| name | google-ads-search-launch |
| tier | playbook |
| calls | ["start","think","site","ads","end"] |
| status | draft |
| description | Reusable Noontide draft playbook for validating an offer with a tightly scoped Google Ads Search launch, a paid-traffic lander, explicit approval gates, and a capped review window. |
google-ads-search-launch
Source playbook: playbooks/google-ads-search-launch/playbook.md. This file
is the Claude Code playbook shell over that runtime-agnostic source.
This is a reusable draft operating playbook, not a one-off ad checklist. Use it
with manual operator judgment when the operator wants to test whether a real
offer can turn paid search intent into leads, calls, bookings, deposits, trials,
or another concrete business outcome.
It is usable today as a manual recipe and run-record template. It is not a
single executable orchestrator, does not automate Google Ads/GTM, and is still
collecting evidence from operator runs.
The playbook's job is to turn one offer into one paid-search proof run:
- define the bet and success criteria;
- prepare or verify the lander and conversion path;
- build a tightly scoped Google Ads Search plan;
- preserve manual approval gates before spend or provider mutation;
- record the run as a push playbook with review evidence and an outcome hook.
It does not publish campaigns, change budgets, upload conversions, publish GTM,
or mutate provider accounts. Those steps stay manual or provider-native until a
future accepted adapter ships with approval gates and smoke evidence.
Playbook Versus Run
This file is the reusable Noontide recipe: an opinionated paid-search
offer-validation sprint. The run belongs in the business repo:
pushes/<push>/playbooks/google-ads-launch-plan.md
Create or update that run file from
templates/push-playbook.md. The run file is the
approval record, provider-boundary record, launch checklist, review-window
record, and outcome hook for a specific offer/push.
This playbook follows the authority model in docs/playbooks.md:
official Google Ads docs are the source of truth for platform facts, Main
Branch owns the provider boundary and run-record shape, and this file owns the
Noontide defaults for testing many offers quickly with tight control. If the
operator forks a default, record the fork and reason in the run file.
Playbook Defaults And Fork Points
Use references/noontide-approach.md as the
style and decision rubric.
For B2B local-services lead-form campaigns, also use
references/b2b-local-services-field-notes.md.
Defaults:
- one offer, one primary conversion, one review window;
- one conversion path chosen up front: call/booking, Stripe/deposit, lead form,
trial, or another concrete action;
- Maximize Conversions for the first launch window when the selected primary
conversion is verified; use Maximize Clicks or manual CPC only as a written
fork when conversion tracking is absent or intentionally delayed;
- exact and phrase match first; broad match only with a written reason;
- geography is explicit: local radius, multi-city service area, statewide,
national, or multi-location; do not reuse local defaults for national tests;
- campaign settings, negatives, sitelinks, callouts, structured snippets, and
skipped assets are part of the plan, not afterthoughts;
- market-intent research precedes asset writing when the offer is new, the
audience/search language is thin, or the geography is unfamiliar;
- a small proof budget is only useful when CPC, conversion rate, and business
value make the result interpretable;
- account history is useful when it exists, but a new offer can start from
offer/customer/keyword reasoning and lander readiness;
- Search Partners, Display Network, AI Max, Final URL Expansion, broad match,
price assets, lead forms, call assets, and automated URL options default to
off or skipped for first proof runs unless the operator records why this run
should fork;
- no spend before the operator approves the lander, conversion action, consent
posture, billing, budget, campaign structure, copy/assets, asset
destinations, and review criteria;
- post-launch judgment is
continue, change, or stop, written as an
outcome or push review.
Required Inputs
- active offer and audience;
- explicit bet or success criterion for the proof run;
- launch push, or permission to create one;
- lander or site repo, if one exists;
- budget cap, review window, geography shape, and business value of the
conversion;
- conversion path: call/booking, Stripe/deposit, lead form, trial, or other;
- market-intent research, competitor/offer positioning, customer language,
objections, proof, and exclusion notes when available;
- provider readiness from
mb status --json --peek and mb connect plan;
- measurement readiness from
mb site check when a site repo exists;
- sanitized account history or read-only provider facts only when already
approved and available.
If no Google Ads account history exists for this offer, say so and continue.
Do not invent prior winners or require pre-offer account scraping.
Flow
- Start and scope. Use
/mb-start facts. Identify the offer, push, bet,
current repo health, provider readiness, and whether this is a new offer or
a rescue of an existing campaign.
- Bet and KPI. Use
/mb-think when the success criterion is vague. Press
for what would make the spend a win, a useful loss, or an inconclusive test.
- Lander and conversion. Use
/mb-site and mb site check before calling
the campaign launchable. Missing measurement can still allow copy drafting,
but not spend approval.
- Market-intent pass. If the offer is new, thin, or entering a new
geography, use
/mb-think to research buyer/search intent, competitor
offers, customer language, objections, proof, and bad-fit exclusions before
writing assets. When 2+ lanes are needed, run them as parallel foreground
research agents and persist findings to research/. If the findings change
durable offer, audience, market, proof, or positioning truth, update the
relevant core/ files after operator approval or record proposed core
updates in the run file.
- Campaign plan. Use
/mb-ads launch-plan and load the Google Ads campaign
plan reference. Generate the full reviewable spec: campaign settings,
bidding, geography mode, ad groups, keywords, negatives, RSA headlines and
descriptions, display paths, sitelinks, callouts, structured snippets,
skipped assets, budget, manual provider steps, and approval gates.
- B2B local-services pass. When the offer is B2B local services, check the
field notes for GA4/GTM/Ads import order, Search-only defaults, geo
presence, Search Partners, Final URL Expansion, AI Max, negative categories,
UI gotchas, and volume calibration.
- Run record. Write or update the push playbook run from the template.
Record defaults used, forks taken, asset rationale, and skipped asset
rationale. Record research files and any core updates made or proposed. Keep
provider refs safe and do not store raw account exports.
- Manual launch. The operator performs Google Ads/GTM/billing/spend steps
manually unless a future adapter is accepted.
- Review. Use
/mb-ads check after the review window. Write the result to
the run record and link an outcome/log file.
- Checkpoint. Use
/mb-end or mb checkpoint to save accepted artifacts.
Current Checks
Run these when applicable:
mb status --json --peek
mb connect plan
mb connect doctor --json
mb site check "$SITE_REPO" --business-repo "$BUSINESS_REPO" --json
mb validate
These checks help the agent stay grounded. They do not prove terminal Google Ads
campaign creation is supported.
Future Surface
The playbook should eventually enable deeper checks when those surfaces exist:
- Google Ads read adapter or sidecar for campaigns, search terms, spend, and
conversions;
- daily paid-search metrics cache tied to bets/pushes;
- deterministic playbook checks for Search Partners, Final URL Expansion,
AI Max, primary conversion presence, and provider review gates;
- playbook health in
mb status;
- provider-specific repair commands;
- dashboard view that hides technical plumbing by default but can expose issues,
branches, PRs, provider refs, and evidence for technical operators.
Until then, keep the run useful: clear inputs, clear approvals, clean manual
steps, and a concrete review decision.