AAAC/organic investigation

Section 04 · Where the phone calls actually come from

Calls & conversion economics

The call data looked flat, then spiked, then collapsed — until it was split by engine and landing page. Underneath: a Yahoo-labeled paid artifact, a genuinely flat Google core, and a real, masked commercial decline.

§ 4.1 — The bubble bought no calls

~20× the sessions, the same calls

The PAA bubble produced ~zero incremental phone calls; losing it should cost ~zero calls.

Corroborated by conversion rates: PAA traffic converts at ~0.1% (Dallas at peak: 21k sessions/mo → 15–40 form submits), and by the page-class decomposition below.

§ 4.2 — The 2025 "spike" that wasn't

Summer call events: 44 → 123 → 35. Calling sessions: flat.

Windows: Jun 1–Aug 13 each year. Split by engine, the divergence disappears — 82 of 2025's 123 events came from a burst that wasn't Google, wasn't organic, and mostly wasn't even in-market. (Event counts also overstate callers: one 2024 session on /woodpeckers produced 12 events. Calling sessions are used throughout.)

R5 confirmed the owner's hypothesis: the burst was untagged Microsoft Ads (Yahoo syndicates Microsoft Ads). It landed on the homepage, converted at 14.9% (vs Google's 5.1% on the same page and window), was geographically out-of-market (Miami 82, Fort Myers 21, Orlando 8 sessions), with degraded engagement. Corrections that followed: TC Bing ads ran from ~May 2025, not "since Jan 2026"; paid masking of TC organic began May 2025; and the burst's 14.9% call rate matches the tagged bing/cpc signature of 10–14% — one consistent, still-unexplained high-rate paid pattern, not two anomalies.

§ 4.3 — Under the flat aggregate

A real commercial decline, confirmed against 2024

Google-organic only, by landing-page class. Calling sessions / sessions / conversion rate:

TC Google-organic call decomposition by landing-page class (Jun 1–Aug 13 windows)
Page classSummer 2024Summer 2025Summer 2026Change driver
Service-area20 / 187 / 10.7%15 / 176 / 8.5%12 / 108 / 11.1%100% traffic (rate stable)
Service/content4 / 185 / 2.2%4 / 191 / 2.1%2 / 78 / 2.6%Traffic
Homepage6 / 459 / 1.3%10 / 197 / 5.1%6 / 107 / 5.6%Traffic loss offset by a real rate step-up (dates to May 2025 — unexplained site/instrumentation change)
Blog0 / 370 / 5798 / 13,259 / 0.06%Bubble donates 8 calls
Commercial rollup31 / 841 / 3.7%29 / 572 / 5.1%20 / 295 / 6.8%Sessions −65% vs 2024

Conversion is not the problem anywhere. Every commercial class converts at or above its 2024 rate (service-area ~10–11% all three years). What's missing is commercial organic traffic: sessions −65% vs 2024, service-area calls 20→12 — purely traffic-driven. The aggregate only looks flat because the homepage rate step-up (+4.6 points) and 13k blog sessions (+8 calls) mask a −11 commercial call loss. GSC corroborates, S25→S26: commercial impressions 160k→34k (−79%), clicks 489→253 (−48%).

§ 4.4 — The masking layer

Why the dashboards said everything was fine

TC total call events, July, by year
17 → 51 → 111
2024 → 2025 → 2026. An all-time high — while the franchisee reported a bad two weeks. The high rests on paid events.
Bing CPC call events, monthly
7 → 88
~97% land on the homepage. Paid call events rose 15→89/mo Apr–Jul 2026 while organic fell — the channel-mix shift that produced the retracted "−80%".
Bing CPC call rate per session
10–14%
~40× the organic rate. Flagged in R1 as suspicious — audit for UET/GTM double-fire or accidental taps before trusting paid ROI.

The open contradiction. If TC total call events are at an all-time high, why did the franchisee report a bad two weeks YoY? Candidate reconciliations: (a) the paid call events are partly phantom (the 10–14% rate); (b) real calls come via Google Business Profile / local pack — invisible to both GA and GSC — and fell there; (c) paid clicks are lower-quality (wrong intent or area). Discriminator: real call logs + GBP insights + a Bing UET/GTM double-fire audit. This is now the top data gap.

§ 4.5 — Dallas: the instrument was broken

The tel-link hijack

A "Bing Ads Body" snippet (installed via the HFCM plugin in wp-admin — not in git) rewrites the phone link so Bing UET can fire before opening the dialer:

// Effect of the snippet, as observed in the live DOM:
document.querySelectorAll('a[href="tel:469-423-9299"]').forEach(a => {
  a.href = "#";                      // ← kills the GTM call_to_click trigger
  a.addEventListener("click", () => {
    uet_report_conversion();           // fire Bing conversion…
    window.open("tel:469-423-9299");   // …then open the dialer
  });
});
// Side effects: breaks long-press / copy / dialer in some browsers;
// "#" scrolls to top on slow JS. Fix: keep the listener, don't rewrite
// the href — UET can fire on click without preventing default.
Dallas call_to_click lifetime
8 events
Fired 8× after the GA property's Oct 2024 creation, then died Oct 23 2024 — the day the hijack snippet took effect. Zero since.
Pageviews vs call events, last 90 days
39.5k / 0
Zero events in both the Dallas property and the legacy rollup (387250340), where Dallas is the largest hostname. GA call data blind since Oct 2024 — and real calls plausibly lost.
Sites with the hijack
1 / 39
Dallas only. 13 sites run passive (harmless) Bing UET. Correlation between the hijack and dead call tracking is perfect.

Related June-2026 Dallas datapoint: quote conversions floored at 4 form_submit + 4 quote clicks — but the quote form was also broken that month (fixed July), so it isn't clean evidence of demand loss.

§ 4.6 — Network blind spots

Where the network can't see at all

Measurement audit, 39-site tracking sweep (R2)
SiteConditionDetail
dallasCall tracking deadtel-hijack (above); GTM noscript fallback also missing (TC has it; same container both sites)
austinGA4 property emptyGTM present but hostname missing from the container lookup; also loads a dead UA snippet pointing at a staging URL in production
pittsburgh, sc-ohioNo usable analyticsNo GTM; dead UA only; empty GA4 properties
fort-pierceZero analytics of any kind
salt-lake-cityWorks, awkwardlyOnly via the legacy rollup, under the old "Click to Call" event name
florenceNXDOMAINThe franchise actually operates at centralky.aaacwildliferemoval.com (works, GA 390759779); the dead florence GSC property is an orphan
san-antonioPossible double-countLoads two GTM containers — its 353 call events/90d may be inflated
phoenix, palm-beachesDouble-collectProperty + legacy rollup both collecting
Not in repo inventoryRollup reveals extra production sites: columbia, kansas, clarksville, rockwall

No GSC property exists for Dallas — its search data comes from the root domain property, page-filtered. Also outside GA entirely: the separate July form-mail outage, in which ~358 leads were lost because the quote form had no persistence before wp_mail().