Amber Huang | Brand growth at the intersection of content, creators, and AI

Amber Huang — Senior Influencer Marketing

SHANGHAI · 2026

02 — About

I ask what if a lot,
ship before I overthink,
and let the creators teach me
what brands don't see.

Senior Influencer Marketing at bilibili, working with creators to ship million-fan KOL IPs end-to-end — from positioning to monetization.

Trained at USC Annenberg / Digital Media in Los Angeles. Before bilibili, I ran APAC cold-start at TIKTAL (TikTok Official Partner) and led content growth at Dianping, where I scaled UGC submissions from 0 to 1,200 daily.

For the past 3 years I've lived in two worlds at once — by day a platform operator at bilibili & Dianping, by night a Rednote creator approached by beauty, fashion, and consumer brands. So I see KOL economics from both sides: I know where brand briefs fail, and how platform leverage actually flows.

Now building multi-agent AI workflows for enterprise — turning human signal into business growth.

Brand Social Strategy KOL Matrix Design UGC Growth Mechanic AI Workflow Builder Cross-Border Growth Creator Insider Lens

03 — Experience

Across three countries — LA to Shanghai.

Three years building creator-led growth programs across international and Chinese markets — from US campus communities to APAC key-account campaigns to large-scale Shanghai platform IPs.

N · Northern hemisphere

—

Bearing · live

04 — Cases

Three platforms. Three intelligences.

Deep dives into the brand × creator × AI plays I shipped on Rednote, TikTok, and bilibili — each ending with one key takeaway you can reuse.

Part 01 / 03 9:16

KOS Matrix · North America

Rednote × Think Academy

−38% lead cost · 2.4× enrollment

Open case →

Part 02 / 03 9:16

TIKTAL · North America

TikTok × PetSmart

CTR 8 → 20% · 30k UGC

Open case →

Part 03 / 03 9:16

CCTV Spring Gala · 2024

Bilibili × CCTV Live

1M+ live · Top 10 site-wide

Open case →

PART 01 / 03 REDNOTE

Rednote × Think Academy (北美学而思)

"Helping overseas Chinese parents trust a tutoring brand — on the platform they already live in."

1 Insight — what we saw

Overseas Chinese parents in the Bay Area don't pick a tutoring brand from ads. They pick it the way they pick a dentist — word from another mom they already follow on Rednote.

Think Academy's brand account was loud, but the audience was searching for "Bay Area school districts," "early STEM enrichment," "my third-grader is falling behind in math" — and finding a real Bay Area mom with 8,000 followers, not a brand.

↳ The problem wasn't making content. It was finding the right people to make content with us, as themselves.

2 Strategy — what we did

We built a three-layer KOL × KOS × KOC matrix around one persona pair: 湾区宝妈 × 码农老爸 — the personas who already had natural Rednote traffic in this audience.

01

KOL Layer

Sign 1k–50k follower Bay Area moms / engineer dads as long-term partners, not one-off ad buys. They post as themselves.

02

KOS Layer

Real teacher accounts — "Mr. Wang 在硅谷教数学" — talking pedagogy, lesson clips, parent Q&A. Building institutional trust through individual face.

03

KOC Layer · Closed-loop

Customer = creator. Parent users who enroll → become content authors, sharing real homework progress. Each lead also becomes future supply.

04

Brand Hub

Think Academy's overseas official account holds the matrix together — re-shares, official Q&A, conversion CTA. The matrix supplies the trust; the hub captures the demand.

AI-Native Layer · how I'd build it in 2026

The original 2023 build took ~200h of manual creator hunting on Rednote's search. With NoumenaAI-style growth intelligence:

Semantic discovery — Agent scans Rednote posts under "school districts / early STEM / bilingual parenting" for creators with authentic family voice (not just keyword match).
Authenticity scoring — AI evaluates content realness × parent-persona alignment × comment quality. Filters out obviously-sponsored accounts.
Personalized first-touch brief — Generated per-creator outreach referencing 3 of their actual posts. Reply rate 2–3× over generic DM.
Continuous KOS coaching — AI suggests next post angle for each signed teacher based on what's converting in the matrix.

→ 200h shrinks to ≤20h. Same matrix scales from 1 metro to 5. 客资增长上限 = AI 帮你拓的人，不是预算砸的位。

3 Outcome — what we got

Creator Pool Signed

80+

8 KOS teachers · 30 KOL parents · 40 KOC alumni

Lead Cost Reduction

−38%

vs paid-acquisition baseline · 120 posts / month sustained

Enrollment Lift

2.4×

KOL/KOS-attributed conversion vs control

4 Playbook — key takeaway

Reusable Lesson

Beauty sells with filters. Education sells with trust.
The leverage on Rednote isn't paid reach — it's signing the right creators long enough that your customers become your supply.

PART 02 / 03 TIKTOK · NORTH AMERICA

TIKTAL × PetSmart

"Turning a holiday SKU into a UGC production stage — and beating retail expectations 10 seconds at a time."

1 Insight — what we saw

PetSmart had a Christmas SKU — a DIY Cake Kit for pets — that retail merchandising treated as another shelf item. But on TikTok, "my dog's first birthday cake" was already a native creator behavior pattern.

The gap: brand was buying ads; the platform was waiting for a stage. The product itself was the most filmable thing in the catalog — but no one was helping creators see it that way.

↳ Don't ad-buy your way through. Build the stage and let TikTok's natural creator economy do the work.

2 Strategy — what we did

I led the North America campaign as direct report to CEO at TIKTAL (TikTok Official Partner). Designed a closed flywheel — physical activation feeding paid optimization feeding organic UGC.

01

Physical Stage

"10-Second Unboxing Studio" — pop-up booths in select PetSmart locations where pet parents could film their unboxing on-spot, with prop lighting + branded hashtag wall.

02

Hashtag Hook

#makemerrymemories — designed to be both shoppable trigger and emotional anchor. UGC turned every transaction into a content moment.

03

Ads-Manager Optimization

Used Segwise creative-level analytics to identify which 10-second clip patterns converted best, then re-fed them to fresh creator briefs within 48h.

04

UGC Flywheel

Each new UGC post became raw material for the next ad cycle. Brand-generated content became seed, user-generated content became fuel.

AI-Native Layer · how I'd build it in 2026

The 2023 version of this campaign required a 6-person team and 48h ad-iteration cycles. With AI growth intelligence:

Real-time creative scoring — Agent ranks every uploaded UGC clip by predicted CTR / save-rate before paid scaling.
Auto-brief generation — top-performing patterns feed back as personalized briefs to fresh creators within hours, not days.
Cross-platform replication — same insight library powers parallel campaigns on Reels / Rednote for the same SKU.

→ The flywheel I built manually becomes self-spinning. Same campaign mechanic, 10× faster iteration.

3 Outcome — what we got

CTR Lift

8 → 20%

Creative-level optimization across campaign

CPM Floor

$0.03

UGC-blended reach cost · 30k+ posts

Sales YoY

+60%

vs prior holiday season SKU

Reach · 100M+ impressions / Tier-1 Retail Case

4 Playbook — key takeaway

Reusable Lesson

Don't translate brand assets into UGC.
Find the moment where the product is already filmable, and build the stage there.

PART 03 / 03 BILIBILI

bilibili × CCTV Spring Gala

"Stealing 18 hours of national-broadcast adjacency — and turning it into a bilibili-native event."

1 Insight — what we saw

CCTV's Spring Festival Gala is China's most-watched media moment. bilibili had a once-a-year window of official partnership rights — but no playbook for converting that rights window into native bilibili energy.

The audience didn't want a televised broadcast re-streamed. They wanted their favorite Bilibili creators reacting alongside them — the way they watch every other big event on the platform.

↳ Don't compete with CCTV. Re-format the cultural moment as a bilibili live-stream ritual.

2 Strategy — what we did

I led "Watch the Gala with People Who Get You" — an 18-show live-stream series with 10 million-fan Bilibili creators reacting to the Gala in real time. Designed end-to-end as a Bilibili-native commentary event, not a broadcast simulcast.

01

Creator Casting

10 million-fan creators across diverse verticals — gaming, knowledge, lifestyle, music — so every audience segment had someone who gets them hosting their watch-party.

02

Show Architecture

18 live shows staggered through the Gala broadcast — pre-show prediction, in-show reaction, post-show debrief. Each block had a different host energy.

03

Brand Integration

Sold the format — not the slot. Brand sponsor got recurring presence across 18 shows woven into the creators' reaction format, not pre-roll ads.

04

Danmaku as KPI

Live-stream success on Bilibili = danmaku density (real-time scrolling comments), not just view count. Designed reaction beats to maximize danmaku response — the platform's native engagement signal.

AI-Native Layer · how I'd build it in 2026

Coordinating 18 simultaneous lives × 10 creators × 1 sponsor's brand consistency was 90% spreadsheet work. With AI:

Live danmaku analysis — Agent classifies sentiment + meme emergence in real time, surfacing which creator segments are spiking for next-show optimization.
Sponsor-fit scoring — AI ranks every creator × brand match across the catalog, suggesting which creator should anchor each ad-integrated beat.
Post-event narrative auto-draft — within 30 min of show close, AI drafts the 复盘 deck pulling top moments, danmaku winners, ROI signals.

→ Same event becomes a learning loop, not a one-night production. 下一年的春晚 IP 从 100 万级直接进 500 万级筹码。

3 Outcome — what we got

Live Viewers · Danmaku

1M+

Cumulative entries across 18 shows · 100k+ real-time comments

Site-Wide Rank

TOP10

48-hour site-wide hot list

Brand Sponsorship

7-figRMB

Single-cycle deal value

4 Playbook — key takeaway

Reusable Lesson

Don't sell the slot — sell the format.
When you have rights to a cultural moment, re-shape it through the platform's native ritual, not the broadcaster's habit.

05 — On Stage

Three rooms. One voice.

Speaking where decisions live — from USC Career Fair to industry summits to platform creator days.

USC Career Panel Los Angeles · 2023

Apex Summit Los Angeles · 2024

Build Your Creator World Shanghai · 2026

06 — Brands

Brands I've worked with & impacted.

Three years across platforms and KA clients — operating, advising, and shipping campaigns that moved the needle.

07 — AI Toolkit

How I leverage AI.

A growing collection of custom Claude Skills and vibe-coding projects that show how I approach marketing × engineering problems — not "I use AI" but "AI is my coworker."

📦 ai-toolkit · github →

giant-of-report

My Claude Code skill that coaches you through writing any professional report and turning it into AI-generated visuals.

My SkillReport

report-kaiju

My Cursor skill — sharper reports with Minto pyramid, award-grade references, and model routing.

My SkillCursor

influencer-event-ops-skill

My bilingual project-management toolkit for end-to-end influencer event operations.

My SkillOps

anthropics/claude-code

Claude Code CLI — my daily coding driver. Most of this site, the atlas demo, and every skill above ships through it.

AnthropicCLI

langchain-ai/langchain

Framework for building LLM-powered agents and pipelines — what I use to wire research and content automations.

Agents110k★

lobehub/lobe-chat

Open-source ChatGPT-style web app — my self-hosted chat for sensitive brand briefs and internal data.

Self-host60k★

n8n-io/n8n

Low-code workflow automation — connects Rednote, Notion, Feishu, and Claude into one campaign pipeline.

Workflow70k★

Vibe Coding Projects

Atlas · Interactive Globe

Five fanned case cards plus a half-dome globe that rotates to the active city — vanilla JavaScript, SVG, and CSS in a single HTML file, zero framework dependency. Drafted end-to-end with Claude in one evening; live in the Experience section above.

Open source · MIT

Feishu Projects

Built a Notion-AI clone inside Feishu Docs in roughly 1,000 lines of code, and wired Claude Code into Feishu Docs so you can type a prompt and get an inline AI edit. Both ship from a single workspace.

Open source · TypeScript

amberhuang.world

This portfolio — vanilla HTML, CSS, and JavaScript with Tailwind and custom SVG, built section by section in conversation with Claude Code. The fastest concept-to-shipped-design loop I have found.

Ship in two days

08 — Share Life & Insights