Csupi Google Ads: completion and operating plan, revision 2

Evidence cut-off: 2026-09-15. This is a decision-ready plan, not a claim that Google has approved the new ads or that a phone click produced a job. Owner for the next edit: Matt after the employee listens to the guide. Account: Csupi Autómentés, Ads customer 4891907898, HUF, Europe/Budapest. Scope: 21 enabled EXACT | target Search campaigns; 23 non-removed legacy campaigns paused; 4 removed history. Read the latest provider state again before any account change. No new Ads, GTM, website, or budget write is authorized by this document alone.

Spoken companion: final employee transcript, private MP3 and WAV master. Charon / Gemini Pro TTS, verified playable duration 13:06 (786.1 seconds, ffprobe). MP3 delivered only to Matt's private Telegram chat, message 1588; no public audio-library upload or HTML review-host publication.

At a glance: outcome, proof and decision

End state / deliverable Current evidence, confidence Success proof and failure check Likely human choice, alternatives Must not do
A. All intended Search paths eligible and actually receiving relevant auctions 21 target RSAs created; all 21 still review-in-progress at last readback. serving_status=SERVING at campaign level is not ad approval or impressions. Pending, high confidence in the blocker; unknown current review outcome. Dated provider export: 21/21 target campaign/ad-group/ad/positive exact-keyword joins, 21/21 approved enabled RSAs, 23/23 legacy paused; separate fresh-day impressions/search terms/clicks per relevant campaign. Zero impressions requires a diagnosis, not an invented click. Explicitly fail if any enabled target ad is disapproved, if an enabled group has no ad/keyword, or a legacy campaign spends. Keep exact/legacy boundary. If approval is routine, wait/read back; if disapproved, fix documented destination/policy or appeal with evidence. No blanket legacy re-enable. No bypass of review, unsafe old ad re-enable, deletion of history, or budget increase to mask a delivery blocker.
B. Geographic, schedule, network, device, keyword and auction controls fit actual service Generic has explicit Budakeszi 40 km proximity; other 20 had no active LOCATION criterion after cleanup. HU language and 3 device rows each, positive exact keyword count 86; 13 Manual CPC, 7 Target Spend portfolio, one Generic Target Spend. Geo and effective portfolio bid partial. Read all 21 campaign settings and active criteria, Ads location/search-term reports, schedule and effective bidding strategy state. Test at least one intended and one clearly unintended locale/search query with non-billing diagnostic tools where possible; compare to client-confirmed service coverage. Diagnose low spend against policy, demand, rank, budget, geo and bid separately. Never declare city targeting from campaign name alone. Probable: Péter confirms served geography/hours; choose modest explicit geography based on actual jobs and search-volume risk, otherwise keep wider coverage with scrutiny. Bid/budget redistribution only within unchanged total after evidence; option to preserve price ceiling if marginal calls poor. No guessed radii, late-night service claim without staffing, broadened match type just to spend, or portfolio switch based on 0.01-HUF child field alone.
C. Phone widget click, not all phone links, is canonical optimization signal Ads GA4 action 6509672057 primary/included; real-call 7641831023, duplicate Ads tap 897422774, ad-call 272393221 secondary/excluded. GTM event tel_szam_katt tag 17 uses broad tel: Just Links trigger, tag 13 same trigger. Action counts MANY_PER_CLICK. A fresh 2026-09-08 to 09-14 GA4 read saw 9 aggregate events, 4 google / cpc; Ads saw 7 All conversions but 0 Conversions for the two proxy actions in that pre-remediation slice. Configured and aggregate activity, not widget-exclusive or runtime-proven. Consent-aware GTM Preview + controlled UI test shows one chosen widget tap fires intended GA4 event once, footer/plain tel: does not increase the primary widget event, consent-denied path does not violate policy; GA4 DebugView event/parameters and later Ads action readback with source, date and reporting delay. For every target campaign, read actual selected account/default or specific goal AND bidding strategy; primary alone is not enough. Verify actual conversion-optimizing bid mode after clean data, not merely primary action mapping. A no-dial event test does not prove genuine Ads attribution. Probable: both floating widget and the paid reveal widget count, but footer/plain number does not; proposed definition, not confirmed. Alternative: floating .widget-phone alone for tighter quality at the risk of missing paid reveal intent. Keep phone taps as an interim proxy; consider a higher-quality action only after matched qualified jobs. No fake gclid, synthetic conversion, real test call without separate arrangement, double tag, unrelated GTM publish or claiming a click is a call/job.
D. Customer-facing ads and landings are policy-safe, honest and reviewable 21 new localized RSAs, 21/21 canonical landing URLs HTTP 200; overview inventories 81 non-removed target ads, pins and possible combinations. Old destination disapprovals recorded; current new ads pending. URLs tested, actual combinations/claims need provider/client confirmation. Provider policy and enabled-ad readback, live HTTP final route/canonical/phone/consent on desktop/mobile, RSA asset/pin comparison against overview, client confirmation of every time/service promise, fresh screenshot of representative preview. Verify root versus live before any deploy. Probable: keep localized copy, remove unverifiable arrival-time promise if not confirmed; use client-approved service language. No new Akár 15 perc promise inferred from historical ad, old broken insertion, misleading preview-as-guaranteed-combination or production overwrite from stale dist/.
E. Spend and business quality are observable without false certainty Three existing daily pools 4,276.30 + 986.80 + 1,315.80 HUF, approximately 199,998.56 HUF at 30.4 days. Generic historical 30-day spend overlapped the change. No verified matched jobs/revenue. Budget high confidence, results low confidence. Weekly dated dashboard: spend, impressions, clicks, relevant search terms, lost impression share budget/rank, widget events, Ads conversions by action, client-confirmed answered calls/qualified leads/jobs, changes and lag. Sum campaign spend versus budget and compare before/after using non-overlapping complete days. Explain zero versus missing data. Matt records a post-change underdelivery alert level after seeing demand; if below that level, the operator diagnoses policy, demand, auction rank, bidding, geo and shared-pool diversion, not just budget. Probable: hold total cap and monitor; redistribute based on qualified leads and auction opportunity. If quality is poor, reduce waste before trying to fill budget. No automatic budget increase, attribution of all calls to Ads, promising fixed monthly spend, or optimizing solely for high click count.
F. Repeatable handoff and recovery C01 plan, receipt, inventory and independent audits exist, but pending gates not closed. Documented partial. Every changed object has before/after ID/status, provider receipt/time, constrained rollback, named owner and next readback. Employee can reproduce a one-page weekly report and tell client what is configured, proven and unknown without another document. Final status PASS only when A-E evidence holds, otherwise precise PENDING/PARTIAL with owner/date. Probable: one 9-14-minute Hungarian audio + this plan become baseline; employee proposes edits after listening, Matt approves only materially changed scope/spend/promise. No irreversible deletion, public hosting of Ads data, or silently converting an optional proof into a reason to abandon independent work.

Campaign coverage map and shared-budget risk

Pool Campaigns Current bidding/budget baseline What remains to prove
Primary (13) Buda, Érd, Diósd, Pomáz, Szentendre, Solymár, Budaörs, M0, M1, M5, M6, M31, M7 Shared 4,276.30 HUF/day; Manual CPC; last provider statuses vary from ELIGIBLE to LIMITED/NOT_ELIGIBLE/PENDING. 13 approved ads, fresh actual auctions, intended geographic reach, honest city/motorway copy, shared-pool allocation.
Secondary (7) Budakeszi, Csepel, Biatorbágy, Pilisvörösvár, Piliscsaba, Zsámbék, Törökbálint Shared 986.80 HUF/day; Target Spend portfolio, historical paused/placeholder-bid concern. Portfolio's fresh effective eligibility/bid and impressions; per-campaign volume so these are not silently starved.
Generic (1) Budakeszi 40 km Own 1,315.80 HUF/day; Target Spend with campaign bid ceiling, 40 km explicit proximity. Search-term relevance and lead quality; don't infer later spend from a partly pre-change 30-day sample.

All 21 have Google Search on and other sampled network flags off in the receipt; presence geo setting appears set, but explicit LOCATION rows are absent outside Generic. The prior audit found five primary campaigns LIMITED by historical disapproved destination ads, while the remediation paused old ads and submitted new ones. These are dated states, not a proof that the live warnings cleared. No target broad/phrase positive keywords in the post-write readback, but Google's exact match can include close variants. Four Generic negative keyword texts are recorded; negatives/search-term coverage is not yet certified as complete. The approved customer-level call asset exists, while direct association expansion validate-only failed: inherited serving is expected, not individually demonstrated.

Evidence chain and stage transitions

CONFIGURED -> REVIEWED -> ELIGIBLE -> IMPRESSING -> CLICKED -> WIDGET EVENT -> QUALIFIED CONTACT -> COMPLETED JOB

Each arrow is a separate observation. The most common false positive is calling a campaign “running” because it is ENABLED or reports SERVING while the new RSA is under policy review. Another is calling a tel: tap a completed call. Do not move to the next label without its dated evidence. Ads conversion data may lag; GTM Preview proves tag behaviour, GA4 DebugView proves collection for a test session, Ads readback proves import/goal, and Péter's records prove business quality. Keep those evidence routes separate.

Stage Checkpoint and evidence If it fails
0. Identity and freeze Customer 4891907898, 21 target IDs, 23 paused legacy IDs, shared budgets and conversion IDs checked immediately before any contemplated mutation. Narrow before-state export and recovery route saved. Stop cross-account writes; reconcile ownership or access first.
1. Policy and auctions 21/21 approved enabled ads and 21 campaign status/diagnostic joins; first complete-day impressions/search terms and observed clicks where demand exists. If review pending, recheck on provider cadence. If disapproved, repair exact cause. If approved but no impressions, isolate demand/geo/bids/budget.
2. Geography and pricing Actual campaign LOCATION rows, served-territory decision, effective portfolio state and lost share; no poor-fit terms dominating. Choose restriction versus coverage consciously; no guessed targeting or unilateral extra spend.
3. Widget measurement Widget-only tag scope, no duplicate or footer primary event, consent checks, GA4 test event, Ads action import and target goal inclusion. Keep signal PARTIAL; debug the failed layer before switching to automated conversion bids.
4. Business loop Client-confirmed answered calls/qualified leads/jobs joined to dated Ads data without assuming all calls came from ads. State business outcome UNKNOWN and collect the weekly log; do not over-credit proxy events.
5. Stable operations First two weekly comparable reports and change receipts; owner can explain each uncertainty and revert an affected change. Continue monitoring only the relevant gate; preserve safe settings and evidence.

Client conversation and probable decision record

  1. Approval/launch: “The new campaigns and pages are prepared, but approval and real impressions must be confirmed separately.” Probable choice: keep exact campaigns and legacy pause, repair only provider-documented rejections.
  2. Where/when: “A city name in a campaign is not a geographic boundary.” Ask Péter where and when trucks answer reliably. Probable choice: only evidence-backed location constraints, balancing too-wide wasted spend against too-narrow low delivery.
  3. Measurement: “A phone widget tap is a useful intent signal, not a completed call.” Probable choice to confirm: floating widget and paid reveal widget, not footer or other bare tel: links. Alternative floating .widget-phone only. The broad trigger must be corrected or precisely segmented before claiming widget-only tracking. Keep matched calls/jobs in a weekly log; do not call a manual CPC campaign conversion-optimizing merely because its goal is primary.
  4. Money/bids: “We have a total daily setup near 6,579 HUF; it need not spend fully.” Probable choice: hold total and redistribute after qualified-demand evidence. First diagnose any underdelivery; then Matt decides a realistic alert level, marginal CPC versus volume, and whether a small cohort can test conversion-aware bidding after trustworthy widget events. Never promise fixed spend or extra money.
  5. Copy: “Google may mix the submitted headlines and descriptions once approved.” Probable choice: retain localized truthful assets; Péter validates service and arrival times rather than us guessing them.
Decision owner Current decision state Most probable choice; live alternative Evidence that closes it
Péter: territory and staffed hours Pending confirmation; 20 non-Generic campaigns lack explicit active LOCATION criteria. Confirm served territory and answering hours, then choose evidence-backed restriction; alternative retain wider coverage to avoid lost demand. Péter's dated confirmation, all 21 active location/schedule rows and actual geographic/search-term report.
Matt: widget KPI definition Primary GA4 phone-click action configured; exact widget surface and consent-safe one-event firing unproven. Floating .widget-phone plus paid reveal widget, excluding footer; alternative floating-only if higher-quality intent. GTM trigger selector, consent-positive/negative Preview, widget-positive/footer-negative event/parameter test, GA4 DebugView and delayed Ads goal readback.
Matt: auction, allocation and underdelivery alert Existing total unchanged; Manual CPC does not automatically bid on conversions and no spend floor was approved. Keep pools and price protection until post-change demand/quality exists, then diagnose underdelivery and set an alert; alternative redistribute within total or test conversion-aware bidding in a small cohort after clean data. Before/after bids, shared budgets, strategy status, complete-day impressions/spend/lost share, Péter's qualified-contact log.
Matt: policy response All 21 new RSAs pending at last C01 policy recheck. Wait and re-read; alternative evidence-backed correction or appeal if a real rejection occurs. Provider policy and enabled-ad join for all 21, exact reason and same-ad later status; actual impressions where demand exists.
Péter: ad claims and job quality Arrival-time discrepancy is historical; no matched jobs/qualified-lead totals proved. Use only confirmed service/time claims and a minimal answered-call/qualified-lead/job log; alternative stronger copy or bid goal only after proof. Written claim confirmation, current asset/landing parity and dated customer call/job records.

The employee's likely post-listening edits are the served territory/hours, the exact widget surface, acceptable qualified-lead definition, emphasis across primary/secondary/Generic, and client-approved ad claims. Record alternatives and tradeoffs beside the changed plan row, not in a hidden task list. A budget rise, historical campaign re-enable, or changed client promise is outside these presumed defaults and needs its own decision. Final Hungarian employee transcript is the spoken companion; LUNA critique corrected approval, location, historical-bid, authority and proof language before audio synthesis.

Ownership, safety and resume point

The previous remediation receipt is a reproducible dated baseline, not a green serving certificate. A dedicated LUNA delivery agent hit model capacity, so the delivery synthesis here rests on the prior audited provider receipt, the operator LUNA brief and official help, not a new post-15:47 RSA policy export. The measurement LUNA did perform a fresh 20:00 CEST Ads/GA4 read, but not a complete new policy export. RSA approval and true post-change impressions remain UNKNOWN until re-read. Ads writes in the next execution require a fresh explicit narrow instruction or approved execution scope, since this project's standing rule is read-only without it. GTM/site publish and production deployment are separate scopes. Private artifacts remain localhost or in the authorized app, not on a public comment host. Do not expose credentials, fabricate Ads clicks/calls or perform payment/lead submission. Preserve the dirty worktree and narrow rollback evidence.

Resume in order of materiality: (1) Google RSA policy readback for all 21, (2) all 20 non-Generic location criteria and client coverage decision, (3) widget-only consent/runtime test and Ads goal selection/effective bidding, (4) post-change serving and first matched client quality log. A failure in one gate does not cancel safe independent checks. The exact implementation may change after the employee hears the audio; this document states measurable end states, not an authority to improvise account writes.

Source ledger