A practical guide for marketers at every level — beginners and experts alike.
If you run paid ads, you know how the job actually feels. You're bouncing between Google Ads, Meta Ads Manager, a spreadsheet, a brief doc, a competitor research tab, and three Slack threads — all before lunch. The actual thinking gets squeezed out by the busywork.
Claude skills are pre-built workflows inside Cowork, the desktop version of Claude. Instead of prompting from scratch every time, a skill already knows the structure of the task. You trigger it, answer a few questions, and get something you can actually use — no prompt engineering, no explaining context repeatedly.
Here are the 10 skills I now use to run everything — from pre-launch research to stakeholder reporting — and exactly how I use each one.
Every platform needs a signal to optimise against: Meta Pixel + Conversions API, Google Tag with conversion tracking, LinkedIn Insight Tag. Without it, you're paying for clicks with no way to know which ones turned into customers. Set this up and verify it's firing before you spend a dollar.
Meta's algorithm needs roughly 50 conversions per week per ad set to exit its learning phase. Google Smart Bidding is similar. If your budget can't generate that volume, optimise for a higher-funnel event (e.g., 'lead form opened' instead of 'purchase') to give the system enough signal to work with.
CPL = Cost Per Lead. CPA = Cost Per Acquisition. ROAS = Return on Ad Spend (revenue per $1 spent — a 3x ROAS means $3 revenue per $1 spent). CTR = Click-Through Rate. CPM = Cost per 1,000 impressions. CPC = Cost Per Click.
Cold = people who've never heard of you (targeted by interest, lookalike, job title). Warm = people who've engaged but not converted. Hot = high-intent visitors (pricing page, cart abandoners). Each layer needs different creative and different messaging.
The first skill I reach for before any campaign is marketing:competitive-brief. It researches competitors' ads, positioning, and messaging — then surfaces the gaps you can actually exploit.
You tell Claude who your top competitors are, what product category you're in, and who your target audience is. It comes back with a structured breakdown of each competitor's value proposition, which emotional and rational angles they're leaning on hardest, and — most usefully — what none of them are saying. That last part is where your brief writes itself.
The Meta Ad Library (facebook.com/ads/library) is free and public — search any brand and see every ad they're currently running. Do this before briefing Claude so you can give it specific examples to work with.
Don't limit competitive research to direct competitors. Look at adjacent categories winning your audience's attention. If you sell project management software and a productivity tool is dominating with a certain message, that tells you what your audience cares about even if it's not a direct competitor.
Run this skill every 6-8 weeks during active campaigns, not just at launch. Competitors shift messaging based on what's working. Catching that shift early is a real edge.
Before I touch creative, I use apollo:prospect to build a precise audience list. You describe your ideal customer in plain English — company size, industry, job title, seniority, geography, even technology signals — and it builds a ranked list of matching decision-makers with verified contact info.
I use these lists to upload custom audiences to Meta for lookalike modelling, build LinkedIn Matched Audiences, and set up Google Customer Match. A clean, precise seed list produces better lookalikes than exporting your full CRM and hoping for the best.
A lookalike audience is when you give a platform (Meta, Google, LinkedIn) a list of your best customers, and the platform finds other users who look statistically similar to them. The quality of your seed list directly determines the quality of your lookalike.
On Meta, 1-2% lookalikes are more precise but smaller in scale. 5-10% are broader and cheaper to reach but less targeted. Test both, but start with 1-2% for efficiency. As third-party cookies continue phasing out, first-party data lists become your most durable targeting asset — build them now.
Run this skill on your highest-LTV, lowest-churn customers — not your full list. The tighter the seed, the sharper the lookalike.
This sounds obvious but it's the step most teams rush. marketing:campaign-plan forces the thinking upfront. You give it your goal, audience, channels, timeline, and rough budget range, and it produces a complete campaign brief — which audience goes in which funnel stage, how to split budget, what to test first, what success looks like by week two and week four.
The brief becomes the document your whole team aligns on before anyone builds anything. It also becomes the source of truth when things go sideways mid-campaign and someone wants to change direction.
The most common mistake is running one campaign to one audience with one message. A proper campaign has at least three layers: cold (awareness), warm (consideration), hot (retargeting). Each needs different creative and messaging. Your cold audience doesn't know you exist — don't lead with a 'buy now' CTA.
Be explicit about which campaign objective you're using. Meta's 'Traffic' objective optimises for clicks; 'Leads' for form fills; 'Conversions' for purchase events. Same budget, same creative, different objective = very different results. Performance Max on Google is intent-capture (people searching for you); Meta Advantage+ is demand-creation (finding people who aren't looking yet). Use them for different funnel stages.
After you get the plan, paste in your competitive brief from Skill 1. Ask Claude to refine the messaging based on the gaps found. That's where the plan gets sharp.
Once the strategy is set, marketing:draft-content handles the writing. Headlines, primary text, descriptions, CTAs — across every platform, at volume, with correct character limits baked in.
| Platform | Headline | Primary / Intro Text | Description |
|---|---|---|---|
| Google RSA | 30 chars (up to 15; 3 shown at once) | N/A | 90 chars (up to 4) |
| Meta | ~40 chars displayed | 125 chars before "See more" (no hard limit) | 30 chars |
| 70 chars (hard limit) | 600 chars (truncates to ~150 on desktop) | 300 chars | |
| TikTok | N/A | 100 chars | N/A |
I ask for 6-7 headline variations per angle. For Google RSA, more asset variety gives the ML system more to test — it picks the combinations, not you. For Meta Dynamic Creative, the same logic applies.
Every ad needs four things: a hook (why stop scrolling?), value (what's in it for me?), proof (why should I believe you?), and a CTA (what do I do next?). Also — your ad is only as good as the landing page it sends to. If your ad promises one thing and the landing page says something different, you'll lose conversions at that step.
Creative fatigue is real and underdiagnosed. On Meta, when frequency exceeds 3-4 on cold audiences, performance typically declines — watch this metric weekly. For Performance Max, provide at least 15 headlines, 4 descriptions, and a mix of image and video assets. Asset variety is what allows PMax to find efficient combinations across placements.
Ask for variations that address different objections — not just different phrasings of the same value prop. 'Too expensive,' 'takes too long to set up,' 'not sure if it fits my team' — each objection needs its own headline angle.
Before any copy goes to brand or compliance review, I run it through brand-voice:brand-voice-enforcement. This checks your ad copy against your brand's voice, tone, and messaging guidelines — and flags anything off with a specific suggested fix.
When you're generating high volumes of copy across multiple campaigns, things drift. An overly casual headline in an otherwise formal campaign. A phrase that contradicts your positioning. This catches it before it becomes a meeting.
Brand voice is how your company 'sounds' in writing — formal vs. casual, technical vs. plain English, bold vs. conservative. Consistency matters because people need to encounter your brand multiple times before they trust it enough to click. If your ad sounds completely different from your website, that friction costs you conversions.
Brand voice drift in high-volume creative production is one of the most common causes of declining CTR over time. When ads start sounding generic because a dozen people have touched the copy, performance suffers. This skill acts as an automated first-pass review that catches inconsistency before it reaches human reviewers.
Do this before brand review meetings, not after. Come in with copy that's already passed the filter and you'll spend less time defending choices.
marketing:brand-review checks for unsubstantiated claims, missing disclaimers, prohibited language by platform, and consistency issues across the full campaign. Google and Meta will disapprove ads with superlative claims like 'the best' or 'guaranteed' without substantiation. In regulated industries — financial services, healthcare, legal, supplements — the rules are stricter and the consequences go beyond a disapproval notice.
Ad disapprovals happen more often than you'd think and can occur mid-flight. Common triggers: 'free' used without clear conditions, before/after claims, anything implying a medical benefit, and personal attribute targeting language (e.g., 'struggling with debt?'). Always check platform policies for your specific product category before launch.
Meta's Conversions API (CAPI) is now essential, not optional. Post-iOS14, browser-based pixel tracking alone misses a significant share of conversions — estimates range from 15-40% signal loss depending on the audience. Without server-side tracking, your algorithm is optimising against a partial picture, and your reported ROAS is inflated relative to actual performance. Set up CAPI before any meaningful spend.
If you're in a regulated industry, paste in your compliance team's specific guidelines alongside the copy. Claude will cross-reference them rather than checking only against general platform rules.
Once a campaign is live, I use data:analyze to pull meaning out of raw numbers. Export your performance data as a CSV and ask a specific question — not 'how is it doing?' but something targeted: 'Which ad sets are beating my target CPA and are under-invested relative to their performance?'
The output isn't a description of the data. It's an analysis with a recommendation: 'Ad Set 3 has a $14 CPA vs. your $22 target. It's receiving 15% of budget. Based on current performance, you could scale to 35% before efficiency degrades.' That's the difference between a dashboard and a decision.
Don't make decisions in the first 3-5 days of a campaign. Platforms have a learning phase during which performance fluctuates as the algorithm tests combinations. Optimising too early resets the learning phase, wastes budget, and locks you into a suboptimal audience. Give each ad set at least 7 days and 50 conversions before drawing conclusions.
Attribution windows matter enormously and are often the source of disagreement between platform reporting and business reality. Meta defaults to 7-day click / 1-day view. Google defaults to 30-day click. These windows overlap — the same conversion can be claimed by both. For a realistic picture, cross-reference CRM data and consider incrementality testing to understand true lift beyond what platform attribution reports.
Ask a decision question, not a reporting question. 'What's our ROAS?' is reporting. 'Should we reallocate $X from LinkedIn to Meta based on this week's CPL data?' is a decision. Claude returns much more useful output on the second type.
The most expensive mistake in paid ads is declaring a winner before you have enough data. data:statistical-analysis tells you whether your test results are statistically significant — or whether you're looking at noise. You give it Variant A and B performance, tell it your confidence threshold, and it tells you: call it now, keep running, or here's how much more data you need.
A 95% confidence level means there is a 5% chance (1 in 20) the difference you're seeing is random rather than real. The minimum data standard for conversion rate tests is at least 100 conversions per variant — not a set number of days. Running a test for 'a week' is not a method.
Test one thing at a time. If you change the headline, the image, the offer, and the audience simultaneously, you won't know what caused the performance difference. A proper A/B test changes exactly one variable and holds everything else constant.
Platforms' built-in A/B test tools — Meta's Experiments feature, Google's Ad Variations — are the cleanest way to run tests because they split traffic at the auction level, preventing audience overlap. Running two ad sets manually in the same campaign and comparing performance is not a true A/B test — the algorithm allocates budget dynamically based on early signals, creating selection bias in your results.
Run this skill before any creative swap or budget reallocation based on test results. It takes 30 seconds and has saved me from killing winners more than once.
Retargeting doesn't have to be just ads. The most effective retargeting I've run combines a paid layer with an email sequence running simultaneously to the same warm audience. Someone clicks your ad, doesn't convert — they now get a retargeting ad and a well-timed email. For higher-ticket offers, that combination moves the needle significantly.
marketing:email-sequence designs and writes the entire flow — timing, branching logic, subject lines, and full body copy for every email. It handles exit conditions (remove on conversion) and branch paths for different engagement scenarios.
Retargeting works because most people don't buy on the first visit. B2B research suggests it typically takes 7-13 touchpoints before a buyer converts. Ads alone rarely create that many touchpoints efficiently. Email adds a lower-cost channel that stays in front of the same person without paying CPM every time.
Frequency management is the hidden lever in retargeting. On Meta, cap retargeting frequency at 5-7 per week for warm audiences. For hot audiences (cart abandoners, pricing page visitors), segment by recency — a 3-day website visitor and a 30-day website visitor should see very different messages. Pair high-frequency ad exposure with email so not every touchpoint costs ad spend.
Align email timing with your retargeting ad schedule. Email 1 goes out the same day as the first ad impression. Email 3 drops when you increase ad frequency. It creates a surround-sound experience rather than two disconnected channels competing for the same conversion.
The last skill in the workflow is marketing:performance-report, and it might be the most politically important one. If you can't communicate results clearly, you lose budget. Leadership cuts what they don't understand.
You feed it your campaign data and context (goals, what happened mid-campaign, who is reading this), and it builds a clean stakeholder-ready document: executive summary, KPI scorecard against targets, trend analysis, wins, honest diagnosis of misses, and ranked recommendations for next period.
Always frame results against the target that was set at the start. 'We got 400 leads' means nothing. '400 leads at a $28 CPL against a $35 target — 20% below goal' means something. The format is always: target, actual, variance, why.
Different stakeholders need different layers of the same data. Your CEO wants three numbers and a recommendation. Your CFO wants cost efficiency and payback period. Your CMO wants channel mix. Your own team wants ad-set-level breakdown and next actions. Brief Claude on the audience before it builds the report and ask for multiple versions from the same dataset.
Include 'what we learned' as a dedicated section, not just 'what happened.' Stakeholders who see you learning as well as reporting will trust your budget more.
These 10 skills aren't a replacement for strategic thinking — they're what happens when strategic thinking doesn't get crowded out by admin. The research is faster, the creative comes out more consistent, the analysis is clearer, and the reporting actually gets done.
The workflow chains together naturally: competitive brief and audience research inform the campaign plan. The plan drives the creative brief. Creative goes through brand and compliance QA. Once live, you're analysing performance, validating tests, running the email retargeting layer in parallel, and producing reports that make the case for continued investment. Every stage is accounted for. And every stage is faster than it used to be.
If you're a marketer building out your paid-ads workflow and want to talk through how to make this work for your specific setup, reach out — I'd love to compare notes.
hello@kerrycokelekoglu.com kerrycokelekoglu.com