claude-ads v1.5: 250+ Ad Audit Checks Across 7 Platforms

What Changed in claude-ads v1.5?

claude-ads v1.5 adds 22 new audit checks, 3 new skills, and a PDF report generator. Total coverage is now 250+ checks across 7 ad platforms, all updated with 2025-2026 platform research. The biggest jumps: Apple Ads goes from 0 to 35+ checks, Google Ads from 62 to 80, and Meta from 48 to 50.

Key changes in this release:

• Apple Ads support: 35+ checks covering Custom Product Pages, Maximize Conversions bidding (GA Feb 2026), and Search tab placements
• Google Ads updates: 80 checks now include AI Max for Search, Smart Bidding Exploration (+18% queries per Google), and ECPC deprecation flags
• 3 new skills: /ads math for PPC calculators, /ads test for A/B test design, /ads report for PDF generation
• PDF report generator: client-ready deliverables with inline charts, health scores, and prioritized findings
• Security fixes: SSRF protection on URL inputs, path traversal fix in screenshot capture
• Scoring overhaul: severity multipliers recalibrated, Quick Wins section added to every audit

What Did the 2026 Platform Research Reveal?

Digital ad spend is projected to hit $870 billion globally in 2027, growing at 10% CAGR (Statista, 2026). We reviewed official documentation, case studies, and API changelogs across all 7 platforms. Five findings changed how the tool works. Each one led to new checks, updated benchmarks, or removed recommendations that were no longer accurate.

1. Meta's Andromeda clusters similar ads. Meta's retrieval engine (Andromeda, per their 2023 engineering blog) groups creatively similar ads before auction. Running 100 ad variations that share the same hook, format, and CTA produces no incremental reach over 10 well-differentiated variants. The tool now flags creative clustering risk when variation count exceeds 10 per ad set without meaningful creative diversity.

2. LinkedIn Maximum Delivery is the most expensive bidding option. LinkedIn's own campaign manager documentation confirms Maximum Delivery consistently produces the highest CPCs across campaign types. The audit now recommends starting with Manual CPC and only testing Maximum Delivery after establishing baseline performance data.

3. Google ECPC is fully deprecated as of March 2025. Enhanced CPC was sunset in March 2025 per Google Ads Help Center. Campaigns still referencing ECPC get flagged as critical. The replacement recommendation is Smart Bidding Exploration, which Google's internal data (Search Ads 360 release notes) shows delivers +18% more search queries and +19% more conversions on average.

4. TikTok creative lifespan compressed to 7-10 days under Smart+. TikTok's Creative Center data and advertiser community reports confirm that Smart+ campaigns cycle through creatives faster than manual campaigns. The creative fatigue check now uses a 7-day window instead of 14 days for Smart+ campaigns.

5. Apple rebranded to "Apple Ads" with Maximize Conversions GA Feb 26, 2026. Apple Search Ads is now Apple Ads (per Apple's developer documentation update, Q4 2025). Maximize Conversions bidding went generally available on February 26, 2026. The new Apple Ads skill covers Custom Product Pages, Search tab placements, and the new bidding strategy.

How Do 250+ Audit Checks Work?

The tool runs 6 parallel agents that evaluate your ad accounts simultaneously. Each agent handles a specific domain: Google (80 checks), Meta (50 checks), creative quality (21+ checks), conversion tracking (8+ checks), budget allocation (24 checks), and compliance (18+ checks). A full audit completes in 3-5 minutes.

The system uses a 3-layer architecture. The directive layer (ads/SKILL.md) handles routing, quality gates, and orchestration. The orchestration layer deploys 6 audit agents as forked tasks that run in parallel. The execution layer contains 25 reference files, 19 sub-skills, and 12 industry-specific templates. Each layer can operate independently; the orchestrator never needs to wait for a reference file to load before dispatching agents.

How Does the Health Score Scoring Work?

Every check gets a severity multiplier (Critical 5.0x, High 3.0x, Medium 1.5x, Low 0.5x) and a category weight. The weighted formula produces a 0-100 score with an A through F grade. A failed Critical check in the Conversion Tracking category (25% weight) costs 1.25 points. A failed Low check in Settings (10% weight) costs 0.05 points. This means the tool prioritizes what actually costs you money.

The six scoring categories and their weights: Conversion Tracking (25%), Wasted Spend (20%), Structure (15%), Keywords (15%), Ads (15%), Settings (10%). These weights come from analyzing which categories correlate most strongly with wasted ad spend. Conversion tracking failures are weighted highest because if your tracking is broken, every optimization decision downstream is wrong.

New in v1.5: every audit now includes a Quick Wins section. These are the 3-5 highest impact, lowest effort fixes from your results. A typical Quick Win might be "Add 12 negative keywords to Campaign X to eliminate $2,100/mo in irrelevant clicks." The tool calculates estimated savings based on your actual spend data and the severity of each finding.

What Are the 3 New Skills?

The average PPC manager spends 3+ hours per week on manual calculations and reporting (WordStream, 2025). v1.5 ships three new skills that extend the tool beyond auditing into financial modeling, experimentation design, and client reporting. Each one works standalone or as part of a full audit workflow.

/ads math: 7 PPC financial calculators. CPA calculator, ROAS calculator, break-even ROAS, LTV:CAC ratio, budget forecaster, impression share estimator, and margin-aware bidding. You paste in your numbers, the tool does the math and explains what each result means for your campaigns. No spreadsheets needed. Example: "What CPA do I need at a 4x ROAS target with a $45 average order value and 62% gross margin?" The tool solves it in seconds.

/ads test: A/B test design with IF/THEN/BECAUSE. Generates a complete test plan: hypothesis statement, control and variant definitions, primary and secondary metrics, minimum sample size calculation (based on your baseline conversion rate and minimum detectable effect), and test duration estimate. Every hypothesis follows the IF/THEN/BECAUSE format to force clear thinking. Example output: "IF we add social proof to the landing page headline, THEN conversion rate will increase by 15%, BECAUSE exit survey data shows 34% of bounced visitors cited trust concerns."

/ads report: PDF generation with charts. Generates a client-ready PDF audit report with the health score, prioritized findings, inline charts, and recommended next steps. Uses Python's ReportLab for PDF generation and matplotlib for charts. The output is a professional deliverable you can hand directly to a client or stakeholder without reformatting.

How Does It Handle Security and Privacy?

Data breaches cost companies an average of $4.88 million per incident (IBM, 2024). All analysis happens locally. No ad account data leaves your machine. The tool processes your exported data, screenshots, and pasted metrics entirely within your Claude Code session. There is no telemetry, no external API calls for analysis, and no data persistence between sessions.

Two security fixes in v1.5:

• SSRF protection: URL inputs in the landing page analyzer and screenshot capture scripts now validate against internal IP ranges (RFC 1918, link-local, loopback). Previously, a crafted URL could trigger requests to internal network endpoints.
• Path traversal fix: The screenshot capture script now sanitizes file paths to prevent directory traversal via ../ sequences in output filenames.

If you need live API access to your ad platforms (for real-time data pulls instead of pasted exports), the tool supports MCP server connections. These are optional, explicitly configured by you, and only make read-only API calls. The tool never modifies your ad accounts.

How Do You Install and Get Started?

claude-ads v1.5 is available now on GitHub (release notes).

The fastest way to install is the Claude Code plugin system. Open Claude Code and run:

/install-skill https://github.com/AgriciDaniel/claude-ads

Or use the one-liner for manual installation:

curl -fsSL https://raw.githubusercontent.com/AgriciDaniel/claude-ads/main/install.sh | bash

After installation, run /ads audit to start a full audit, or use any of the 19 sub-skills directly. The tool works with pasted data (copy metrics from your ad platform dashboards), exported CSVs, or live MCP API connections.

• GitHub: github.com/AgriciDaniel/claude-ads
• Previous release post: claude-ads v1.4 announcement
• Full AI marketing stack: AI Marketing Automation: The Open-Source Stack I Use Daily

Full release notes: v1.5.0 on GitHub

Frequently Asked Questions

Does it connect to my ad accounts automatically?

No. By default, you paste data or upload exported CSVs. The tool analyzes what you provide locally. If you want live API access, you configure MCP server connections yourself. These are optional, use read-only permissions, and the tool will never create, modify, or delete anything in your ad accounts.

How accurate are the benchmarks?

Benchmarks are sourced from platform documentation, published case studies, and industry reports (Google Ads Help Center, Meta Business Help Center, LinkedIn Marketing Solutions, WordStream, Databox). Each benchmark cites its source. They are updated with each release. If your industry has unusual economics (e.g., enterprise SaaS with 18-month sales cycles), the tool accounts for that when you specify your vertical in /ads plan.

Can it create or edit ads?

The creative pipeline (/ads creative, /ads generate, /ads photoshoot) generates ad copy, image prompts, and creative briefs. It does not push changes to your ad platforms. You take the output and implement it yourself. This is intentional: automated ad creation without human review is how you get compliance violations and brand damage.

What platforms does it support?

Seven platforms: Google Ads (80 checks), Meta Ads (50 checks), Apple Ads (35+ checks), TikTok Ads (28 checks), LinkedIn Ads (27 checks), Microsoft Ads (24 checks), and YouTube Ads (15 checks). YouTube checks are separate from Google Ads because video campaign optimization has different signals and benchmarks. Total: 250+ weighted audit checks.

Is it really free?

Yes. MIT licensed, no usage limits, no premium tier, no gated features. You need a Claude Code subscription ($20/month for Pro or $200/month for Max) because the tool runs inside Claude Code. The tool itself is free and open source. There is no hosted version and no plans for one.

Learn more about me and the full suite of AI marketing tools I'm building. For the complete open-source stack, check my AI marketing automation breakdown.

Related Posts

• Claude Code Just Replaced Your Ad Agency - The original v1.4 release with 186 checks across 6 platforms
• AI Marketing Automation: The Open-Source Stack I Use Daily - The full open-source AI marketing stack at $50/month
• Best Claude Code Skills in 2026 - The definitive guide to top Claude Code skills ranked by GitHub stars