Ahrefs: traffic -32% while tracked keyword volume fell 33.9% and positions held flat. The crawl cannot explain the trend (it's a 1-URL sample), and the trend itself doesn't require an onsite explanation. One complication worth flagging: Brand Radar shows heavy AI Overview presence on exactly Bupa's target queries, so AI answers absorbing clicks may contribute to traffic erosion beyond pure demand loss — but without GSC CTR data, that mechanism is unproven.
The crawl says nothing; don't let it into the narrative. 1 URL, 200 OK, depth 0. The 0% indexable figure is a sample artefact. Meanwhile every other source screams that architecture matters. The single biggest gap in this engagement is that the ~34K URL site has never been crawled.

The strongest cross-source narratives

1. Bupa's organic visibility is a single point of failure. AI citation is doing all the work; conventional authority is absent (0–1 refdomains on the pages carrying the load); the platform is concentrated (90.3% Google); the questions are concentrated (top 10 = >50% of volume); and demand is shrinking (-33.9%), which removes "wait for volume to recover" as an option. One change to Google's AI Overview sourcing and the pages driving 73% of organic traffic have no rankings cushion to land on.

2. The site's internal authority is pointed at the wrong assets. While VHIS comparison questions are exactly where AI demand sits (Brand Radar), Bupa's internal linking funnels 752 dofollow links to a VHIS glossary PDF rather than to answerable pages. The pages AI actually cites get neither external links nor concentrated internal equity. This is verifiable by the client's team in week 1 with any crawler.

3. The paid surge is masking organic erosion. +806% paid spend against -32% organic traffic is a commercial pivot papering over a structural visibility problem. In a shrinking-demand market, capture rate within AI answers is the only organic growth lever left — which is precisely where Bupa is concentrated and fragile.

Featured finding

Bupa's entire organic channel rests on a handful of AI-cited pages that have no authority backstop — and the site's own internal linking starves them further.

The mechanism: Ahrefs shows the critical-illness-insurance-vs-medical-insurance page drives 1,939 monthly visits as the AI Overview source for "危疾" (vol 1,800, position 1) and "危疾保" (vol 2,500) — on a single referring domain. The pregnancy page is the AI Overview source for "懷孕保險" on 0–1 refdomains. Brand Radar confirms 90.3% of AI visibility volume rides Google alone, with the top 10 questions carrying over half of it. Live checks prove the near-zero Perplexity/Gemini/Copilot presence is not a blocking issue — access is open; the content simply isn't being cited elsewhere. And Ahrefs internal-link data shows the equity that could harden these pages is instead flowing to PDFs (752 dofollow links) and legal footers.

Why this matters more than anything else in the data: it is the only finding three sources converge on, it exposes the majority of current organic value (73.1% of traffic in 10 pages) to a single-platform sourcing decision Bupa does not control, and the shrinking demand curve (-33.9% volume) means there is no alternative recovery path. Fixing it is concrete: harden and interlink the AI-cited pages, reclaim dead URLs with existing authority, and build citation presence on the platforms where access is already open. It is fast, verifiable, and the root cause is already identified.

Data gaps visible only in combination

The pages carrying 73% of traffic have never been crawl-verified. Ahrefs and Brand Radar tell us exactly which pages matter; the 1-URL crawl cannot confirm a single one is indexable, canonical-clean or template-sound. Full 34K crawl is the gateway to every other onsite claim.
No GSC data. Cannot separate demand decline from AI-Overview CTR compression, and the branded/generic split is broken in Ahrefs (0 branded terms is wrong). True generic visibility is weaker than it looks — by an unknown amount.
No server logs. Live checks are single-IP, UA-string only. Production Googlebot/AI-bot behaviour is unverified.
Locale architecture unknown. /tc/ confirmed via internal links; /en/ existence, hreflang, and whether /contact/ and /services/ exist under locale paths are all unverified until the crawl.

Ranked priorities

GEO: De-concentrate AI visibility — harden the cited pages, extend to platforms where access is proven open.
Onsite: Full 34K crawl (indexation, templates, depth), redirect the dead pages holding DR 59–78 links, fix internal-link equity flow away from PDFs/legal.
Risk cleanup (not acquisition): jobsdb = 53.6% of all live backlinks; average refdomain DR 11.5; spam anchors present. Frame as fragility, never as "build more links" — the data shows links are not the bottleneck.
Commercial context: paid +806% is masking erosion; name it.

Next step requiring signature: GSC access + server logs + sign-off on the full crawl. One week, and every open question above closes.