# Linkable Asset > Linkable-Asset Skill — data studies that pull links (Lesson 7) - Skill: `lionelndong/linkable-asset` (Agent Skill) - Install (CLI): `npx skillmds add lionelndong/linkable-asset` - Raw SKILL.md: https://api.skillmd.com/api/skills/lionelndong/linkable-asset/raw - Safety review: pending (external: skill-scanner PASS, skillspector PASS) - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: AI & ML - Author: lionelndong (https://skillmd.com/u/lionelndong) - Updated: 2026-08-19 - Page: https://skillmd.com/skills/lionelndong/linkable-asset --- # Linkable-Asset Skill — data studies that pull links (Lesson 7) We are a small, young domain (DR 21). Lesson 7 of the course is blunt: **don't expect natural links while you're small — you have to build linkable assets.** The single best asset a product company owns is **its own data**. Almost nobody links to a product page; lots of people link to *"X% of users do Y"* statistics, charts, and trend reports. So this skill converts our aggregate PostHog + Stripe numbers into a **DATA STUDY** that journalists, bloggers, and roundup writers cite — each citation a referring domain — and then **points that new authority at our money clusters** so the link-juice does real work. See `STRATEGY.md` Lesson 7 + the off-page note (§"How the strategy maps to the pipeline"). > **What "winning" looks like:** the study earns referring domains we didn't have to ask for, our > **Domain Rating climbs**, and because the study **internally links down** to keystone + money/product > pages, that DR lifts the whole cluster's rankings — `max_targetable_kd` rises and harder keywords > unlock (`STRATEGY.md` §2). A data study that gets links but links to *nothing of ours* is a near-miss: > the authority leaks instead of compounding. **Links earned + DR up + juice routed to clusters = win.** This is **NOT** a normal blog article. It skips the keyword-cluster pipeline (it's an *asset*, not a cluster member) but it MUST still clear the uniqueness gate (`STRATEGY.md` §5 — our data IS the information-gain element) and carry our brand. It runs only in **Phase P4 / off-page**, after the core blogs are proven and traffic is rising. --- ## Two HARD RULES (read before anything else) ### RULE 1 — AGGREGATE ONLY. NEVER PII. (non-negotiable) Every number in the study is a **count, rate, average, or trend over a population** — never a fact about an identifiable person. This is what makes our data brand-safe and publishable at all: a mainstream site can cite *"43% of AI-companion users chat daily"* and never touch a real customer. We are an **adult brand** — leaking even one user's behavior is both a privacy breach and a reputational landmine. There is no "small exception." If a metric could be tied back to one person, it does not ship. **SAFE (aggregate — publish these):** - `"68% of users who create a companion send a message within the first hour."` - `"Average companion conversation runs 14 messages."` - `"NSFW image generations grew 3.2× quarter-over-quarter."` - `"Voice replies are used by 1 in 4 active companions."` - `"Weekend sign-ups run 22% higher than weekday."` **UNSAFE (PII / re-identifiable — NEVER, no matter how 'interesting'):** - ❌ Any email, username, display name, user id, `distinct_id`, Stripe `cus_…`/`sub_…`/charge id, IP, or device id. - ❌ Any **single-person** row, quote, transcript snippet, or "one user did X" anecdote. - ❌ **Small-cohort** stats that re-identify: a bucket of < **100 people** (k-anonymity floor), or "100% / 0% of [tiny segment]", or a combination of filters (city + plan + day) that narrows to a handful. Suppress or widen the bucket instead. - ❌ **Free-text** properties (message bodies, prompts, support replies) — even "aggregated," text leaks. Only count/bucket structured fields; never surface the words. - ❌ Geographic precision finer than country, or any timestamp precise enough to single someone out. **How this skill enforces it:** - The scaffold (`scripts/data_study_draft.py`) only ever runs **`count()` / `avg()` / ratio** HogQL — it never `SELECT`s a person field, and it applies a **minimum-cohort floor** (default `MIN_COHORT = 100`): any metric whose denominator is below the floor is **dropped and listed as suppressed**, not published. - Stripe figures used in the study are **aggregate totals/rates already computed** (e.g. "paid conversion rate", "MRR growth %") — pulled from the scorecard / finance aggregates, never raw customer rows. **Mask anything that smells like an id** (`cus_…`, emails) on sight; if you can't aggregate it, you can't use it. (Mirrors the cockpit PII-redaction doctrine.) - The reviewer (you / the operator) does a final **PII sweep** before approval — see the checklist. ### RULE 2 — APPROVAL-GATED. DRAFT → OPERATOR APPROVES → PUBLISH. NEVER AUTO. This skill **produces a draft and stops.** It never publishes, never pushes to Strapi, never calls `/format-for-publish --auto-publish`, never opens outreach. A data study makes a public, quotable claim about our company under our brand — that is a board-level statement, so a human signs off. The flow is exactly: ``` 1. /linkable-asset → pulls SAFE aggregates, writes a DRAFT study + a data appendix. (scaffold: scripts/data_study_draft.py) 2. OPERATOR REVIEW → reads the draft, runs the PII sweep, sanity-checks every number against source, confirms the internal-link routing. APPROVES or edits. 3. (only after approval) → the normal publish path runs as a DRAFT in Strapi for a final human read (/format-for-publish WITHOUT --auto-publish), then the operator clicks Publish. Promotion/outreach is a SEPARATE approved step. ``` If you are running autonomously, you **stop at step 1** and surface the draft path + a one-line "ready for operator approval" — you do **not** advance to step 2 or 3 yourself. --- ## The linkbait dimensions (Lesson 7 — what makes an asset link-worthy), mapped to Contagious STEPPS Pick the angle that our data can actually support. A study is strongest when it hits **two or more**. Lesson 7 names four linkbait dimensions; each maps onto a **Contagious / STEPPS** principle (`/contagious-why-things-catch-on`, Jonah Berger) — and STEPPS is the engineering manual for *why* each one earns a share/link. **The data study is our strongest linkbait, so engineer it explicitly for shareability:** | Dimension (Lesson 7) | STEPPS principle | What it is | Our data study version | |---|---|---|---| | **Numbers** | **Social currency** | A concrete, quotable statistic people repeat — and sharing an exclusive number makes the sharer look in-the-know | "The state of AI companions 2026: N data points from M conversations" — the headline stat IS the link. Our **exclusive** data is pure social currency: only we have it, so citing us signals insider knowledge. | | **Utility** | **Practical value** | "News you can use" — a benchmark/reference others measure themselves against | "AI companion engagement benchmarks" — what 'normal' usage looks like, so writers cite us as the yardstick. Most brand-safe, most-cited shape. | | **Emotion** | **Emotion** (high-arousal) | A surprising / counter-intuitive finding — arousal (surprise, awe), not mere positivity, drives sharing | a trend that defies the assumption (e.g. "voice adoption outpaced text faster than anyone expected") — with the real number behind it. Lead with the *gap* between expectation and reality. | | **Stories** | **Stories** (Trojan Horse) | A narrative the data tells over time, with our brand **integral** to the telling | "How AI-companion behavior shifted across 2026" — a trend line, not a snapshot; build it so the story can't be told without naming Pleasur.AI as the source (valuable virality, not a "fool in the pool"). | **Engineer the share, don't hope for it (STEPPS moves to apply on top of the dimension):** - **Inner remarkability + Social currency** — find the one genuinely *surprising* number in the data and make it the headline; the more remarkable, the more it gets repeated. An exclusive stat only we hold is the strongest social currency a product company has. - **Practical value** — frame the study as a **benchmark/yardstick** readers and writers can act on and measure against; package it scannable (`:::stat` / `:::stat-group`) so the useful bit is effortless to lift and cite. - **Brand-integral story** — if the angle is a trend, make Pleasur.AI inseparable from the finding so every citation carries the brand (valuable virality). **Numbers + Utility (Social currency + Practical value)** is our bread-and-butter: a *benchmark report* ("here's what typical engagement looks like") is the most-cited, most brand-safe shape — it's useful, it's quotable, and it positions Pleasur.AI as the category's data authority without a single PII risk. --- ## Internal-link routing — the juice MUST flow DOWN to our clusters (Lesson 7) A linkable asset that earns links but keeps the authority to itself is wasted. The entire point is to **funnel the new DR down to the pages that make money.** Every study MUST link to: 1. **The relevant cluster keystone(s)** — read `content-pipeline/0-keywords/cluster-map.md` for the live keystone slug per cluster. A companions data study links down to the `companions` keystone; an image-gen study to the `image-gen` keystone; etc. (clusters: `companions`, `image-gen`, `chat-roleplay`, `voice-calls`, `tools-compare` — see `clusters.md`). 2. **The money / product page** the cluster showcases (the product the data is *about* — link the feature page so a reader who came for the stat can convert). 3. **2–4 supporting articles** in the same cluster that deepen a finding (descriptive anchors, not "click here"). Anchors are **descriptive and natural** (`our AI companion creator`, `NSFW image generation`), never keyword-stuffed. This is the same internal-linking discipline `format-for-publish` applies — the study is just an unusually high-authority source page, so its outbound internal links matter more than most. > Mechanism: external sites link to the **study** (it's safe + quotable) → the study passes that > authority **internally** to the keystone + money pages → those pages climb. That's how a brand that > can't get links *to its product* still grows the product's rankings. --- ## Input `/linkable-asset [cluster-id]` (optional cluster focus; defaults to the highest-data cluster). Reads: - `content-pipeline/scorecard/snapshots/.json` — already-aggregated per-article + conversion data (a safe, PII-free source the scorecard produced). - `content-pipeline/scorecard/latest.md` — aggregate conversion/compounding context. - `content-pipeline/0-keywords/cluster-map.md` — live keystone slug per cluster (for the routing). - `content-pipeline/0-keywords/cache/brand-dr.json` — our live DR (the asset's *reason for existing*: raise this; frame the study's ambition around it). - `examples/authors.md` — the byline (a data study is **Sloane Avery**, the Analyst / data-benchmark persona — `persona: sloane-avery`). - live aggregates via the scaffold (PostHog `count()`/`avg()` only; Stripe aggregates only). ## Process 1. **Pick the angle + cluster.** Choose a linkbait dimension (above) our data genuinely supports; pick the cluster the study is *about* (so routing has a home). Prefer **Numbers + Utility** (a benchmark) unless a real, surprising trend justifies Emotion/Stories. 2. **Pull SAFE aggregates only.** Run the scaffold: ```bash doppler run -- python scripts/data_study_draft.py [cluster-id] ``` It emits a draft outline + a **data appendix** of aggregate metrics (counts/rates/trends), with every below-floor cohort **suppressed and listed**. It does **not** publish anything. 3. **PII sweep (RULE 1).** Read every number. Confirm each is population-level, each cohort ≥ the floor, no field, anecdote, or free-text could identify a person. Drop or widen anything borderline. If in doubt, cut it — a thinner study that's unimpeachable beats a richer one that leaks. 4. **Write the study around the data.** Lead with the headline stat (the thing people will quote). Structure: hook → top-line findings (the quotable numbers, in `:::stat` / `:::stat-group`) → trend(s) → what it means → **methodology disclosure** (`:::methodology`: source = our own product analytics, sample size, date range, definitions, the aggregation + privacy note "no individual user data; all figures are aggregate counts/rates over ≥N users"). Byline Sloane Avery. 5. **Route the juice (Lesson 7).** Add the internal links DOWN to the keystone(s) + money/product page + 2–4 supporting articles (descriptive anchors). This is a required section, not optional. 6. **Uniqueness check (`STRATEGY.md` §5).** Our data is the information-gain element — confirm the study says something the SERP doesn't already have (it does, by construction: it's *our* numbers). 7. **STOP — hand to the operator (RULE 2).** Save the draft + appendix; surface the path and "ready for operator approval." Do **not** publish, push, or promote. ## Output `content-pipeline/linkable-assets//` - `draft.md` — the data-study draft (byline comment + H1 + `:::stat`/`:::methodology` + the internal-link routing section). A DRAFT — not a publish package. - `data-appendix.md` — the aggregate metrics the draft is built on (every figure population-level; suppressed cohorts listed) + the methodology/privacy note, for the operator's sanity-check. - `APPROVAL.md` — the gate record: the PII-sweep checklist (unticked), the routing targets, and the explicit "operator must approve before publish" line. Publishing before this is signed is a process violation. ## DO / DON'T **DO** - Lead with **one quotable number** — the headline stat is the link magnet. - Aggregate everything; state the **sample size + date range + definitions** openly (credibility = links). - Frame it as a **benchmark / reference** others measure against (Utility) — the most-cited shape. - **Link DOWN** to the keystone + money page + supporting ring (route the juice — Lesson 7). - Keep it **on-brand** (Sloane Avery byline; Pleasur.AI named as the source of the data). - **Stop at draft** and let the operator approve. **DON'T** - ❌ Surface ANY individual user/customer datum, id, email, free-text, or small-cohort (< floor) stat. - ❌ Auto-publish, push to Strapi live, or kick off outreach — ever. Drafts only; operator gate. - ❌ Treat it like a keyword article (it's an asset) — but DON'T skip uniqueness or the brand. - ❌ Earn links to a page that **links to nothing of ours** (authority leak — route it down). - ❌ Inflate, cherry-pick, or round dishonestly — a debunked stat is a reputational + SEO loss (`STRATEGY.md` anti-patterns: don't chase the spike; don't ship what you can't stand behind). - ❌ Expect this to work before the core blogs are proven (Lesson 7 is sequenced into Phase P4). ## Quality checklist - [ ] **Every** figure is aggregate (count/rate/avg/trend over a population) — zero individual data. - [ ] **No PII** anywhere: no email/username/user-id/`distinct_id`/Stripe-id/IP/device, no free-text, no single-person anecdote, nothing finer than country / coarse date. - [ ] **Every cohort ≥ the minimum floor** (default 100); below-floor metrics suppressed + listed, not shipped. - [ ] Headline carries a **quotable Numbers/Utility stat**; ≥ 2 linkbait dimensions hit. - [ ] **Engineered for shareability via STEPPS (`/contagious-why-things-catch-on`): the surprising/exclusive number is the headline (Social currency + inner remarkability), the study reads as a citable benchmark/yardstick (Practical value), and — if a trend — the brand is integral to the story. Name which STEPPS principle(s) the study is built to hit.** - [ ] `:::methodology` discloses source (our product analytics), sample size, date range, definitions, and the explicit **aggregate-only / no-PII** note. - [ ] **Internal-link routing present:** links DOWN to the cluster keystone(s) + money/product page + 2–4 supporting articles, descriptive anchors (Lesson 7 juice flow). - [ ] Byline is **Sloane Avery** (`persona: sloane-avery`) — the data-benchmark persona. - [ ] Passes uniqueness (`STRATEGY.md` §5) — our data is the information-gain element. - [ ] Output is a **DRAFT + appendix + APPROVAL.md** — nothing was published, pushed, or promoted. - [ ] `APPROVAL.md` states the operator must approve before any publish (RULE 2). ## Discipline - **Aggregate or it doesn't ship.** When a number can't be made population-level without re-identifying someone, it's cut. The credibility of the whole asset depends on us never being the brand that leaked its users (`STRATEGY.md` — we refuse reputational losses dressed up as growth). - **The operator is the publish gate, by design.** This skill's job ends at a reviewed-ready draft. Auto-publishing a claim about our company is exactly the kind of high-blast-radius action that stops for a human (it mirrors the cockpit's guardrail doctrine). - **An asset that doesn't route juice is a miss.** The reason we build it is DR → clusters. If the draft doesn't link down to money/keystone pages, it isn't done (Lesson 7). - **Phase-gated.** Linkable assets belong to the off-page phase (P4), after the blogs prove out — not a thing we do while the core clusters are still being seeded (`STRATEGY.md` Lesson 8 / off-page note).