SEO Navigator° Knowledge Pack

Google Ads — Knowledge Pack

You need to sign in to load or edit knowledge packs. The first email you register becomes the workspace admin.

Internal · SEO Navigator · Auto Detailing Vertical

Detailing Google Ads — MASTER Benchmark + Knowledge Pack

ONE workbook that captures (a) the SEO Navigator skill defaults, (b) published 3rd-party benchmarks, (c) the agency's aggregated client averages, and (d) per-client overrides — for every Google Ads parameter the agency tunes.

Pack version v3 MERGED Vertical Auto detailing (ceramic, PPF, tint, paint correction) Owner Tung (media buyer) · Sang (strategist) · Brooklyn (CRM)

ATwo use cases

This pack runs in two directions at once — per-client tuning, and a feedback loop that improves the agency's skill defaults.

Use case A

Per-client onboarding

Tung fills the This Client column for a new client (Sun Stoppers, Elite Finish, etc.). Empty cells = use skill default. When the agent runs any Google Ads skill, populated overrides win.

Use case B

Skill-improvement loop

Sang rolls up actual values across all active client workbooks into the Agency Aggregate columns quarterly. Variance Flag fires when agency avg diverges >20% from skill default. A flagged row = signal to update the skill markdown.

BCell legend

Light blue
Per-client input — fill in for this client. Replace [FILL IN] placeholders.
Light green
Agency aggregate — rolled up across all clients quarterly by the strategist.
Light gray (italic)
Skill default or published benchmark — read-only reference from the skill markdown today.
Light beige
Published 3rd-party benchmark (LocaliQ, WordStream, etc.) — context only.
Variance flag
Agency aggregate differs from skill default by >20% — signal to update the skill markdown.
🔒 Locked sections
Append-only logs (Decision Log, Pipeline Negatives). Never delete rows, even on failures.

CHow the AI agent uses this

When any Google Ads skill is invoked for a client, the agent reads this Knowledge Pack first. It treats populated This Client cells as overrides. If empty → fall back to Agency Aggregate. If that's empty → fall back to Skill default (which is what's in the skill markdown today).

Tip › Fields auto-save as you type. Use Export JSON to back up or pass to another buyer. Use Import JSON when starting a new client from an existing template.
30-Second Briefing

Quick Reference Card

The buyer-of-record fills these out and updates whenever any of them change. Pin to the top of the account folder.

Account: [unset]
If you had 30 seconds with a new buyer before they pulled this account on Monday, this card is what you'd say.
Account
[Client name] — [Google Ads ID]
Primary metric
e.g. Booked appointments with deposit paid
CPA ceiling — overall
$
CPA ceiling — highest-LTV service
$
Conversion value rule
e.g. Cap at $1,500 in Ads feed
Three things never to do — 1
Three things never to do — 2
Three things never to do — 3
Top diagnostic rule
e.g. CPA spike → tracking, search terms, budget, LP, client-side BEFORE bid strategy
Last big lesson
Next planned change
Date + what
Identifiers & Ownership

Metadata

Client name
Client slug
Google Ads account ID
MCC link status
Primary services on Ads
Monthly budget at last review
Account age (with us)
Account age (overall)
Pack owner (media buyer)
Strategist on account
CRM lead
Last full review date
Next scheduled review
60-second briefing Pending — Account Brief

01 — Account Snapshot

If a new buyer had to take over tomorrow with only this section, what would they need to know?

What this client actually sells
In their own words, NOT industry-speak. e.g. 'Tinted comfort and protection for car owners who park outside in Charlotte heat.'
Why they hired us
What 'winning' looks like
In their own words. e.g. '10 booked ceramic appointments a month, not 100 quote requests'
What they DO NOT want, even if it lowered CPA
e.g. 'Cheap quick-detail volume that clogs the calendar'
Three things that surprise people — 1
Three things that surprise people — 2
Three things that surprise people — 3
Conversion value & LTV math Pending — Revenue per Lead

02 — Service Economics & LTV

Per service: skill default proxy value, agency aggregate across clients, this client's true value. Variance flag fires when agency aggregate ≠ skill default by >20%.

Skill default
Published benchmark
Agency aggregate
This client
Variance >20%

AConversion value per service= avg ticket × close rate

→ scroll table horizontally
Parameter Skill default SKL 3rd-party benchmark PUB Agency avg AGY Agency range (low–high) AGY n (clients) Last reviewed THIS CLIENT CLI Variance Notes

BLTV math (narrative)

Definition of LTV used here
e.g. 'First-job revenue + (12-mo repeat rate × avg repeat ticket) — only first-party data'
Highest LTV service (and why)
Loss-leader / low-LTV service we still run
e.g. 'Basic detail at $175 — door opener for ceramic upsell'
Payback period (months)

CPer-client derivation table

Fill avg ticket / cost / close rate / repeat to compute client-specific conversion value.

→ scroll table horizontally
Service Avg ticket ($) Avg cost ($) Gross margin ($) Close rate Repeat rate 12mo Computed conv. value Confidence Notes
What Smart Bidding must not touch Partial — Revenue pending

03 — Sacred Rules

If 2% of jobs make up 60% of revenue, an unconstrained tROAS will chase that long tail and starve volume. Lock down the rules here.

ARevenue concentration

% jobs vs % revenue
e.g. 'Top 2% of jobs = 60% of revenue'
Is the long tail protected by a value cap?
If yes, conversion value cap rule
e.g. 'Cap at $1,500 in Smart Bidding feed; raw revenue tracked in CRM'
Conversion action where cap is applied
e.g. 'Action B — Closed Sale (Capped)'

BOff-limits items (require strategist sign-off)

ItemRationale CLI
Brand campaign budget
LSA budget
Specific campaign (name it)
Specific audience or RLSA list
Specific conversion action
Other

CMandatory always-on rules

SettingSkill default SKLClient override CLI
Display NetworkOFF (always)
Search PartnersOFF (default; tested quarterly)
Location targetingPresence only — no exceptions
Auto-applied recommendationsAll OFF unless explicitly approved
Ad rotationOptimize: Prefer best performing
Auto-taggingON (required for tracking)
Trust levels & signal quality Pending — Revenue per Lead

04 — Conversion Quality Map

What each conversion action actually means for THIS client. Which signals are trustworthy, which are noisy, which are dangerously misleading.

→ scroll horizontally
Conversion action What fires it True close → $ rate (%) CLI Trust level Notes

ASignals to demote or kill

Events that fire as conversions in Ads but don't lead to revenue. e.g. '15-sec phone calls fire the call event but are 80% wrong-number / sales calls.'

Signal #1
Signal #2
Signal #3
Signal #4
Signal #5

BSignals to elevate

e.g. 'Booked-and-deposit-paid event from GHL. Imported as Stage 3 offline conversion. Weighted highest for Smart Bidding.'

Signal #1
Signal #2
Signal #3
Signal #4
Signal #5
Append-only — never delete Pending — Revenue-by-keyword

05 — Pipeline-Quality Negatives

Search terms that look great in Google Ads (good CTR, conversion-action firing) but close at near-zero in the CRM. They make dashboards look healthy while quietly destroying the account. Maintain weekly.

🔒 APPEND-ONLY · Pipeline-Quality Negatives · Never delete rows
Search term / pattern Why it triggers a "conversion" in Ads Why it doesn't close Action taken Date added

APatterns to watch for on this account

e.g. 'Fleet inquiries — high lead volume, low close rate because we don't service municipal contracts. Pre-emptively neg fleet, city, government, municipal.'

Pattern #1
Pattern #2
Pattern #3
Pattern #4
Pattern #5
Fix-order per scenario

06 — Diagnostic Runbook

Order matters. The default rule is NEVER to touch bid strategy first — that's almost always the symptom, not the cause.

Research timing · devices · seasonality · geography Partial — radius profile pending

07 — ICP Behavioral & Geography + Ring Profiles

Replaces generic skill defaults with what THIS client's customers actually do. Overrides the radius-optimizer rings, remarketing-engine windows, and campaign-builder ad scheduling.

AResearch-to-buy timing

When does the ICP research?
e.g. 'Sundays 6–10pm + Tuesday lunch'
When does the ICP book?
e.g. 'Mondays 8–11am the day after researching'
Avg gap first click → booked appt
e.g. '3.2 days'
Implication for ad scheduling
e.g. '+30% bid Sun 18:00–22:00 and Mon 08:00–11:00'
Implication for RLSA window
e.g. 'Hot-intent window 72 hours, not 7 days'

BDevice patterns

Primary research device
Primary booking device
Implication

CSeasonal patterns

SeasonDemand pattern (skill default)Action we take CLIPriorityNotes

DGeography & distance willingness

Avg miles driven for primary service
e.g. '8 miles for ceramic, 3 miles for full detail'
Furthest paying customer last 6 mo
e.g. '47 miles for $4,200 PPF Full Body'
Implication for radius rings
e.g. 'Override default. Inner 0–10mi, mid 10–25mi for ceramic, extend outer to 50mi for PPF only.'

ERing profiles & bid modifiers

→ scroll horizontally
Parameter Skill default SKL Pub benchmark PUB Agency avg AGY Agency range n Last reviewed THIS CLIENT CLI Variance Notes
Pre-decide once · stop re-deciding weekly

08 — Google 'Recommendations' Playbook

Default for SEO Navigator clients = Refuse unless explicitly approved.

RecommendationSEO Navigator default SKLThis client (override) CLIReason
Auto-apply settings on this account › Document the actual checkboxes. Default override: Ad & assets, Measurement, Remove redundant keywords, Remove conflicting negative keywords.
Primary metric · CPA range · strategy progression

09 — Bidding & CPA Targets

APrimary success metric

The one we optimize for, in conflict with all others.

Primary metric
e.g. 'Booked appointments with deposit paid. Not leads, not phone calls.'

BAcceptable CPA range by service

→ scroll horizontally
Parameter Skill default SKL Pub benchmark PUB Agency avg AGY Agency range n Last reviewed THIS CLIENT CLI Variance Notes

CFloor / Target / Ceiling CPA per service

Service Floor CPA ($) CLI Target CPA ($) CLI Ceiling CPA — kill above ($) CLI Skill default range SKL Notes

DBid strategy progression

Current strategy
Why this strategy now
What triggers next graduation
What triggers a step back

EBid strategy decision tree — skill default, for reference

Conv vol / moStart strategyGraduate whenGraduate toNotes
<15Maximize Clicks (w/ max CPC cap)15+ conv in 30 daysMaximize Conversions (no target)Most new detailing accounts
15–30Maximize Conversions (no target)30+ conv in 30 daysMax Conv w/ Target CPA
30–50Target CPA (= actual + 10% buffer)50+ conv & values trustworthyTarget ROAS
50+Target ROAS or tCPA—Portfolio bidding across campaignsFull Smart Bidding capability

FWhen to override Smart Bidding manually

Override rule #1
e.g. 'If sale closes >$5,000, do NOT feed full revenue to Google. Cap at $1,500 in offline import.'
Override rule #2
Override rule #3
CPC benchmarks & match-type graduation Partial — no industry benchmark for some rows

10 — CPC & Match-Type

ACPC benchmarks by service / cluster

→ scroll horizontally
Parameter Skill default SKL Pub benchmark PUB Agency avg AGY Agency range n Last reviewed THIS CLIENT CLI Variance Notes

BBudget allocation by intent tier

TierIntentMatch typesSkill default % SKLThis client % CLINotes

CMatch-type graduation gates — skill default

From → ToVolume gateOther gatesHold timeNotes
Exact+Phrase → add Broad≥30 conv in 30–60 days & CPA within avg benchmarkNegative list mature; LP CVR stable30–60 daysMOST important rule for low-volume detailing
Maximize Clicks → Max Conv≥15 conv / 30 daysTracking verified30 days
Max Conv → Target CPA≥30 conv / 30 daysSet tCPA = actual + 10%30 days
Target CPA → Target ROAS≥50 conv / 30 daysConversion values trustworthy30 days
Performance-auditor KPIs

11 — KPI Benchmarks

Skill default = the 'Average' band in the 4-band benchmark from the performance-auditor skill. Variance flag fires when agency aggregate diverges >20% from skill default.

AKPI numeric benchmarks

→ scroll horizontally
Parameter Skill default SKL Pub benchmark PUB Agency avg AGY Agency range n Last reviewed THIS CLIENT CLI Variance Notes

BFull 4-band KPI benchmark

MetricBelow AvgAverageGoodExcellentSource

CWasted spend categories

CategoryDetectionTypical waste %FixAgency observed % CLINotes
Quick lookup per skill Pending — clarify skill behaviors

12 — Skill Override Map

When an agent runs a specific Google Ads skill on this account, what's the override? Empty Override column = use skill default.

SkillDefault behavior SKLOverride for this client CLIPointer (detail tab)
Reports · language · sensitivities · cadence Pending — client interview

13 — Client Communication

AWhat they care about in reports (priority order)

Priority 1
Priority 2
Priority 3
Priority 4
Priority 5

BWhat they DO NOT want to see

Avoid #1
e.g. 'Impression share, ad strength scores, anything that doesn't tie to revenue'
Avoid #2
Avoid #3
Avoid #4

CTheir language ("they say → they mean")

They say…They mean…

DSensitivities

Sensitivity #1
e.g. 'Don't mention competitor names X / Y / Z'
Sensitivity #2
e.g. 'CPA is a trigger word — they want to talk in cost-per-booked-job'
Sensitivity #3

ECommunication cadence

Weekly check-in
Yes/No — day, channel, owner
Monthly report
Format: Looker / PDF / Slack
Emergency contact
Name + channel
Append-only — never delete Pending

14 — Decision Log

Every non-trivial change, dated. Append-only — never delete entries, even if they describe a failure. Goal: 6 months from now we can answer "why did we do that?" and "have we tried this before?" without guessing.

🔒 APPEND-ONLY · Decision Log · Never delete rows
DateWhat changedWhyExpected outcomeActual outcome (fill 14–30 days later)Lesson / next move
🔒 APPEND-ONLY · Closed loops · Won't repeat without new info
DateClosed loop entry

The Knowledge Pack System

How data flows from this Knowledge Pack into the SEO Navigator Claude skill suite — and how that loop improves the agency's defaults over time.

LAYER 1 · INPUTS LAYER 2 · KNOWLEDGE PACK LAYER 3 · CLAUDE SKILLS Client (Tung) Onboarding inputs Per-client overrides Strategist (Sang) Quarterly aggregation Across all clients CRM (Brooklyn) Revenue per lead Close rates · LTV Published data LocaliQ · WordStream Industry benchmarks Google Ads — Master Knowledge Pack v3 17 tabs · 4 data columns per row · Variance flag auto-fires when Agency ≠ Skill default by >20% Skill default Published benchmark Agency aggregate This client override Variance flag loop → update skill .md campaign-builder reads → 09 keyword-architect reads → 10 conversion-tracker reads → 02 + 04 performance-auditor reads → 11 + 06 radius-optimizer reads → 07 remarketing-engine reads → 07 When a skill runs: read THIS CLIENT first → else Agency aggregate → else Skill default

Skill ↔ Tab map

Each Claude skill below pulls overrides from specific tabs in this pack before falling back to its built-in default.

google-ads-campaign-builder
Tier-based campaign structure with bid strategy & budget allocation
Reads: 09 Bidding & CPA · 03 Sacred Rules
google-ads-keyword-architect
3-tier intent buckets, exact + phrase first, match-type graduation
Reads: 10 CPC & Match-Type · 05 Pipeline Negatives
google-ads-conversion-tracker
Conversion actions, proxy values, offline import setup
Reads: 02 Economics & LTV · 04 Conversion Quality
google-ads-performance-auditor
100-point audit using 4-band KPI benchmarks
Reads: 11 KPI Benchmarks · 06 Diagnostic Runbook
google-ads-radius-optimizer
Tiered rings + bid modifiers by market density
Reads: 07 ICP Behavioral & Geo (E. Ring Profiles)
google-ads-remarketing-engine
5-tier audience ladder + seasonal re-engagement
Reads: 07 ICP Behavioral & Geo (A. timing, C. seasons)
google-ads-copy-lab
15 messaging angles, RSA generator, A/B framework
Reads: 13 Client Communication · 01 Account Snapshot
paid-organic-crosswalk
Overlap dashboard between Ads keywords & organic rankings
Reads: 13 Client Communication (excluded items)

Conversion trust ladder

The agent treats these three stages differently when feeding signals to Smart Bidding.

1

Onsite signals

Stage 1 · Medium trust
  • Phone Call ≥60s — High
  • Form Submit — Med (varies)
  • Online Booking — High
  • Click-to-Call — Excluded ($0)
2

CRM stage offline import

Stage 2 · High trust
  • Qualified Lead (SQL)
  • Appointment Booked
  • Imported via GCLID from GHL
3

Closed sale w/ revenue

Stage 3 · Gold standard
  • CRM stage → closed-won
  • Revenue value (capped per rule)
  • Weighted highest for Smart Bidding

Diagnostic priority

When the account misbehaves, fix order matters. Never touch bid strategy first — it's almost always the symptom.

1

Tracking integrity

Fired counts vs CRM lead counts match? Auto-tagging still ON? GTM container live? No new conversion action accidentally promoted?

2

Search term drift

Last 7 days vs prior 7 days. Any new high-spend irrelevant queries? Check against Tab 05 Pipeline Negatives.

3

Budget changes

Anyone bumped budgets? A sudden +50% on a Smart Bidding campaign re-triggers learning.

4

Landing page

Page speed, form working, page even loading? Check Search Console for sudden indexing issues.

5

Client-side issues

Phone working? Voicemail full? Booking calendar full? Receptionist on vacation?

6

Only now — bid strategy

Has it been stable >14 days? Has anything been bulk-edited? Reference Tab 09 D for graduation thresholds.

The variance flag loop

The same workbook drives two parallel feedback loops. Together they prevent skill drift and keep the agency's defaults grounded in observed reality.

FORWARD LOOP · Skill → Client

Every time a skill runs for a client, it reads THIS CLIENT first. If empty, falls back to Agency aggregate. If empty, falls back to Skill default. The buyer can override anywhere in the chain.

REVERSE LOOP · Clients → Skill

Quarterly, Sang rolls This Client values across all active accounts into the Agency Aggregate column. When that aggregate diverges >20% from Skill default, the variance flag fires — signal to update the skill .md.

Why this matters › Without the reverse loop, skill defaults rot. They were written from one buyer's instinct at one moment in time. The aggregate column converts the agency's accumulated reality into something Claude can read.
Citations & references

Sources

SourceTypeDateWhat was used / URL
✓ Saved