marketing says what the product is and who it's for, in words a buyer believes — then creates, captures and attributes demand for it. The launch. Being found, in search, in an AI answer, in an app store, in someone else's newsletter. The page that converts the visit, the email that arrives, the ad budget and whether it did anything. What makes it different is that it grades its own evidence: marketing advice mixes real, replicated findings with folklore repeated until it sounded measured — so every claim here says how solid it is and where it stops applying. The default reader is one person building a digital product, with a small budget and no analytics team — which changes the answers, not just their scale.
runs onClaude CodeCodexCursorAntigravityopencodeGrok BuildHermes
attribution-and-measurement.mdsurface-selfserve-saas.md
2 of 20 loaded · read fully
Route before acting. One job, at most one base surface, overlays added only when they apply. If the request names no business model, the skill does not stall — it assumes self-serve SaaS and says which it assumed.
The router is the skill. There is no fixed campaign pipeline to run start-to-finish — each job stands alone and enters where your request is. The animation traces one path (the flagship, deciding what a measurement can honestly support); the sections below map the whole surface it routes across.
The scope is the full external-demand surface — positioning, audience, launch, SEO, AI search,
content, landing pages, email and lifecycle, paid, social and community, partnerships and PR, app
store optimization, and measurement — across every place demand actually happens: search and AI
answer engines, inboxes, ad platforms, app stores, launch platforms, integration marketplaces.
Those platforms are the terrain, never the subject — the techniques are the same; the
surfaces say how each one reshapes them. marketing writes copy, requirements, briefs,
plans and measurement specs. It writes no production code, and it does not compose the page
it briefs.
everything between the product and the person who hasn't heard of it
the "Not this skill" table — eleven asks this skill declines by design
The default reader is one person, with a small budget and no analytics team. That is not a simplification of enterprise practice — it changes the answers. Platform attribution over-claims by construction, a valid lift test needs scale almost nobody has, and marketing-mix modelling needs years of weekly data. What remains at that scale is real, and this pack teaches it instead of pretending the enterprise toolkit scales down.
SKILL.md is a router, not a script. Every request selects the smallest sufficient route: one primary job — the fourteen below — combined with at most one base surface that reshapes how the job applies to how the product is actually bought, plus additive overlays when there is no audience yet or when a model authors at scale. Read the selected references completely; load one or two, never the whole pack.
| facet | options | rule |
|---|---|---|
| ① Primary job | attribution-and-measurement ⭐ · positioning-and-messaging · audience-and-segmentation · marketing-science · go-to-market-and-launch · seo-strategy · ai-search · content-strategy · landing-pages-and-conversion · lifecycle-and-email · paid-acquisition · organic-social-and-community · partnerships-pr-and-affiliates · app-store-optimization | Exactly one. Pick the single job the request needs. The flagship answers the question the rest of the field avoids — what actually happened, and what can you honestly know at your scale. Often the answer is "less than the dashboard claims," and saying so is the job. |
| ② Base surface | surface-selfserve-saas (default) · surface-b2b-sales-assisted · surface-developer-tool | At most one. The surface reshapes what each job means for how the product is actually bought — it points to the job references, it doesn't redefine them. Many questions (what to claim, which consent regime applies) are surface-independent. |
| ③ Additive overlays ⭐ | surface-zero-audience — no list, following or customers yet surface-agentic — a model authors or decides a spend or send at scale |
Additive. Each stacks on top of a base surface, never instead of one. Read marketing-science as a fourth, orthogonal move whenever a claim invokes "brand science" or a named statistic — it grades the claim so the other references need not re-derive it. |
Each job is one reference, read fully only when its route is selected. Attribution is the flagship because every other channel reference assumes this honesty hierarchy — not the platform dashboard — is the arbiter of "did it work." This is the whole surface, not a headline slice.
| I need to… | Read | Contribution |
|---|---|---|
| Judge whether a channel, campaign or launch actually worked — and what my scale can honestly support ⭐ | attribution-and-measurement.md |
Flagship. The honesty hierarchy of measurement methods; why platform-reported ≠ incremental; why almost nobody can run a valid lift test; the n=1 answer — UTM discipline, self-reported attribution, geo-split lift, pre-committed decision rules |
| Say what the product is, who it's for, and why it beats the alternatives | positioning-and-messaging.md |
The front-door job. Dunford's five components as craft, never as science; category and competitive framing; the messaging ladder; copywriting frameworks; the staged repositioning rollout and defensibility hierarchy |
| Decide who to market to, or challenge a persona that names no buying situation | audience-and-segmentation.md |
Category entry points over persona theater; the JTBD seam with product; ICP as a funnel not a checkbox; light-buyer reality; the consent-regime checkpoint before any segment is emailed |
| Grade a "brand science" claim, a named statistic, or a budget-split rule of thumb | marketing-science.md |
The adjudication layer. The evidence ladder; the laws with their domain boundaries; the 95-5 derivation and its disclaimer; the Thomaz dispute presented, not settled; no SaaS dataset exists, and n=1 is outside all of them |
| Choose a go-to-market motion, or plan and run a launch | go-to-market-and-launch.md |
Four motions and the brutal middle; launch ≠ release ≠ deploy; major/minor/silent launch tiers; Product Hunt's enforceable rules (never solicit upvotes); the failed-launch pattern; owned/rented/borrowed channels |
| Decide what to rank for, or check an SEO claim against what Google actually documents | seo-strategy.md |
Keyword-intent mapping; topic clusters; four myths Google's own dated docs kill; the structured-data requirement marketing specifies and frontend implements; the FAQ/HowTo removal as a staleness parable; zero-click reality |
| Decide what to do about visibility inside ChatGPT, Perplexity, AI Overviews or Copilot | ai-search.md |
The platform divergence (Bing formally adopts GEO; Google formally rejects it); the crawler matrix per bot; llms.txt falsified by four independent evidence classes; the one peer-reviewed GEO anchor and its limits; citability vs. surfacing |
| Decide what content to produce, brief a writer, or set an editorial cadence | content-strategy.md |
Distribution beats production; what a brief must specify, including its information-gain claim; scaled-content abuse is purpose-tested, not authorship-tested; the traffic-collapse convergence; the copy-fidelity seam with design |
| Brief a landing, campaign or pricing page — the claims, the proof, the objections | landing-pages-and-conversion.md |
The requirements brief marketing owns and design composes against; Attention Ratio and the one-goal doctrine; trust and proof devices with placement as the lever; message-match; the ethics/dark-patterns table |
| Get marketing email delivered, and decide whether a list is lawful to send to | lifecycle-and-email.md |
Deliverability as engineering first; SPF/DKIM/DMARC and DMARC's supersession; provider divergence; the four consent regimes and the jurisdiction decision rule; stream separation; where lifecycle hands off to success |
| Decide whether to spend on ads, which format, and when it's premature | paid-acquisition.md |
What ATT changed and didn't; black-box campaign types — documented controls, undocumented effect, zero independent incrementality evidence; when paid is premature; the format taxonomy; the standard-vs-incremental death-spiral rule |
| Choose between a brand account, a founder account, or an owned community | organic-social-and-community.md |
Brand-account reach is dying and founders aren't — stated at its real evidence tier; owned vs. rented; the Orbit model with its maintenance status; DevRel is not developer marketing; an explicit list of what stays folklore here |
| Earn distribution through someone else's audience — press, marketplaces, affiliates, co-marketing | partnerships-pr-and-affiliates.md |
The site-reputation-abuse warning that kills a class of partnership pitch; marketplace listings and their traction gates; affiliate mechanics; FTC disclosure; the liveness check before naming any journalist-query service |
| Get a mobile app found and installed — listing copy, keywords, store experiments | app-store-optimization.md |
Also the pack's mobile-surface authority. Hard character limits per store; two different keyword-stuffing enforcement philosophies; Custom Product Pages ≠ Product Page Optimization; per-market localization; what stays unverified |
A fifteenth routing row is a diagnostic entry point rather than a
fourteenth reference: traffic, rankings, conversions or inbox placement fell and the cause isn't
known routes to seo-strategy.md
first — intent-and-coverage diagnosis before volume — and then to whichever channel reference the
decline actually touches. Full router table & invariants:
SKILL.md.
One base surface, at most, reshapes every job for how the product is actually bought — the same
positioning job resolves differently when a rep closes the deal than when anyone can sign up
unattended. The two overlays are additive — they stack on top of whichever base you picked,
never replace it — and carry a distinct violet identity throughout this page, the same convention
automation, operate, quality and data use for
their own additive overlays.
Ask an agent whether a campaign worked and it reads the number the platform reported. Nobody checked whether those conversions were incremental — whether they would have happened anyway. That question is the one the field mostly avoids, and answering it honestly is this pack's flagship job. Platform-reported conversions answer "what fired last," never "what was incremental." The gap between the two is not a rounding error. In eBay's own large-scale paid-search field experiment, naive estimation put ROI above +4,100% uncontrolled and above +1,400% with time and geo controls; switching the ads off in randomized markets measured −63% ROI (95% CI [−124%, −3%]) — a sign flip, not a margin. The mechanism generalizes past search ads: people clicking high-intent ads were mostly going to buy anyway, which is why retargeting is the format most likely to overstate its own incrementality.
| Method | What it answers | What it cannot |
|---|---|---|
| Platform-reported conversions | Which touchpoint fired last before a tracked conversion | Whether the conversion was incremental — would it have happened anyway |
| Multi-touch attribution | A credit split across observed touchpoints | Causality; blind to unobserved, offline or organic touchpoints; degrades under consent gating and ATT |
| Self-reported attribution | Directional channel awareness, very cheaply | Precision — it carries recall and social-desirability bias, in a known direction |
| Holdout / conversion lift | The true incremental effect of exposure vs. a matched control | Anything at all without real statistical power — see the ceiling below |
| Geo-lift | Incrementality at market level; survives ATT and cookie loss with no user tracking | A trustworthy null without enough independent geo units and a stable outcome metric |
| Marketing-mix modelling | Cross-channel contribution and diminishing-returns curves over the longer run | Causal ground truth on its own; and it cannot forecast — Google's Meridian docs say it is "designed for causal inference, not prediction" |
MMM has the same shape of gate from the other direction: Robyn (Meta, MIT) documents roughly two years of weekly data at a 1:10 variable-to-observation ratio; Meridian (Google, Apache-2.0) says two years for a geo-level model is "too low to estimate the model reliably" and recommends three. Neither names a dollar threshold — the gate is data history, which is why MMM is unreachable for an early-stage product regardless of ad spend.
And a number can be incremental and still be the wrong thing to optimize. A
campaign that shows a clean incrementality result while only harvesting buyers already in-market
has passed a measurement test without answering the growth question. Reading an incrementality
result alongside "did this reach buyers new to the category" — connecting measurement mechanics to
the brand-growth evidence in marketing-science — is a connection the wider genre
leaves open, and this pack's job.
Almost every individual fact in this pack exists somewhere else. What does not exist anywhere else is a pack that grades those facts and connects them to each other — so the wedge is named as adjudication and integration, not discovery.
The canon mixes decades-replicated law, self-selected award-entry corpora, and arithmetic heuristics with a disclaimer attached. They are not cited the same way here.
Store limits, spam thresholds, crawler policies, supported rich results and platform guidelines all change with no deploy on your side. The pack teaches the check, not the list — and it never names a number in order to forbid it.
These govern every route, whichever references it loads.
Copy is delivered as copy. Requirements are delivered as requirements. Neither is delivered as production code. A marketing artifact is incomplete unless it carries the audience and buying situation it addresses, the one goal, every claim with a named proof source, the evidence tier and domain boundary of any law it cites, an as-of date on every volatile fact, the consent basis if anything gets sent, the measurement method and what that method cannot answer, and the sibling packs it hands to.
The five components filled as craft, not science — competitive alternatives, unique attributes, the value they enable, who cares most, and the market frame the product is set in.
What the page must claim and prove — the one goal, the objections, the proof devices — handed to design to compose against. Never a layout.
The launch as its own event — launch ≠ release ≠ deploy — with its tier chosen deliberately and the platform rules that are actually enforceable named before the day.
The UTM taxonomy, event and conversion definitions, and attribution window that data implements — plus the method's confidence level and what it cannot answer.
The consent basis per jurisdiction, decided before the copy is written, alongside the sender-authentication baseline and transactional/marketing stream separation.
marketing operates independently when invoked alone, and uses compatible upstream artifacts
without silently overriding them. It is rarely terminal: handoff.md maps every seam
between this pack and the rest of the family from marketing's side — what it produces for each
sibling, what it consumes, and the line it does not cross.
audience: <buying situation> one_goal: <what this makes happen> claims: [claim, named proof source, evidence tier, domain boundary] volatile_facts: [fact, as_of, re_verify_check] consent_basis: <if anything is sent> measurement: method + what it cannot answer
Landing-page requirements, copy, E-E-A-T content requirements and the structured-data
requirement plus entity map. design composes the page — its surface-website
product-narrative beats stay canonical there; frontend implements. Marketing
writes no production code.
The attribution and measurement requirements — what needs tracking and why. Marketing does not build or certify the pipeline and writes none of the SQL that computes attribution; it owns the requirements spec.
A qualified hypothesis for a funnel experiment. The design and the readout are
growth's regardless of who proposed the test — marketing never reports a growth
experiment as its own finding.
The AARRR note. A practitioner who knows acquisition → activation → retention → referral → revenue will notice this pack covers acquisition and stops. That is a deliberate seam, not a gap: AARRR spans three packs — marketing owns acquisition, growth owns the experiments that improve conversion across the funnel, success owns retention. Read the boundaries as where responsibility for each letter moved, not as territory this pack failed to claim.
The rest of the family — each an independently installable pack with its own guide:
Install once. It's a plain SKILL.md router — no flags, no config, no scripts — so it
activates on natural-language phrasing ("is any of this ROAS actually incremental," "write the
positioning for this before I build the page," "is this list lawful to email in the EU") rather
than a fixed command.
The same install runs on any Agent Skills
host. Codex installs to ${CODEX_HOME:-$HOME/.codex}/skills and triggers with
$marketing; agents remains a separate cross-agent installation target.
| host | install target | command |
|---|---|---|
| Claude Code | ~/.claude/skills | ./install.sh claude |
| Codex | ${CODEX_HOME:-$HOME/.codex}/skills | ./install.sh codex |
| Cross-agent path | ~/.agents/skills | ./install.sh agents |
| Cursor CLI | ~/.cursor/skills | ./install.sh cursor |
| Antigravity (IDE + agy) | ~/.gemini/…/skills | ./install.sh antigravity |
| opencode | ~/.config/opencode/skills | ./install.sh opencode |
| Grok Build | ~/.grok/skills | ./install.sh grok |
| Hermes | ~/.hermes/skills | ./install.sh hermes |
Prefer npx skills add gabros20/marketing-skill when you have Node — it maps supported clients itself.
More docs: docs/installation.md · docs/usage.md · docs/recipes.md · SOURCES.md — attribution, licences, and what this pack claims for itself.