google-ads-audit

name

google-ads-audit

description

Assemble a complete Google Ads account audit as a single scrolling HTML narrative document. Orchestrates all Google Ads analysis skills, orders findings by dollar impact, and renders them as a professional audit report with Executive Summary, KPI Scorecards, per-section findings tables + callouts, What's Set Up Right, and a prioritized Action Plan. Use this skill when the user asks to audit, review, or health-check a Google Ads account. Output is a Porter report (single HTML page), not a dashboard.

Google Ads Audit — Assembly Orchestrator

What this builds

A single-page HTML narrative document delivered as a Porter report. Not a dashboard — no charts. The document scrolls from Header to Action Plan, with each section backed by live data and ending in a specific recommendation.

Reference audits — read these to calibrate tone, density, and finding logic only. They predate this template and ship the older hand-written teal/navy CSS, so do not copy their look:

• Harper Health (Search, small account): report_id 69af1804-c1d8-40e7-8cd2-b584c7cf27c0
• PolicyMe (Search + DG, large account): report_id 80939d61-817f-45e5-b96a-25c8ba34b039

Visual standard = this skill's own template (pages/main.html + the Porter Design System themes), NOT those reference reports. Every audit renders from that template in one data-theme (default cream) so the output is consistent run to run. Deploy via references/deploy.md.

---

Step 0 — Gather inputs

Input	How to get it	Why
account	list_accounts (connector: google-ads)	All queries need customer_id-login_customer_id
date_range	Ask user or default to last 90 days + prior 90 days	90 days gives enough signal; 30 days only for small-spend accounts
brand_terms	Ask user (optional)	Required for brand vs non-brand split in search terms
campaign_types	query_data → google_ads_campaign_advertising_channel_type	Gates which sections apply

Run campaign fingerprint first — determines which sections to include or skip:

• All Search → full audit
• PMax present → skip Keywords/Search Terms sections; note the gap
• Demand Gen present → add DG section
• Shopping present → add product P&L note (out of scope for this skill version)

Then run a pre-flight access check — BEFORE Step 1, never skip. Two quick calls confirm the connector can actually serve the whole audit, so a permission hole surfaces up front instead of as a hole in section 06+ after the work is done:

1. Bid-strategy / budget route — call the campaign.list connector action (a tiny GAQL: a few campaign rows + campaign_budget.amount_micros). It is the preferred route for true bid targets and budgets when it works.
2. Geo route — a 1-row query_data on ["google_ads_geo_target_region","google_ads_cost_micros"].

Branch on the result:

• Both OK (HTTP 200 / rows) → proceed; the Bid-strategy and Geography sections are safe.
• campaign.list returns 403 / "not configured" / a manager-account permission error → this account has report access but no Google Ads API access for this customer. Fall back to the GAQL report.query route for targets; if that also fails, tell the user plainly and offer a partial audit (Bid-strategy dropped, footer gap noted). Never let it surface as a silent hole mid-build — that is the failure mode this check exists to prevent.
• Either returns reauth_required → surface the reauth URL and resume after the user reconnects.

Reconciles the campaign.list question: earlier notes claimed campaign.list "fails for ~75% of accounts." Live runs show it working as the reliable route on accounts where API access is configured, and 403-ing where it is not. So don't assert either way — the pre-flight tests it per account and branches. (The report.query GAQL route stays as the documented fallback.)

---

Step 1 — Run all analysis skills in parallel

Run these concurrently after Step 0. Each returns findings + a verdict.

Skill	Output needed
measurement/conversion-tracking	Is the conversion signal trustworthy? List of primary actions, shallow ones, missing values
campaigns/spend-allocation	Per-campaign spend / conversions / CPA / IS — who's wasting, who's capped
campaigns/impression-share	IS earned + lost-to-budget + lost-to-rank per campaign
campaigns/bid-strategy	Strategy type + target vs actual per campaign
campaigns/value-based-bidding	Is value bidding set up correctly?
campaigns/campaign-settings	Location = Presence? Search Partners / Display off?
ads/metrics	QS 3 pillars per keyword (not numeric — categorical only)
search-terms/performance	Winning / Watch / Waste by spend
search-terms/n-grams	Top waste buckets by 1/2/3-gram pattern
search-terms/classifier/branded	Brand leak detection + competitor terms
search-terms/negatives	Layer 1 (presence) + Layer 2 (brand containment) — feeds the search terms section
search-terms/match-types	Spend concentration by Broad/Phrase/Exact — is budget too heavy on one match type?
ads/copy	RSA Ad Strength + asset labels (PENDING/GOOD/LOW/BEST)
ads/assets	Sitelinks / callouts / snippets / images present?
ads/alignment	Keyword → Ad → Landing chain coherence
ads/landing-cro	Live page scrape: value prop / CTA / proof / form friction
segmentation/audience/demographics-audit	Age/gender bid adjustment opportunities
segmentation/audience/geography	Province/city CPA vs account average
segmentation/audience/devices	Desktop vs mobile efficiency gap
segmentation/time/cyclical	Day-of-week + hour performance vs average
campaigns/demand-gen	DG campaign audit vs Search baseline (only if DG campaigns exist)

---

Step 2 — Score and order sections by impact

Each skill returns a verdict and a dollar estimate. Score sections before writing:

Severity rule — assign based on $ at stake:

• HIGH — affects >15% of total spend, or conversion tracking is broken (always HIGH)
• MED — real impact but not the biggest lever; or structural fix with medium spend exposure
• LOW — housekeeping, optimization when ready, or low-volume signal

Ordering rule:

1. measurement/conversion-tracking is always 01 — a broken conversion signal corrupts every other number. Run and write this first regardless of severity score.
2. Remaining sections ordered by HIGH → MED → LOW, then by spend at stake within each tier.
3. What's Set Up Right and Action Plan always come last.

---

Step 2.5 — Verify every finding before you write it (anti-hallucination gate)

The audit is read as fact by a client, so a false flag is worse than a missed one. Before a finding goes in the document it must pass all four:

1. Cite the source. Every HIGH/MED finding traces to a specific pull — name the field + window it came from. If you can't point to the pull, it doesn't ship.
2. Cross-check with a second lens. Re-derive the claim a different way before trusting it:
   • "Action didn't fire / isn't tracked / 0 conversions" → check all_conversions too AND a wider/earlier window. conversions=0 with all_conversions>0 = secondary, fires (not absent); a recent-month 0 may be reporting lag (offline/PURCHASE import late). Say the verified state, never "fired zero" off the primary field alone. (This exact trap produced a false "Paying Customers fired zero" finding in testing — the action fired ~11/mo as a secondary.)
   • "Campaign spent $X for 0 conversions" → confirm it's the primary conversions field and re-check against a settled window before calling it pure waste.
   • Any rate (CTR/CPA/ROAS/CVR) → recompute from base counts; never trust an aggregate ratio field.
   • Quality Score → categorical pillars only; numeric QS from query_data is corrupted.
   • Bid targets / "primary" status → primary_for_goal is True for every action (connector bug); derive primary from conversions>0, derive targets from campaign.list (per the Step 0 pre-flight).
3. Calibrate the wording to the evidence. Absolute words — "zero", "never", "not tracked", "broken" — are only allowed when the cross-check confirms them. When the data is a snapshot or ambiguous, downgrade: "fires but secondary", "0 in May (provisional — verify in-account)".
4. When unresolved, flag don't assert. If a finding can't be corroborated, render it as a c-review / "verify in-account", not a c-broken. Never fabricate a number to fill a section.

---

Step 3 — Write the document

Document skeleton (in order, no exceptions)

1. Header
2. Executive Summary
3. KPI Scorecards
4. Legend
5. Sections (ordered by impact per Step 2)
6. What's Set Up Right
7. Action Plan
8. Footer / Methodology

---

1. Header

<header>
  <div class="brandrow">
    <div>
      <span class="eyebrow">Google Ads Account Audit</span>
      <h1>[Client Name]</h1>
      <p class="sub">[Business description — 1 line]</p>
    </div>
    <span class="ga-badge"><span class="ga-dot"></span> Google Ads</span>
  </div>
  <div class="metarow">
    Period · Account ID · Campaign type · Markets/geography · Audit date · Currency
  </div>
</header>

---

2. Executive Summary

Structure:

• <span class="tag">Executive summary</span>
• <h3> — One diagnosis headline. Lead with the biggest finding and the dollar number. Never generic ("results were mixed"). Example: "$483K spent and CPA fell 32% — but only because the conversion signal got noisier, not because buying got smarter."
• 2–3 paragraphs. Para 1: what the top-line looks like and why it's misleading (or not). Para 2: where the money leaks. Para 3: what's working.
• <div class="do"><b>Do first:</b> — exactly 3 actions, the ones with most $ at stake. Reference them to sections.

Tone rules:

• Every sentence names a number or a named campaign/keyword/page. No vague statements.
• Lead with the problem, not the context.
• "Do first" = the 3 actions that, if done this week, address the majority of waste.

---

3. KPI Scorecards

6 cards, always:

1. Spend (period)
2. Conversions
3. CPA or ROAS (whichever is the primary optimization metric)
4. Clicks or Impression Share
5. One account-specific key metric (e.g. Conversion Value, CPL, IS Budget Lost)
6. One account-specific concern metric (e.g. Value/conversion, Zero-conv spend %)

Each card:

<div class="scard">
  <div class="lab">[Metric name]</div>
  <div class="val num">[Value]</div>
  <span class="delta d-good|d-bad|d-flat">▲/▼ [delta] vs prior</span>
  <span class="cap">[1-line context that explains the delta]</span>
</div>

---

4. Legend

Always include immediately before the sections. Never skip — the client needs to know what the chips mean.

Broken = active problem costing money now
Review = flag — verify or monitor before acting
Pass = working correctly — no action needed
HIGH = largest $ at stake · MED = real impact, lower urgency · LOW = housekeeping

---

5. Section anatomy (every section follows this pattern)

<div class="section" id="[slug]">
  <div class="section-head">
    <span class="section-num">N</span>
    <h2>[Section title — question form preferred: "Is the conversion signal trustworthy?"]</h2>
    <span class="chip c-broken|c-review|c-ok|c-na">[Verdict]</span>
    <span class="sev sev-h|sev-m|sev-l">[HIGH|MED|LOW]</span>
  </div>

  <p class="lead"><b>[The single most important finding in bold.]</b> [Supporting numbers in the same sentence.]</p>

  <!-- Data table -->
  <div class="tablewrap"><table>
    <thead>...</thead>
    <tbody>
      <!-- Each row: item name + evidence label + metrics + verdict chip + recommended action -->
    </tbody>
  </table></div>

  <!-- Callouts (use as many as needed, in this priority order) -->
  <div class="callout co-fix"><span class="ico">Fix</span><div class="ct"><b>[Action title]</b> [Where · What · Why format]</div></div>
  <div class="callout co-watch"><span class="ico">Watch</span>...</div>
  <div class="callout co-info"><span class="ico">Note</span>...</div>

  <!-- Bridge to next section (one line) -->
  <p class="sec-bridge">[Connect this finding to the next section.]</p>
</div>

Callout types:

Type	Class	When to use
Fix	co-fix	Immediate action — specific change with expected outcome
Watch	co-watch	Monitor — signal present but not enough volume to act yet
Note/Context	co-info	Explains a data limitation or provides useful context
Resolved	co-win	Correction of a prior version or confirmed fix

Table recommendation column format — always Where · What · Why:

• Where: the exact location in the Google Ads UI or the URL
• What: the specific change to make
• Why: the dollar or quality reason

Verdict chips:

• c-broken = Broken (active waste, misconfiguration, or broken tracking)
• c-review = Review (flag, monitor, opportunity — not urgent)
• c-ok = Pass (working correctly, no action needed)
• c-na = N/A or Limited Data (connector limitation, auth issue)

---

Sections — what each one covers

01 · Conversion tracking (always first)

Skill: measurement/conversion-tracking
Check: Are conversions down-funnel (policy approved, purchase, form fill) — not page views or app starts? Are values assigned? Are deprecated Universal Analytics goals still counting? Is offline/CRM import active?
Bridge: a noisy conversion signal corrupts bid targets → leads to 02.

02+ · Spend allocation & budget

Skill: campaigns/spend-allocation + campaigns/impression-share
Check: Which campaigns are capped (lost to budget)? Which are losing to rank? Who spends the most and converts the least? Per-campaign spend / conv / CPA / IS table. IS budget-vs-rank split.

· Bid strategy

Skill: campaigns/bid-strategy + campaigns/value-based-bidding
Check: Target vs actual (tCPA, tROAS). Broken if Maximize Conversions/Value with no tCPA/tROAS target AND rank-lost IS > 40% — the absent guardrail IS the misconfiguration. Is value-based bidding set where conversion values exist? Are $0-value conversions feeding value bidding?
Caveat: pull bid targets via the campaign.list connector action when the Step 0 pre-flight shows it works (the reliable route on accounts where Google Ads API access is configured); fall back to GAQL report.query if campaign.list 403s. A returned 0 on target fields = no guardrail (meaningful signal). Numeric target values from any route are corrupted (fan-out artifact) — trust only the binary "set vs not set".

· Quality Score

Skill: ads/metrics
Check: 3 pillars per keyword (Ad Relevance, Expected CTR, Landing Experience). Pull via GAQL keyword_view — these fields are absent from query_data/list_fields in most accounts (route dependency, not a field gap). Numeric QS via GAQL keyword_view IS reliable (1-10 per keyword). Numeric QS via query_data is NOT (aggregates to nonsensical values — never use). Focus on keywords with highest spend + below-average pillars.

· Search terms & negatives

Skills: search-terms/performance, search-terms/n-grams, search-terms/classifier/branded, search-terms/negatives

This section has three distinct blocks. Run all four skills in parallel and render them as subsections.

---

Block 0 — Negative keyword presence check

Goal: establish whether negative keywords exist at all before analyzing waste. This is a binary check that can produce the highest-severity finding in the section.

How: run search-terms/negatives (Layer 1 + Layer 2) first. Cross-reference campaign list vs campaigns with negatives to find zero-negative campaigns.

Verdict table columns: Campaign · Type · Negative count · Verdict chip · Severity

Finding	Verdict	Severity
Any Search campaign with 0 negatives	c-broken	HIGH
PMax with <50% of Search campaign negative count	c-review	MED
1–9 negatives in a Search campaign	c-review	MED
All campaigns have 10+ negatives	c-ok	—

Callout — zero negatives:

co-fix: "Add negative keywords to [Campaign name]"
Where: Google Ads UI → [Campaign] → Keywords → Negative keywords
What: Add intent-filter negatives first (free, cheap, DIY, jobs), then brand containment
Why: A campaign with zero negatives serves every query that matches any keyword — including irrelevant, low-intent, and competitor brand searches

Callout — PMax gap (if present):

co-info: "PMax campaign has [N] negatives vs [N] average in Search campaigns"
Why: PMax doesn't support ad-group-level negatives. Apply shared negative lists that cover the same intent filters used in Search.

---

Block 1 — Waste & negatives

Check: What % of non-brand spend drove zero conversions? Top individual waste terms + worst-value n-gram buckets. Competitor terms: conquest or accident?

Table columns: Search term · Campaign · Match type · Spend · Conversions · CPA · Verdict chip (Waste / Watch / Win)

Callout rules:

• co-fix if any single term > 2% of total spend with 0 conversions → "Add as negative exact in [Campaign]"
• co-fix if a competitor term appears with significant spend → "Conquest or accident? Verify intent — if accidental, add negative. If conquest, isolate in its own ad group with tailored copy."
• co-watch if a term has conversions but CPA > 2× account average → flag for bid adjustment before negative

---

Block 2 — Brand containment

Goal: detect own-brand search terms being captured by non-brand campaigns (the "contain_brand" leak). This is the structural problem Resolve audits flag as "Brand and Non-brand Separation."

Check using search-terms/classifier/branded:

1. Classify every search term as brand / competitor / generic.
2. Flag any brand-class term served by a campaign NOT in the brand campaign list (contain_brand = true).
3. Calculate total spend and conversions on those leaked brand terms.

Verdict logic:

• If contain_brand_spend > 5% of total non-brand campaign spend → c-broken / HIGH
• If contain_brand_spend 1–5% → c-review / MED
• If 0 contain_brand terms found → c-ok / PASS

Table columns: Search term · Served by (campaign) · Should be in (brand campaign) · Spend · Conversions · Match type of triggering keyword

Callout — always include if any contain_brand terms exist:

co-fix: "Add brand terms as negative keywords in generic campaigns"
Where: Each non-brand campaign listed in the contain_brand table
What: Add [exact match] negatives for each brand-class term found — use the exact term strings from the classifier output, not paraphrases
Why: Brand searches convert at lower cost when isolated; generic campaigns inflate CPL by competing with the brand campaign's own intent
Match type rule: always negative exact [brand term] — negative broad risks blocking legitimate generic variants

Structural recommendation (include when contain_brand_spend is HIGH):

co-fix: "Restructure into separate brand and non-brand campaigns if not already done"
Where: Google Ads campaign settings
What: Create a dedicated brand campaign (if none exists) with brand keywords on exact match. Add all brand terms as negative exact in every generic campaign.
Why: Without separation, Smart Bidding learns a blended CPA that undervalues brand intent and overweights generic CPL — the account can't optimize each correctly.

Bridge: brand containment fixes the intent routing; the next section checks whether the landing pages those terms reach are converting correctly.

· Match types

Skill: search-terms/match-types
Check: Is spend over-concentrated in one match type (usually Broad)? Is any type sitting untested at $0?

Always open with the concentration table — one row per type, never skip a type even if $0:

Match type	Spend	Share	Conversions	CPA	Verdict
Broad	…	…%	…	…	chip
Phrase	…	…%	…	…	chip
Exact	…	…%	…	…	chip

Blend guardrail (mandatory): if the account has more than one primary conversion action (e.g. Form Fill + Purchase, or Trial + MQL), mark every CPA value as directional and do NOT assert which type is "most efficient." Porter cannot split qualified actions by match type — the per-type CPA is a blend that can invert the true ranking. State this explicitly.

Verdict per type:

• c-broken — a type holds ≥70% of spend with no concentration benefit (CPA not clearly better), or a type is untested at $0 with obvious upside
• c-review — concentration exists but CPA advantage is visible; test advised before rebalancing
• c-ok — no concentration issue

Drill-down rule: only drill into individual keywords when a type is flagged c-broken or c-review. Show top 5 keywords by spend for the flagged type only.

Callout — concentrated:

co-fix: "Rebalance spend off [Match type]"
Where: Google Ads → Keywords tab → filter by match type
What: Shift 20–30% of [Match type] budget to [Exact/Phrase] — run as parallel ad groups first
Why: [N]% of spend on one match type increases query irrelevance risk; diversifying gives bid optimization more signal

Callout — untested type:

co-watch: "Pilot [Match type] on top ad groups"
Where: Google Ads → best-performing ad groups
What: Add 5–10 [Phrase/Exact] keywords mirroring existing Broad winners; run for 30 days
Why: [Match type] has $0 spend — no data on whether it performs better or worse than the current blend

Bridge: match type distribution explains part of the search term waste pattern — over-indexing on Broad generates more irrelevant queries.

· Landing page CRO

Skills: ads/landing-cro, ads/alignment
Check: Live scrape each unique landing page. 5 criteria: value prop / CTA / differentiation / proof / form friction. Flag message-match breaks (keyword searches for X, lands on page about Y).

· Audience & demographics

Skill: segmentation/audience/demographics-audit
Check: Age/gender efficiency vs account average. Undetermined bucket size (often 30%+ of spend). Which segments deserve bid up/down.

· Geography

Skill: segmentation/audience/geography
Check: Province/region CPA vs account average. Language/market mismatch (e.g. English ads in French Quebec). Best and worst geos.
Caveat: Porter Canada geo = no map atlas. Use bars, not geo bubble.

· Ad assets (RSA copy)

Skills: ads/copy, ads/assets
Check: Sitelinks / callouts / snippets / images present? RSA asset labels (PENDING = not enough data yet; note this honestly). Pinning abuse. Ad Strength.
Known gap: sitelinks and snippets often return blank from the Porter connector → flag for manual UI verification, never report as missing.

· Device & ad schedule

Skills: segmentation/audience/devices, segmentation/time/cyclical
Check: Desktop vs mobile CPA/ROAS gap. Day-of-week conversion pattern vs spend distribution. Weekend efficiency vs weekday.
Rule: don't recommend bid adjustments if < 30 conversions per device — low-confidence signal, say so explicitly.

· Campaign settings

Skill: campaigns/campaign-settings
Check: Location = Presence (not "presence or interest")? Search Partners OFF? Display Network OFF on Search campaigns? These are pass/fail checks.

· Demand Gen (only if DG campaigns exist)

Skill: campaigns/demand-gen
Check: DG campaign CPA/ROAS vs non-brand Search baseline. Is it earning its place?
Known gap: asset-group detail not available via Porter connector → note in section, direct to UI.

---

6. What's Set Up Right

Never skip this section. Always 4–6 cards of things that are genuinely working.
Use <div class="passgrid"> with <div class="passcard">.
Only include things that are actually confirmed by data — not generic praise.
Examples: "100% Search campaigns", "Offline CRM import active", "Location set to Presence", "Landing pages strong".

---

7. Action Plan

Ranked by $ at stake (not by section order). Each item:

<div class="item">
  <div class="rank"></div>  <!-- CSS auto-numbers these -->
  <div class="tc">
    <div class="tt">[Action title — imperative verb]</div>
    <div class="td">[2-3 sentences: what exactly to do, where, expected outcome]</div>
    <span class="tmeta">[N] · [HIGH|MED|LOW] · Est. [time] in [tool]</span>
  </div>
</div>

Maximum 7 items. If there are more, fold low-severity ones into a single "housekeeping" item.

---

8. Footer / Methodology

Always include:

• Data source (Porter Metrics Google Ads connector, date pulled)
• Period and comparison
• What was excluded and why (QS numeric, auction insights, asset groups if DG, any reauth failures)
• Conversion volume caveat if total conversions < 50 in the period

---

Known connector gaps — handle gracefully

Always acknowledge in the relevant section, never pretend data is missing:

Gap	How to handle
Auction Insights competitor overlap	Note as unavailable, direct to Google Ads UI → Auction Insights tab
Demand Gen asset-group detail	Note as unavailable, direct to UI → Demand Gen campaign → Asset groups
Sitelinks / snippets / images	Flag for UI verification, do NOT report as absent
Bid targets (tCPA / tROAS)	Pull via the campaign.list connector action; the Step 0 pre-flight tests it per account. It works where Google Ads API access is configured and 403s where it isn't (an access check, not "broken ~75%"). On a 403 fall back to GAQL report.query. A returned 0 or null on target fields = no guardrail set (meaningful signal — not a data error). Trust only the binary "target set vs not set" — numeric values from any route are corrupted by fan-out artifact.
Numeric QS via query_data	Never use — query_data aggregates values across ad groups, returns numbers like 16, 35, 46, 205 (outside the 1-10 scale). Use GAQL keyword_view instead: adGroupCriterion.qualityInfo.qualityScore returns correct 1-10 per keyword.
QS 3 categorical pillars	NOT available via query_data/list_fields in most accounts — this is a route dependency, not a field gap. Use GAQL keyword_view: adGroupCriterion.qualityInfo.searchPredictedCtr, .creativeQualityScore, .postClickQualityScore
Network settings booleans all False	When target_search_network, target_content_network, target_partner_search_network all return False for every campaign including active ones — this is a Porter rendering artifact, not real account config. Do NOT report as "correctly OFF". Add co-info callout directing to Google Ads UI → Campaign → Settings → Networks.
Geographic data (reauth)	If unavailable, infer from campaign structure + note the gap
Geographic view — city/region granularity	geographic_view returns country-level criterion IDs only (integers: 2840=US, 2124=Canada). Sub-national data (city, region, DMA) requires user_location_view, not geographic_view. Canada has no Porter atlas — use bar tables, not geo bubble.
Search terms — sort order	order_by is not supported in query_data — results return in default order, not sorted by spend. Sort results client-side after retrieval. Cannot guarantee "top by spend" ordering from query_data directly.
PMax search terms	Zero rows returned from search_term_view for PMax campaigns — confirmed Google API limitation, not a Porter gap. Direct client to Google Ads UI → Performance Max campaign → Insights tab → Search terms.
Age + gender in single query	Cannot be combined — universal Google API restriction. Run two separate queries: age_range_view and gender_view. PMax does not expose either view at all.
Landing page URL field	google_ads_final_url is not a valid field name. Use google_ads_unexpanded_final_url as the proxy field.

---

Design system & deploy

Do not write a stylesheet per account. Render from the template pages/main.html — it already carries the full audit layout and pulls every colour from the Porter Design System tokens (porter-design/dist/porter-tokens.css, from the public repo portermetricsample/porter-design). You fill the content; you never invent CSS. Pick one data-theme on the #audit wrapper (cream default · white · blue · purple). This is what makes every audit look identical run to run.

To understand the token mapping before editing the template's layout, read references/design.md (element-by-element: fonts, tokens, chips, tables, callouts, action plan).

The three non-negotiables (baked into the tokens — keep them if you ever touch the template):

• Bricolage Grotesque for KPI values and h2 headings. Hanken Grotesk for body and tables. IBM Plex Mono only for section numbers, chips, badges, and .num cells.
• No gradients, no drop shadows — 1px hairline borders only.
• The brand purple is the single accent — section numbers, eyebrows, verdict chip borders, "Do first" accent.

Step 4 — Deploy as a hosted Porter report

After the standalone HTML is written + verified, deploy it per references/deploy.md: make the file deploy-safe (wrap in <div id="audit" data-theme="…">, scope the style under #audit, strip the dev theme switcher) → scripts/to_porter_bundle.py (inlines the tokens + assembles the 3-file bundle) → get_report_template('blank') → create_report (config.charts:{}, /* porter:no-compare */, visibility:"PRIVATE"). Reply with the local file path and the report URL.

---

Relationship to other skills

Skill	Relationship
google-ads/dashboard/	Same analysis skills, different output (live charts vs narrative document)
account-audit/run-audit/	Predecessor — superseded by this skill for the HTML document format
google-ads/slides/	Future: takes this document's HIGH/MED findings and compresses into exec slides