What if everything your ad agency charges you $5,000 to $10,000 a month to audit could be done in your terminal, in minutes, for free? That’s the core premise of claude-ads — an open-source tool built by developer Agrici Daniel that uses Claude Code to run 250+ automated checks across seven major advertising platforms simultaneously.
The tool has already racked up 42,000 views on its demo video and 2,400+ GitHub stars. It’s one of those projects that, once you see it in action, is hard to unsee.
The Problem with Traditional Ad Audits
Let’s be honest about what most agency ad audits actually look like. Someone junior logs into your ad accounts, screenshots some graphs, pastes them into a deck with the agency’s logo, and calls it an audit. You get a PDF, maybe a call, and a bill for thousands of pounds.
The uncomfortable truth is that most human audits realistically cover 15 to 20 checks. They catch the obvious stuff — a campaign left on broad match, a forgotten ad group spending budget with no conversions — but the deeper structural issues? The creative fatigue patterns? The cross-platform audience overlap? Those require more time than any agency is billing for at that rate.
claude-ads runs 250+ weighted checks across Google Ads, Meta, YouTube, LinkedIn, TikTok, Microsoft Ads, and Apple Ads. In minutes. With no conflict of interest.
How the Audit Actually Works
When you run /ads audit in Claude Code, six parallel agents launch simultaneously, each dedicated to a specific domain. They don’t run sequentially — they run in parallel — which is why the whole thing takes minutes rather than weeks.
Here’s a breakdown of what each platform gets:
- Google Ads (80 checks) — Conversion tracking validation, Enhanced Conversions, Consent Mode V2 compliance, search term waste analysis, negative keyword gap detection, Quality Score optimisation, Smart Bidding strategy audit (flagging deprecated ECPC), Performance Max asset density, AI Max for Search evaluation, and CTV measurement gaps.
- Meta Ads (50 checks) — Pixel and Conversions API health with EMQ scoring, Andromeda creative diversity analysis (a Similarity Score above 60% suppresses your ad retrieval), creative fatigue detection, audience overlap, Advantage+ Sales structure, and placement performance trends.
- YouTube Ads (15 checks) — Format evaluation across Skippable, Non-Skippable, Bumper, Shorts, and Demand Gen. Hook quality in the first five seconds, ABCD creative framework compliance, CTV measurement strategy (75% of YouTube ad spend now lands on CTV), and frequency management.
- LinkedIn Ads (27 checks) — ABM targeting efficiency, Thought Leader Ads evaluation (CPC of £2–4 versus £13+ for standard), lead gen form completion rates, bidding strategy review, CRM attribution, and EU compliance checks.
- TikTok Ads (28 checks) — Spark Ads versus standard creative performance, hook rate analysis, Smart+ modular controls, GMV Max requirements for TikTok Shop, Search Ads toggle evaluation, and Events API Gateway setup.
- Microsoft Ads (24 checks) — Google Ads import validation, Copilot placement evaluation, LinkedIn Profile Targeting performance, CTV and vertical video ad checks, and Consent Mode compliance for EEA/UK.
- Apple Ads (35+ checks) — Campaign structure across BOFU/MOFU/Search Match, bid health, Custom Product Pages, Maximize Conversions bidding, AdAttributionKit dual attribution, and TAP placement coverage.
That’s not a checklist a human is going to get through in an afternoon. Or even a week.
The Scoring System
Every check is weighted, not just counted. The scoring formula applies severity multipliers — Critical issues score 5.0x, High 3.0x, Medium 1.5x, Low 0.5x — against category weights to produce a 0–100 health score with an A through F grade.
The category weights reflect where real money is at risk: Conversion Tracking carries 25%, Wasted Spend 20%, Structure and Keywords 15% each, Ads quality 15%, and Settings 10%. A single broken conversion tracking setup will tank your overall grade regardless of how clean everything else looks. That’s intentional — because broken tracking is the one thing that makes all other optimisation pointless.
What you get out the other end isn’t a slide deck. It’s a prioritised list of issues ranked by estimated revenue impact, each with the specific problem described, a severity rating, exact fix instructions, and estimated monthly savings. Quick Wins are flagged separately for fixes under 15 minutes.
On a £40,000/month ad budget, the critical findings alone — conversion gaps, budget allocation errors, audience overlap — typically identify 15–30% in wasted spend. That’s £6,000–£12,000 found in a single command.
What’s New in v1.5
The latest release pushed the check count from 186 to 250+, added Apple Ads as the seventh platform, and introduced three new skills that go beyond auditing:
- /ads math — A PPC financial calculator covering CPA, ROAS, break-even analysis, impression share, budget forecasting, LTV:CAC ratios, and MER. Works offline with pasted data from exports, no live API connection required.
- /ads test — An A/B test designer using an IF/THEN/BECAUSE hypothesis framework, with statistical significance calculations, sample size tables, and platform-specific testing guides for Meta, Google, LinkedIn, and TikTok.
- /ads report — A professional PDF generator that produces a formatted report with health score gauge charts, platform comparison bars, and pass/fail distributions. Something you could hand to a client.
v1.5 also brought SSRF protection and security hardening — important for a tool that’s going to be touching your ad account data, even locally.
The Architecture Behind It
claude-ads is built on three layers. The directive layer is the orchestrator — the SKILL.md file that handles routing and quality gates. Below that is the orchestration layer with 19 sub-skills routing to the right analysis. At the bottom sits the execution layer: 10 agents, 25 reference files, and 12 industry templates, all loading on-demand so there’s no bloat.
The parallel architecture is what makes the speed possible. Sequential analysis of seven platforms would take considerably longer. Running six agents simultaneously compresses that to minutes.
Installation
Two options:
Via the plugin marketplace (recommended):
/plugin marketplace add AgriciDaniel/claude-ads
/plugin install claude-ads@agricidaniel-claude-adsOne-liner install:
curl -fsSL https://raw.githubusercontent.com/AgriciDaniel/claude-ads/main/install.sh | bashAfter that, /ads audit runs the full multi-platform analysis. You can also run single-platform deep dives with /ads google, /ads meta, and so on. The tool works with data you provide — exports, screenshots, pasted metrics. For live API access, it pairs with MCP servers like Google Ads MCP, Adspirer for Meta, and GrowthSpree for LinkedIn.
The whole thing is MIT licensed. No usage limits. No premium tier. No ad account data leaves your machine — all analysis runs locally.
What This Means for the Industry
There’s a version of this story where AI tools are replacing everything. That’s not quite what’s happening here. The strategic layer of paid media — creative direction, understanding unit economics, knowing which channels to even be on — that still requires human judgement. What claude-ads automates is the audit part: the systematic checking of 250+ things that should or shouldn’t be true about your account setup.
The agencies that will thrive are the ones that use tools like this to run deeper audits faster, then spend their human time on the strategy and creative work that actually requires a brain. The ones that don’t adapt will keep charging for a process that’s now automatable.
For in-house marketers and freelancers, the value proposition is even clearer. You now have access to an audit depth that previously required either a significant agency retainer or an enormous amount of manual time. Neither is a realistic option for most teams.
The tool is on GitHub at AgriciDaniel/claude-ads with 2,400+ stars and climbing. Worth running against your accounts before your next agency review.
Leave a Reply
You must be logged in to post a comment.