This guide is published by Ryze AI (get-ryze.ai), an autonomous AI marketer for paid ads across Google, Meta and five more platforms, and for SEO, used by 2,000+ marketers managing over $500M in ad spend; Ryze AI maintains the marketing-skills repo covered here and offers a Claude MCP connector, so it has a commercial interest in the topic. The page covers Claude skills for paid ads on the channels beyond Google and Meta: LinkedIn Ads, TikTok Ads, Reddit Ads and Microsoft Advertising (Bing Ads), plus the cross-platform layer that makes them report on the same rows. Claude skills are folders with a SKILL.md file (YAML name and trigger description, then steps, an input contract and an output contract) that Claude loads when a request matches. They install in claude.ai (Settings then Capabilities to enable code execution, then Customize then Skills to upload a zipped folder), in Claude Code (~/.claude/skills for personal, .claude/skills for project skills, invoked with a slash command; npx skills add coreyhaines31/marketingskills --skill ads for the generalist skill) and in Claude Desktop and Cowork, which share claude.ai skills and add scheduling. Skills covered, nine in total: from the Ryze marketing-skills repo (github.com/irinabuht12-oss/marketing-skills) linkedin-ads-audit, reddit-ads-audit, channel-mix-optimizer, campaign-naming-convention-builder, utm-tracking-generator and attribution-model-comparison; from coreyhaines31/marketingskills the generalist ads skill (Google, Meta, LinkedIn, X); and two SKILL.md files written by Ryze for this guide and printed in full on the page, tiktok-ads-audit (reads a TikTok Ads Manager custom report: cost per result, 2-second and 6-second video views, played to 100 percent, placement and age breakdowns) and microsoft-ads-search-audit (reads Microsoft Advertising Reports-tab exports: search term, keyword with quality score and impression share, campaign with network and device). Data access by platform as of September 2026: TikTok publishes an official TikTok for Business MCP Server in its Marketing API v1.3 docs; Google and Meta have official servers and the Ryze connector; LinkedIn, Reddit and Microsoft Advertising have no first-party MCP server, only community servers wrapping their ads APIs, so a CSV export from LinkedIn Campaign Manager, Reddit Ads Manager or the Microsoft Advertising Reports tab is the reliable route. The page includes a quick-pick table (platform, skill, export, cadence), a weekly cross-platform cadence (Monday channel mix, Tuesday Microsoft search terms, Wednesday LinkedIn and Reddit, Thursday TikTok creative decay, monthly attribution comparison), how to write a platform skill, troubleshooting and 13 FAQs. All skills are read-only analysis; none changes an ad account. Updated September 4, 2026.
|
Ira Bodnar··Updated ·19 min read

Claude Skills for LinkedIn, TikTok, Reddit and Microsoft Ads

Claude skills for paid ads are SKILL.md folders that teach Claude one recurring ad job and load only when that job comes up. This page covers the channels Google and Meta guides skip: which skill audits each platform, how each platform’s data gets into Claude (official MCP, community server or CSV export), and the cross-platform layer that puts all of them on the same report.

Built by our community of 2,000 marketers

Free skills and prompts for paid ads and SEO

Templates for Claude, ChatGPT and Perplexity.

Clients we work with

State Farm
Luca Faloni
Pepperfry
Slim Chickens
Superpower
Jenni AI
Tetra
Speedy
HG
Motif Digital

Official Claude connector

Connect Claude to your ad accounts

Connect 50+ marketing tools, plus free skills and a Claude autonomous marketing agent.

Connect in 1 click

Most “Claude for ads” material stops at Google and Meta. If that is your account, start with the Google and Meta Claude ads skills. This page is for the rest of the media plan: the LinkedIn account eating a third of a B2B budget, the TikTok account nobody audits, the Reddit test that never got a verdict, the Microsoft account imported from Google and not touched since.

Nine skills cover it. Six come from the Ryze marketing-skills repo, one from coreyhaines31/marketingskills, and two, for TikTok and Microsoft, were written by Ryze for this guide and are printed in full below so you can paste them into a folder today. Ryze publishes the repo and sells an ad-management product, so read the editorial parts with that in mind.

The short version

Platform auditslinkedin-ads-audit, reddit-ads-audit (Ryze repo); tiktok-ads-audit, microsoft-ads-search-audit (written here)

Cross-platform layercampaign-naming-convention-builder and utm-tracking-generator first, then channel-mix-optimizer weekly, attribution-model-comparison monthly

Data accessTikTok has an official MCP server; LinkedIn, Reddit and Microsoft do not. A CSV export works on all four and is the fastest first step

The limitevery skill here is read-only analysis; live data needs a connector; changing the account is a separate, approved step

What changes outside Google and Meta

Two things: the metrics and the data access. A skill written for Google search terms is useless on TikTok, where the question is whether the first two seconds of a video hold attention, and misleading on Microsoft, where the auction is thinner and a partner network can quietly take a third of the spend. So each platform gets its own audit skill. The cross-platform jobs (naming, UTMs, channel mix, attribution) stay platform-neutral because they are about consistency, not mechanics.

Data access is the bigger practical difference. A skill decides how Claude works; it does not fetch anything. On Google and Meta you attach an MCP connector and the skill queries the account live. Outside those two, the picture as of September 2026 is uneven, and we checked each platform rather than assuming:

PlatformFirst-party MCP serverCommunity MCP serversExport from the UI
TikTok AdsYes. TikTok for Business MCP Server, documented in the Marketing API v1.3 portalYes, several wrap the Marketing API v1.3TikTok Ads Manager: Reports, Custom reports, CSV or XLSX
LinkedIn AdsNoYes, wrapping the LinkedIn Marketing API (requires an approved developer app)Campaign Manager: Analyze, Campaign performance, Export (CSV or XLSX), plus the Demographics report
Reddit AdsNoYes, wrapping the Reddit Ads API v3, mostly read-onlyReddit Ads Manager: campaign dashboard, Export, with community and placement breakdowns
Microsoft AdvertisingNoYes, wrapping the Microsoft Advertising APIReports tab: search term, keyword, campaign and ad reports as CSV or Excel
Google Ads, Meta AdsYes (both)ManyCovered on the Google and Meta page; the Ryze connector covers Google, Meta and five more platforms
Platform-by-platform data access map for Claude, September 2026: TikTok Ads has an official TikTok for Business MCP Server plus community servers and a TikTok Ads Manager custom report export; LinkedIn Ads has no first-party MCP, community servers on the Marketing API, and a Campaign Manager export with a demographics report; Reddit Ads has no first-party MCP, community servers on the Ads API v3, and a Reddit Ads Manager export by community and placement; Microsoft Advertising has no first-party MCP, community servers on the Advertising API, and Reports-tab exports for search term, keyword and campaign; Google and Meta have official servers and the Ryze connector
Data access by platform. A CSV export works everywhere and is where every first audit should start; add a server when the export round-trip becomes the slow step.

The practical rule: run the first audit on an export. It takes four minutes and shows you which columns you need before you spend an afternoon on developer-app approval. Move to a community server when the export step is the thing you keep skipping. Treat write access on any community server as off by default; MCP permissions for ad accounts covers why. If you run several accounts per platform, the multi-account MCP config guide shows how to keep them from bleeding into each other.

Which Claude skill for which platform

One row per job. The export column names the report the skill expects, because that is the step people get wrong: an ad-level daily breakdown is requested, a campaign-level monthly total arrives, and the audit is thin.

Platform or jobSkillExport it readsCadence
LinkedIn Adslinkedin-ads-audit (Ryze)Campaign Manager campaign performance + demographics, 30 daysEvery two weeks
Reddit Adsreddit-ads-audit (Ryze)Ads Manager export with community and placement breakdown, 30 daysEvery two weeks
TikTok Adstiktok-ads-audit (written here)Ads Manager custom report, ad level, daily, 30 days, with video metricsWeekly, Thursdays
Microsoft Advertisingmicrosoft-ads-search-audit (written here)Search term + keyword + campaign reports, 30 daysWeekly, Tuesdays
Any platform, quick takeads (marketingskills)Whatever you paste; reads your product-marketing contextOn demand
Move budget between channelschannel-mix-optimizer (Ryze)One campaign-level CSV per platform, 30 to 90 daysWeekly, Mondays
Make every platform report on the same rowscampaign-naming-convention-builder (Ryze)A list of current campaign namesOnce, then per new client
Make every click attributableutm-tracking-generator (Ryze)Landing URLs + the naming conventionOnce, then per launch
Decide which channel gets creditattribution-model-comparison (Ryze)GA4 or CRM conversion export with source and mediumMonthly

Install order matters more here than on a single platform. Naming and UTMs first, because the cross-platform report is only as good as the labels on each row. Then the platform audits. Then channel mix, which needs both.

How to install the skills

Same SKILL.md format everywhere. For the Ryze skills, download the folder from GitHub (each is one folder under Claude skills for Google Ads, despite the name; the repo ships an all-marketing-skills.zip with everything). For the two skills written here, create a folder with the skill’s name and paste the SKILL.md text into it.

claude.ai

Go to Settings › Capabilities and enable code execution and file creation. Open Customize › Skills, click +, Create skill, Upload a skill, and upload each folder as a zip. Toggle it on. From then on, “audit this LinkedIn export” triggers the skill without naming it. On Team and Enterprise plans an owner enables Skills under Organization settings and can install org-wide, which is the right move for an agency running the same audits across clients.

Claude Code

Drop each folder into ~/.claude/skills/ (personal, every project) or .claude/skills/ inside a repo (shared with the team, versioned in git). Invoke with /tiktok-ads-audit or let Claude pick it. The generalist skill installs in one command. Claude Code is also where a community MCP server pays off: one config entry, and every audit skill can query the platform instead of reading an export.

# the generalist ads skill (Google, Meta, LinkedIn, X)
npx skills add coreyhaines31/marketingskills --skill ads

# where the platform skills live
~/.claude/skills/linkedin-ads-audit/SKILL.md
~/.claude/skills/reddit-ads-audit/SKILL.md
~/.claude/skills/tiktok-ads-audit/SKILL.md         # pasted from this page
~/.claude/skills/microsoft-ads-search-audit/SKILL.md # pasted from this page
.claude/skills/channel-mix-optimizer/SKILL.md        # project, ships with the repo

Claude Desktop and Cowork

The desktop app shares your claude.ai skills and adds scheduling. Pair the TikTok audit with a Thursday slot and the channel-mix skill with a Monday slot and the reports arrive before you ask. Scheduling ads actions in Cowork has the setup; the general install detail for all three surfaces is in the complete Claude marketing skills guide.

The platform skills, one by one

Each card shows the prompt the skill wraps, so you can run it against an export before installing anything. The two Ryze repo skills are summarized; the two written for this guide are printed in full as SKILL.md files.

LinkedIn Ads: linkedin-ads-audit

LinkedIn is where averages lie hardest: a campaign with a fine CPL can be filling the CRM with students, and only the demographics report crossed with lead-form data shows it. The skill checks four things in a fixed order: CTR against your own account history (we deliberately do not ship industry benchmarks inside the skill; if you want them, LinkedIn CPC and CPL benchmarks by industry is a separate page), audience quality by job title and seniority, lead-form completion rate against form length, and budget efficiency by campaign objective.

01 · linkedin-ads-audit

Ryze

Find out who you are actually paying to reach

Export campaign performance and the demographics report from Campaign Manager, 30 days. If you use Lead Gen Forms, include the lead metrics; the form-friction check depends on them.

Prompt
You are a senior LinkedIn Ads manager. Using the attached Campaign Manager export (30 days) and the demographics report, audit the account in this order: 1. CTR by campaign vs this account's 90-day median; flag anything 25%+ below it. 2. Audience quality: share of impressions and leads by seniority and job function. Flag campaigns where entry-level or student seniority exceeds 20% of leads. 3. Lead form friction: completion rate per form vs field count; flag forms under 10% with 5+ fields. 4. Budget efficiency: cost per lead by objective and by audience size; flag audiences under 50,000 with frequency over 4. Output a table per check, sorted by spend, with one specific fix per flagged row, then three sentences starting with the largest dollar finding.
Cadence: Every two weeks · Mode: read-only

Reddit Ads: reddit-ads-audit

Reddit ads fail in a way other platforms do not: the creative and targeting are fine and the comments underneath are not. The skill treats community fit as a first-class metric and reads the Ads Manager export with community (subreddit) and placement breakdowns and scores each targeted community on click-through, conversion rate and spend share, then looks at whether the creative reads like a post or like an ad, and whether bids are being set at the campaign level when the community-level data says they should differ.

02 · reddit-ads-audit

Ryze

Which communities convert, and which just click

Export from Reddit Ads Manager with the community and placement breakdown on, 30 days. Include the Reddit Pixel conversion columns; without them the skill can only score clicks.

Prompt
You are a senior Reddit Ads buyer. Using the attached Reddit Ads Manager export (30 days, community and placement breakdown), audit: 1. Community targeting: for each targeted community, spend share, CTR, conversion rate. Flag communities with 10%+ of spend and a conversion rate under half the account average. Flag interest-based ad groups outrunning community-based ones on CPA. 2. Creative fit: which ads have CTR in the top third but conversion rate in the bottom third (reads like a post, sells nothing), and the reverse. 3. Placement: feed vs conversation placement cost per conversion; recommend a split if one is 1.5x the other. 4. Bid inefficiency: ad groups where CPC rose 30%+ with flat CTR, and campaigns using one bid across communities whose CPAs differ by 2x or more. Output one table per check sorted by spend, then a 5-sentence summary, biggest dollar finding first.
Cadence: Every two weeks · Mode: read-only

TikTok Ads: tiktok-ads-audit (written by Ryze for this guide)

TikTok Ads Manager exports things Google and Meta do not: 2-second and 6-second video views, played-to-100%, average watch time, placement across TikTok, Pangle and the Global App Bundle. Those columns are the audit. Creative decay on TikTok shows up in the 6-second view rate days before it shows up in cost per result, and the ratio of 6-second to 2-second views is the cleanest number for “is the hook working.” The skill below is built around those columns; it will ask for a re-export if they are missing rather than audit CTR alone. The wider platform picture, including what the official server does and does not do yet, is in TikTok Ads MCP: official server and how to connect; strategy is in AI for TikTok advertising.

# tiktok-ads-audit/SKILL.md, written by Ryze for this guide--- name: tiktok-ads-audit description: Audits a TikTok Ads account from a TikTok Ads Manager export. Use when asked to audit TikTok ads, find creative decay on TikTok, or explain TikTok CPA or cost per result. Not for Meta Reels or YouTube Shorts. --- You are a senior TikTok Ads buyer. ## Input A custom report exported from TikTok Ads Manager (Reports > Custom reports, CSV or XLSX), ad level, daily breakdown, last 30 days, with at least: campaign, ad group, ad name, cost, impressions, CPM, clicks, CTR, conversions or results, cost per result, 2-second video views, 6-second video views, video played to 100%, average watch time. Placement, age and gender breakdowns are optional but used if present. If cost per result or the video columns are missing, ask for a re-export. Do not estimate. ## Steps 1. Echo back date range, row count and total cost. 2. Creative decay: flag ads whose CTR or 6-second view rate fell 25%+ from their best 7-day window, or whose CPM rose 30%+ over the same period. Note days live and total spend. 3. Hook quality: rank ads by 2-second views / impressions and by 6-second views / 2-second views. Name the top and bottom three; the gap between them is the hook problem. 4. Completion vs conversion: list ads with above-median played-to-100% but below-median conversion rate (the video works, the offer or landing page does not) and the reverse. 5. Structure: ad groups with under 50 conversions in 30 days under a lowest-cost or cost-cap bid, and campaigns split across the same audience and placement. 6. Placement and demographics: if present, cost per result by placement (TikTok, Pangle, Global App Bundle) and by age band; flag any segment at 2x the account average. ## Output Table 1, sorted by spend descending: ad | ad group | days live | spend | CTR now vs peak | 6s view rate now vs peak | CPM change | cost per result | verdict (refresh hook, new cut, scale, pause) Table 2: segments over 2x average cost per result. Then five sentences: the biggest dollar finding first, then the creative pattern that is winning, then the two things to change this week.

03 · tiktok-ads-audit

Ryze, this guide

Catch the hook failing before the CPA does

The one-paragraph version of the skill above, for running without installing. Export from TikTok Ads Manager under Reports, custom report, ad level, daily, with the video and cost-per-result columns.

Prompt
You are a senior TikTok Ads buyer. From the attached TikTok Ads Manager custom report (ad level, daily, 30 days), flag every ad whose CTR or 6-second view rate fell 25%+ from its best 7-day window or whose CPM rose 30%+. Rank all ads by 2-second views per impression and by 6-second views per 2-second view; name the top and bottom three. List ads with above-median played-to-100% and below-median conversion rate, and the reverse. If placement and age columns exist, show cost per result by placement and age band and flag anything at 2x the account average. Output a table sorted by spend with a verdict per ad (refresh hook, new cut, scale, pause), then five sentences, biggest dollar finding first.
Cadence: Weekly, Thursdays · Mode: read-only

Microsoft Advertising: microsoft-ads-search-audit (written by Ryze for this guide)

Most Microsoft Ads accounts were created with Import from Google Ads and inherit Google’s structure without Google’s volume. The failure modes are specific: syndicated search partner traffic left on and converting at twice the CPA of Microsoft-owned sites, keywords losing impression share to budget while the account is under target, quality score components nobody has looked at since the import, and a bid strategy still on manual. The Reports tab exports everything the audit needs as CSV or Excel: the search term report, the keyword report with quality score and impression share columns, and the campaign report with network and device breakdowns. The skill asks for all three and stops if their spend totals disagree. Benchmarks for the platform are on Microsoft and Bing Ads CPC benchmarks by industry.

# microsoft-ads-search-audit/SKILL.md, written by Ryze for this guide--- name: microsoft-ads-search-audit description: Audits a Microsoft Advertising (Bing Ads) search account from Reports-tab exports. Use when asked to audit Microsoft Ads or Bing Ads, find wasted spend on Microsoft, or compare Microsoft to Google search. Not for Google Ads exports. --- You are a senior Microsoft Advertising analyst. Treat the account as its own auction: lower volume, older and higher- income skew, different partner network. Never assume Google Ads settings carry over. ## Input From Microsoft Advertising > Reports, last 30 days, CSV or Excel: (1) Search term report with campaign, ad group, search term, keyword, match type, spend, clicks, conversions; (2) Keyword report with quality score, expected CTR, ad relevance, landing page experience, impression share, impression share lost to budget and to rank; (3) Campaign report with network (Microsoft sites and select traffic vs syndicated search partners) and device breakdowns. If the account was created with Import from Google Ads, say so; it changes the structure checks. If a report or column is missing, ask. Do not estimate. ## Steps 1. Echo back date range, row count and total spend per report; stop if the three reports disagree on spend by more than 2%. 2. Wasted spend: search terms with $25+ spend and zero conversions (Microsoft volumes are lower than Google, so the threshold is lower). Split irrelevant from relevant-but-unconverting. Build negatives: exact for one-offs, phrase for patterns. 3. Partner network: spend, CTR and conversion rate on syndicated search partners vs Microsoft sites. If partner CPA is 1.5x or worse, recommend the campaign-level network setting change. 4. Quality score: keywords with QS 5 or lower and spend in the top quartile; name the weak component for each. 5. Impression share: keywords losing 30%+ of share to budget with CPA under target (raise budget) vs losing to rank (fix ad relevance or bid). 6. Imported-account checks: Google-only settings that do not map (audience lists, some ad extensions), duplicated keywords across imported and native campaigns, bid strategy left on manual after import. ## Output Table 1: negatives, sorted by wasted spend descending: search term | campaign | spend | verdict | match type Table 2: network and device split with CPA per cell. Table 3: keywords by quality score with the weak component. Then a 4-sentence summary: total flagged spend first, the network decision, the budget decision, the one structural fix.

04 · microsoft-ads-search-audit

Ryze, this guide

Treat it as its own auction, not Google's shadow

The prompt form of the skill, for a first run on exports. Pull the three reports from Microsoft Advertising, Reports, for the same 30 days.

Prompt
You are a senior Microsoft Advertising analyst. Using the attached search term, keyword and campaign reports (same 30 days): first confirm the three agree on total spend within 2%. Then (1) list search terms with $25+ spend and zero conversions, split irrelevant vs unconverting, and build exact and phrase negatives; (2) compare syndicated search partners vs Microsoft sites on CTR, conversion rate and CPA, and say whether to change the network setting; (3) list keywords with quality score 5 or lower in the top spend quartile with the weak component named; (4) split keywords losing 30%+ impression share to budget (with CPA under target) from those losing to rank; (5) if the account was imported from Google Ads, list settings that did not map and duplicated keywords. Three tables, then four sentences: flagged spend, network decision, budget decision, one structural fix.
Cadence: Weekly, Tuesdays · Mode: read-only

Get a free instant audit

Get a free, instant read on your paid ads or SEO — and fix it right away.

Paid ads audit

  • Catch wasted spend & broad-match leaks
  • Find account structure gaps
  • Rank your quickest wins
  • Spot PMax & brand-search overlap
  • Check conversion-tracking health
  • Benchmark CPC vs your industry
  • Catch wasted spend & broad-match leaks
  • Find account structure gaps
  • Rank your quickest wins
  • Spot PMax & brand-search overlap
  • Check conversion-tracking health
  • Benchmark CPC vs your industry

Free · no credit card · instant

SEO audit

  • Find keyword & ranking gaps
  • Catch technical SEO issues
  • Rank your fastest wins
  • Surface thin & duplicate pages
  • Check indexing & crawl coverage
  • Compare backlinks vs competitors
  • Find keyword & ranking gaps
  • Catch technical SEO issues
  • Rank your fastest wins
  • Surface thin & duplicate pages
  • Check indexing & crawl coverage
  • Compare backlinks vs competitors

Free · no credit card · instant

The cross-platform layer

Four skills that do not care which platform the row came from. Two of them are setup work you do once; two run on a schedule. They are the difference between five separate audits and one media plan. The generalist ads skill from marketingskills sits alongside them: it covers Google, Meta, LinkedIn and X in one file and reads the shared product-marketing context, which makes it the right tool for “write me a LinkedIn variant of this Meta ad” and the wrong tool for an audit.

Naming and UTMs: the two you run first

Cross-platform reporting fails on labels, not math. If LinkedIn calls a campaign Q3_Webinar_Enterprise, TikTok calls it webinar test 2 and Microsoft still has the Google import name, no skill can line them up. The campaign-naming-convention-builder takes your current names from every platform and proposes one scheme (platform, objective, audience, geo, creative version, date) with a rename map. The utm-tracking-generator then builds UTM parameters that mirror the scheme, so GA4 and the CRM see the same segments the platforms do. Run both once; rerun the naming skill when a new client or channel arrives.

05 · campaign-naming-convention-builder

Ryze

One naming scheme across every platform

Paste the current campaign names from every platform. It returns the scheme and a rename map you can apply platform by platform.

Prompt
You are a paid media operations lead. Here are the current campaign names from LinkedIn, TikTok, Reddit, Microsoft, Google and Meta: [PASTE]. Propose one naming convention with fixed, ordered segments (platform | objective | audience | geo | creative version | launch date), using delimiters that survive every platform's UI and CSV export. Then output a rename map: current name | proposed name | platform, and list the names that could not be parsed and what is missing from them. Keep segment values to a controlled vocabulary and print that vocabulary.
Cadence: Once, then per new channel · Mode: read-only

06 · utm-tracking-generator

Ryze

Every click lands with the same labels

Give it the naming scheme and the landing URLs. It returns tagged URLs per platform and the GA4 event names that match, with the platform-specific placeholders where the ad platform fills values in.

Prompt
You are a marketing analytics lead. Using this campaign naming convention: [PASTE SCHEME] and these landing URLs: [PASTE], generate UTM-tagged URLs for LinkedIn, TikTok, Reddit and Microsoft Advertising. Use utm_source and utm_medium values that match how GA4's default channel grouping classifies paid social and paid search. Put the campaign name in utm_campaign exactly as the convention writes it, and use each platform's dynamic placeholders for utm_content where the platform supports them. Output a table: platform | landing page | full URL, then the GA4 event names these campaigns should fire, and a checklist of what breaks if a UTM is edited inside the platform later.
Cadence: Once, then per launch · Mode: read-only

Channel mix: the weekly one

The channel-mix-optimizer is the skill that justifies the other eight. It reads one campaign-level export per platform and estimates marginal CPA per channel: how the last increment of spend converted, not the average. A channel with a great average CPA can be the wrong place for the next dollar if its last 20% of spend converted at twice the rate. The skill flags channels past diminishing returns and proposes a reallocation that holds total budget flat, with the projected change in conversions and a note on which moves need a stabilization window. It recommends; it does not move anything.

Example output of the channel-mix-optimizer skill across five paid channels with illustrative numbers: Google Search, Meta, LinkedIn, TikTok and Microsoft Ads, each with monthly spend, average CPA, marginal CPA and a proposed shift; LinkedIn is flagged past diminishing returns with a marginal CPA well above its average, spend is proposed to move from LinkedIn and Meta into Microsoft Ads and TikTok, total budget is held flat, and the projected conversion change and a 7-day stabilization note are shown
Illustrative output of channel-mix-optimizer. The numbers are an example, not a client account. The column that matters is marginal CPA; the average column is what most budget meetings look at instead.

07 · channel-mix-optimizer

Ryze

Move the next dollar to the cheapest next conversion

One campaign-level CSV per platform, 30 to 90 days, same date range, same conversion definition. If the conversion definitions differ across platforms, say so in the prompt; the skill will normalize to the one you name.

Prompt
You are a performance marketing lead managing $[X]/month across Google Search, Meta, LinkedIn, TikTok and Microsoft Advertising. Attached: one campaign-level export per platform, last [30/60/90] days, same date range. The conversion to optimize is [NAME]; where a platform reports a different conversion, normalize to it and say how. For each channel: total spend, conversions, average CPA, and marginal CPA (how the top 20% of daily spend converted vs the rest). Flag channels whose marginal CPA is 1.5x their average or above target. Propose a reallocation that holds total budget flat and maximizes conversions, showing current vs proposed spend per channel, projected conversions, and which shifts need a 7-day stabilization window before judging. State every assumption.
Cadence: Weekly, Mondays · Mode: read-only

Attribution comparison: the monthly one

Every platform claims the conversion. LinkedIn reports it, TikTok reports it, Microsoft reports it, and GA4 gives it to organic. The attribution-model-comparison skill does not pick a winner; it shows how credit shifts across last-click, first-click, linear, time-decay and data-driven views for the same period, so you can see which channels only look good under one model before you change the one the channel-mix skill uses. It reads a GA4 or CRM conversion export with source, medium and campaign, which is why the UTM skill runs first. The tool landscape for stitching this together is on cross-platform ad attribution tools.

08 · attribution-model-comparison

Ryze

See which channels only win under one model

Export conversions from GA4 (Advertising, Model comparison, or a raw export with source, medium, campaign) or from the CRM with the UTM fields, one month.

Prompt
You are a marketing measurement analyst. Using the attached conversion export (source, medium, campaign, conversion date, value), compare how conversion credit for the last month is distributed across channels under last-click, first-click, linear, time-decay and, if present, data-driven attribution. Output a table: channel | credit under each model | platform-reported conversions. Flag channels whose credit swings more than 30% between models and channels whose platform-reported number is over 1.5x their best-case modeled credit. End with a recommendation on which model the weekly channel-mix decision should use and the one metric to watch if that model is wrong.
Cadence: Monthly · Mode: read-only

A cross-platform weekly cadence

About ninety minutes a week if each skill has a slot, much more if they run whenever something already feels wrong. The cadence assumes exports; with connectors and Cowork the scheduled rows run themselves. A role-by-role version of the same week, for buyers who also carry Google and Meta, is in the media buyer’s guide to Claude.

Cross-platform weekly cadence for Claude paid-ads skills: Monday 25 minutes, channel-mix-optimizer on all platforms plus a pacing check; Tuesday 15 minutes, microsoft-ads-search-audit on the search term, keyword and campaign reports; Wednesday 20 minutes alternating weeks, linkedin-ads-audit and reddit-ads-audit; Thursday 15 minutes, tiktok-ads-audit for hook and creative decay; first week of the month 45 minutes, attribution-model-comparison and a naming and UTM re-check; on demand, the generalist ads skill for a variant or a quick take
About ninety minutes a week plus a monthly hour. Diagnostics run before the budget meeting, not after the CPA moves.
SlotSkillInputDecision it feeds
Monday, 25 minchannel-mix-optimizerOne campaign export per platform, 30 daysThis week's budget shifts
Tuesday, 15 minmicrosoft-ads-search-auditSearch term, keyword, campaign reportsNegatives, network setting, budget caps
Wednesday, 20 minlinkedin-ads-audit / reddit-ads-audit (alternate weeks)Campaign Manager or Ads Manager exportAudience and community changes, form fixes
Thursday, 15 mintiktok-ads-auditAds Manager custom report, ad level, dailyWhich creatives to cut, refresh or scale before weekend spend
Monthly, 45 minattribution-model-comparison, then naming and UTM re-checkGA4 or CRM conversion exportWhich model the Monday decision uses; any drift in labels
On demandads (marketingskills)Whatever you pasteA variant, a quick read, a brief

The honest test of the cadence is which slots get skipped. A skipped slot means the output is not driving a decision (rewrite its output contract so it ends in a decision) or the slot is wrong (move the slot, not the habit). If Monday is the slot that keeps slipping because six exports is too many, that is the signal to add connectors, in this order: TikTok (official), then whichever community server covers your biggest remaining spend. For the mechanics of chaining them, see automating cross-platform ads with Claude and MCP and connecting Google, Meta, TikTok, GA4 and Shopify to Claude.

Write your own platform skill

The two skills printed on this page were written the way you would write one for Pinterest, Snap, Amazon Ads or Apple Search Ads: from the platform’s own export, not from a generic ads template. The order that works:

  • Start from the export, not the prompt. Open the platform’s report builder and list every column it can give you. The columns that exist nowhere else (6-second views on TikTok, syndicated partner network on Microsoft, seniority on LinkedIn, community on Reddit) are the audit. Name them in the input contract exactly as the platform labels them.
  • Write the description as a trigger and a fence. “Audits a TikTok Ads account from a TikTok Ads Manager export. Not for Meta Reels.” The fence stops it firing on the wrong platform, which is the most common failure once you have four audit skills installed.
  • Make step one an echo. Date range, row count, total spend. It catches the filtered or mismatched export before the analysis does.
  • Put the platform’s specific failure modes in the steps. Not “check for waste” but “compare syndicated partners against Microsoft sites and recommend the network setting if partner CPA is 1.5x.”
  • End in a decision. The output contract fixes columns, sort order and a verdict vocabulary (refresh, new cut, scale, pause). Then a sentence count and the rule that the biggest dollar finding comes first.
  • Test on a month you already audited by hand and diff. Where the skill disagrees with you, one of you is wrong; find out which, and fix the file, not the prompt.

Keep SKILL.md under a page; benchmark tables and the column glossary go in separate files inside the folder. The Claude Code version of this loop, where the skill pulls its own export through a connector and writes the report to a file, is in Claude Code marketing workflows.

Troubleshooting

The failures on these four platforms are mostly input failures. Five patterns cover nearly all of them.

The audit is thin

The export is at the wrong level. TikTok and Reddit audits need ad level with breakdowns; Microsoft needs three separate reports. Re-export with the columns the input contract names, and make the skill echo the row count so you see the problem first.

The wrong platform skill fires

Two descriptions overlap (both say "audit paid social"). Add the fence: name the platform in the description and say what it is not for. Name the skill explicitly once to confirm, then fix the description.

TikTok video columns are missing

Custom reports only include video metrics if you add them. In TikTok Ads Manager, edit the custom report, add 2-second views, 6-second views, played to 100% and average watch time, and export again.

Microsoft totals disagree

The three reports were pulled for different date ranges or one has a filter applied. The skill stops at a 2% mismatch on purpose; re-pull all three for the same range with no filters.

Channel mix numbers look wrong

Different conversion definitions per platform, or different attribution windows. Name the conversion in the prompt, tell the skill to normalize, and run the attribution comparison before trusting a reallocation over 15% of any channel's budget.

One boundary to keep in view: every skill here reads and recommends. Applying the change, on four platforms with four UIs, is still a person’s job unless an execution layer does it. Ryze AI runs the same analyze-decide loop and applies changes with an approval step; that is a product decision separate from installing skills, and the skills work the same either way.

Frequently asked questions

What are Claude skills for paid ads?

Claude skills for paid ads are SKILL.md folders that teach Claude one recurring ad job: audit a LinkedIn account, find wasted Microsoft Ads spend, propose a channel-mix shift. Claude loads a skill only when a request matches its description. The Google and Meta skills live in the Ryze marketing-skills repo; this page covers the skills for LinkedIn, TikTok, Reddit and Microsoft Ads plus the cross-platform layer.

How do I connect paid media data to Claude?

Three routes, by platform. TikTok publishes an official TikTok for Business MCP Server in its Marketing API docs. Google and Meta have official servers or the Ryze connector. LinkedIn, Reddit and Microsoft Advertising have no first-party MCP as of September 2026, so you either use a community MCP server wrapping their ads APIs or export a CSV from the platform UI and attach it to the chat.

Can Claude audit LinkedIn Ads?

Yes. Export campaign and demographic reports from LinkedIn Campaign Manager as CSV, attach them, and run the linkedin-ads-audit skill from the Ryze marketing-skills repo. It checks CTR against your own account history, audience quality by job title and seniority, lead-form completion rate, and budget efficiency, then returns a scored table with one fix per finding. Live queries need a community LinkedIn Ads MCP server.

Is there a TikTok Ads MCP for Claude?

Yes. TikTok documents an official TikTok for Business MCP Server in the Marketing API v1.3 developer portal, described as a bridge for connecting AI agents to the TikTok Ads platform. Community servers wrapping the Marketing API existed before it. If you do not want to set up either, export a custom report from TikTok Ads Manager as CSV or XLSX and attach it; the tiktok-ads-audit skill on this page reads that format.

How do I connect TikTok Ads to Claude?

Either add the official TikTok for Business MCP Server or a community TikTok Ads MCP server to Claude Desktop or Claude Code and authorize it with a TikTok Business app, or skip the connector and export from TikTok Ads Manager under Reports. For a first audit the export is faster. The skill runs the same on either input; only the data-fetch step changes.

Can Claude audit Reddit Ads?

Yes. Reddit Ads Manager exports campaign, ad group and ad reports with breakdowns by community, placement and device. Attach the export and run the reddit-ads-audit skill from the Ryze repo. It scores community targeting, creative fit for a comment-driven feed, subreddit-level performance and bid inefficiencies. Reddit has no first-party MCP; community servers wrap the Reddit Ads API v3 for live access.

Does Claude work with Microsoft Ads (Bing Ads)?

Yes, through exports or a community MCP server; Microsoft has not shipped a first-party MCP for Microsoft Advertising as of September 2026. Export search term, keyword, campaign and quality score reports from the Reports tab, attach them, and run the microsoft-ads-search-audit skill written on this page. It treats the account as its own auction, not as a copy of Google Ads.

Which platforms have an official MCP server for Claude?

Google Ads (self-hosted, read-only), Meta Ads (hosted at mcp.facebook.com) and TikTok (TikTok for Business MCP Server in the Marketing API docs) publish first-party servers. LinkedIn, Reddit and Microsoft Advertising do not; their ads APIs are covered by community MCP servers. Ryze's own connector covers Google, Meta and five more platforms. For every platform, a CSV export still works and is the fastest first step.

What is cross-platform ad reporting with Claude?

One report that puts LinkedIn, TikTok, Reddit, Microsoft, Google and Meta on the same rows: spend, conversions, CPA, and the marginal CPA of each channel's last increment of spend. The channel-mix-optimizer skill produces it from one CSV per platform. It only works if campaign names and UTMs follow one convention, which is why the naming and UTM skills come first in the setup order.

What is the channel-mix-optimizer skill?

A Ryze marketing-skills file that reads 30 to 90 days of campaign data from every paid channel, estimates marginal CPA per channel, flags channels past diminishing returns, and proposes a reallocation that holds total budget flat. It outputs current versus proposed spend per channel and the projected change in conversions. It recommends; moving the budget is a separate, approved step.

Where do I install these skills?

In claude.ai, enable code execution under Settings then Capabilities, open Customize then Skills, click plus, Create skill, Upload a skill, and upload the skill folder as a zip. In Claude Code, drop each folder into ~/.claude/skills for personal use or .claude/skills in a repo, and invoke with the slash command. The generalist ads skill installs with npx skills add coreyhaines31/marketingskills --skill ads.

Do I need a different skill for each ad platform?

For audits, yes. LinkedIn is judged on lead quality by seniority, TikTok on video completion and creative decay, Reddit on community fit, Microsoft on search terms and partner-network waste. One generalist audit misses all of that. For the cross-platform layer (naming, UTMs, channel mix, attribution comparison) one skill covers every platform, because those jobs are about consistency rather than platform mechanics.

Can Claude make changes in LinkedIn, TikTok, Reddit or Microsoft Ads?

The skills on this page are read-only analysis; none pauses a campaign or moves a budget. Some community MCP servers expose write tools, and we recommend keeping them off until a read-only weekly loop has run for a month. An execution layer such as Ryze AI applies changes with an approval step; that is a separate product decision from installing skills.

Let AI run your ads

Google, Meta + 5 platforms · audits 24/7

Start free trial
Live results across
2,000+ clients

Paid Ads

Avg. client
ROAS
0x
Revenue
driven
$0M

SEO

Organic
visits driven
0M
Keywords
on page 1
48k+

Websites

Conversion
rate lift
+0%
Time
on site
+0%
Last updated: Sep 8, 2026
All systems ok