AI won't replace performance marketers, but it replaces the parts of the job nobody was doing well anyway: the daily account read, the fatigue diagnosis, the search-term review, the character-count check on every ad asset. What it doesn't replace is the decision about whether to run ads at all, which is an economics question and not a targeting one. The gtm-skills Performance Marketer pack is 30 free, open source Claude Skills covering paid social and paid search, and every one of them is read-only or drafts a paused change.
You're the person who has to decide, this month, whether more money goes into ads. The only scoreboard you have is the one the platform wrote about its own performance. Your best-performing campaign has a return figure that would be great if it were true, and a nagging feeling that the customers behind it would have bought anyway.
Here's what changes with AI in that job, what doesn't, and the 30 free skills that cover the parts that do. Also the part most guides skip: the three questions that decide whether you should be running ads at all.
The gate before the tactics
Most paid-ads advice starts with how. The honest starting point is whether. Paid ads is a loop: a campaign has to earn enough to pay for the next campaign. When the loop closes, ads compound. When it doesn't, ads are a subscription to losing money, and no targeting trick fixes that.
| The question | Why it decides everything downstream |
|---|---|
| What is a new customer worth in the first month, cash in hand? | Not lifetime value. Month one. If the money comes back in 30 days you can run the loop on your own cash instead of floating spend for quarters. |
| What can you afford to pay for a customer? | If month-one value is $80, you cannot pay $100 and call it growth. Write the number down. Every skill in the pack reads it. |
| Can you spend 2-3x that per angle without flinching? | Signal is lumpy. Zero sales on the first $100 and three on the next is normal. If the test budget can't survive the noise, the test will lie to you. |
If the gate says no, fix the offer, the price, or the follow-up first. Ads amplify a working machine, they don't build one. Running them early doesn't just waste money, it produces a false conclusion ("ads don't work for us") that costs you the channel later.
There's one sanctioned exception. A small daily angle test is legitimate even pre-scale, as a messaging lab. You're not buying growth, you're buying knowledge about which pain, which person, which promise. That knowledge compounds across everything, including your organic content and your landing pages.
What AI actually replaces in this job
The parts genuinely at risk were never the hard part. They're the parts that require doing the same disciplined thing every day, which is exactly where humans lose to software.
| Task | Automatable? | Why |
|---|---|---|
| The daily account read | Yes | Same comparison, same five checks, every morning. Nobody sustains this manually past week three. |
| Diagnosing creative fatigue | Yes | It's a two-condition rule applied consistently. Humans apply it inconsistently and usually too early. |
| Classifying a search-terms export | Yes | Business fit first, evidence second, hundreds of rows. Tedious, mechanical, high-value. |
| Counting characters on 15 headlines and 4 descriptions | Yes | There is no judgment in it and it is skipped constantly. |
| Deciding whether the loop closes | No | An economics question about your business, not a question about the account. |
| Deciding an angle is exhausted rather than tired | No | Requires knowing what you've already tried and what the market has heard. |
| Saying the offer is the problem | No | The most valuable sentence in paid media, and the one nobody with an account to keep says out loud. |
The five rules the skills enforce
These aren't preferences. Each one exists because its absence has a predictable failure mode, and each is written into the skills as a guardrail rather than a suggestion.
- The offer is fixed. The angle is the experiment. An angle is who the ad is for, plus the pain it names, plus the outcome it promises. One offer can carry four completely different angles. You test angles against each other, and variations only come after an angle wins.
- Your ad is the targeting now. With interest categories retired and exclusions removed, delivery aims by reading the creative. Specificity is the lever you still hold. "For everyone" gives the machine nothing to aim with.
- Spend enough to see signal before judging. Two to three times your target cost per result, per angle, before a verdict. An ad that got $12 of a $400 test wasn't tested, it was starved.
- Read daily, act rarely. Daily edits reset learning and kill compounding winners. But not looking is how budgets quietly burn. The resolution is a read-only audit every morning and changes on a schedule.
- Keep it simple. One campaign, one broad ad set, three to five genuinely different angles as separate ads. Low budgets die of too much structure, not too little.
What's in the pack
Twenty-eight skills, adapted from Kelpi's MIT-licensed Meta and Google ad skill collections and rebuilt to this repo's conventions. Sixteen cover paid social, twelve cover paid search. Every skill that touches an account is read-only or drafts a paused change.
| Stage | Paid social | Paid search |
|---|---|---|
| Understand the market | the-brand-kit-reader, the-verbatim-miner, the-ad-library-miner | the-search-term-miner, the-search-intent-mapper |
| Decide the message | the-promise-sharpener, the-angle-spread, the-copy-formula-picker | the-search-ad-writer, the-quality-score-fixer |
| Build and launch | the-creative-producer, the-targeting-verdict, the-campaign-drafter, the-lead-form-builder, the-product-set-builder | the-query-promoter, the-negative-keyword-builder |
| Watch it daily | the-morning-ad-audit, the-fatigue-check | the-delivery-triage |
| Decide weekly | the-angle-scoreboard, the-scale-pacer | the-search-scorecard, the-search-week-review, the-bid-strategy-picker, the-change-plan-builder |
| Check the measurement | the-pixel-audit | the-conversion-goal-audit |
The two measurement skills sit at the bottom of that table and should probably sit at the top. Every other row inherits whatever the tracking says, and the most common expensive mistake in paid media is a confident decision made on a number that was already wrong.
The guardrails that do the real work
Skills are easy to write and hard to make safe. Four of these carry rules that exist specifically to stop a plausible-looking output from costing money.
- the-negative-keyword-builder refuses to run without a protected-query list. A too-broad negative is the only mistake in paid search that leaves no evidence: the demand it removes never shows up as a row anywhere. So every proposed negative is collision-checked against the searches that must keep running, before it's added rather than after.
- the-fatigue-check keeps a "looks tired, is not" section. Half the value is naming the auction shift, the seasonality, the tracking break, or the starved test that's the likelier cause. Impatience kills more winners than fatigue does.
- the-angle-scoreboard won't present platform-reported return as business truth. It pairs it with blended cost per customer from your own records, separates click-through from view-through, and ends with at most three decisions the data supports plus the ones it doesn't.
- the-scale-pacer checks the loop math before proposing any scaling. If the winner isn't profitable at scale on your numbers, it says so instead of scaling it. Scaling an unprofitable winner faster is the most expensive thing a budget skill can do.
The honest gap
Every skill here reads what the ad platform reports about itself. The platform grades its own homework: attribution flatters, view-through inflates, and an excellent reported return can sit on top of zero incremental revenue. The skills say so, loudly and in every relevant output, but saying so doesn't fix it.
Closing that gap needs a second source of truth: conversions recorded independently and carried through to what the customer actually paid. That's what turns a platform return figure into blended cost per customer, and it's the number that decides whether the loop closed. If you're already thinking in those terms, the revenue attribution problem is the same one from a different angle.
Worth saying plainly: Intempt does not ingest ad-platform spend natively today. The join between spend and revenue is real and it's something you bring the numbers to, not something that happens by itself. Pretending otherwise would be exactly the flattery these skills are built to refuse.
Install it
One command installs the whole catalogue, all eight jobs. Or take the Performance Marketer pack on its own and ignore the rest.
# the whole catalogue
npx skills add sidchaudhary/gtm-skills
# just one skill
npx skills add sidchaudhary/gtm-skills/skills/performance-marketer/the-morning-ad-auditStart with the-morning-ad-audit if you already run ads. Start with the-brand-kit-reader if you're at zero. If you want the rest of the catalogue first, the full skills directory has all 96 across the eight jobs, and the ad copy generator is the free tool that pairs with the copy skills in this pack.
Frequently asked questions. Answered.
No, not the role. It replaces specific tasks inside it: reading the account every morning, applying the fatigue rule consistently, classifying a search-terms export, counting characters on 15 headlines, drafting negatives with collision checks. Those are real hours and they're the hours most often skipped. What it doesn't replace is the judgment about whether the loop closes, which angle is worth a fourth execution, and when the honest answer is that the offer is the problem rather than the targeting.
