💡CTF's most-shared Facebook post has nothing to do with the Foundation's work. Shares are bought with generic self-help content — the people sharing cannot tell anyone who CTF is.CTF 最多人轉發嗰條 FB 帖,同基金會做嘅嘢完全冇關係。轉發嗰班人根本講唔出基金會做緊乜,所以 273 個讚永遠變唔到 followers。基金會分享量最高的帖文與其工作完全無關,分享者無法說明基金會的角色,故未能轉化為追蹤者。
➡P0Rebuild Facebook as the follower engine: 328 → 2,000 page likes in 90 days, using content that names CTF's own work.FB 專頁重做追蹤者引擎:90 日內 328 → 2,000 個 like,內容一定要講得出基金會自己做緊乜。將 Facebook 重建為追蹤者引擎:90 日內由 328 增至 2,000,內容須明確呈現基金會工作。
02 — DIAGNOSIS診斷診斷
Two platforms. Two completely different illnesses.
兩個平台,兩個完全唔同嘅病。
兩個平台,兩種截然不同的問題。
Instagram is a conversion problem — traffic arrives and leaves. Facebook is an existence problem — there is no audience to convert. One calendar cannot fix both.
Reel reach multiple ×0.79, highest of all four. But 95% of posts get zero comments and 84% hide the like count. Reach arrives; nothing converts.Reel 觸及倍數 ×0.79,四間最高。但 95% 帖零留言、84% 帖收埋 like 數。流量到咗,冇嘢轉化到。Reel 觸及倍數 ×0.79,四者最高;惟 95% 帖文零留言、84% 隱藏讚好數。
👥 Facebook — 存在問題
Shares per post 7.44 (2nd of four) on a page with 328 likes. The content travels; there is no base for it to travel from.每帖分享 7.44(四間第二),但個專頁得 328 個 like。內容行得郁,但冇個基座俾佢出發。每帖分享 7.44(四者第二),惟專頁僅 328 個讚好,缺乏基礎受眾。
🧵 Threads — 空白
Nobody in this set has an account, including CTF. Text-first, zero production cost, and the only channel where arriving first still counts.四間全部冇開,包括 CTF。文字為主、零製作成本,亦係呢個界別唯一仲有「做第一個」價值嘅渠道。四者均未開設,包括基金會。以文字為主、零製作成本,亦為唯一先行者仍具價值的渠道。
💡CTF already wins the metric that is hardest to buy (reach) and loses the one that is easiest (audience size). Most brands have the opposite problem — so this is a distribution fix, not a creative reinvention.CTF 已經贏咗最難買嘅嘢(觸及),輸咗最易買嘅嘢(追蹤者基數)。大部分品牌啱啱相反——即係話呢個係分發問題,唔使推倒重來換 creative。基金會已在最難取得的指標勝出,卻於最易累積者落後。此屬分發問題,毋須重構創意方向。
➡P0Split the brief: Facebook builds audience, Instagram converts it. Never run one calendar across both.拆開兩份 brief:Facebook 起底層,Instagram 做轉化。唔好用同一條 calendar 打兩個平台。分拆為兩份簡報:Facebook 建立受眾,Instagram 專注轉化,不宜共用排期。
💡Community Chest posted 28 times in 7 months and still leads comments per post (6.46). On Facebook, conversation is driven by substance, not volume. CTF posts 39 times and gets 0.72.公益金 7 個月得 28 帖,每帖留言 6.46 仍然全場第一。喺 FB,帶對話嘅唔係出帖量,係內容有冇實料。CTF 出 39 帖,每帖得 0.72。公益金七個月僅 28 帖,每帖留言 6.46 仍居首。於 Facebook,帶動對話者為內容實質而非發帖量。
➡P0Stop measuring the page by post volume. One Q3 target: 328 → 2,000 page likes, driven by two substance posts a week.FB 唔好再用出帖量做 KPI。Q3 淨係盯一個數:328 → 2,000,用每週兩條有實料嘅帖推。不應以發帖量為指標。第三季只設一個目標:328 增至 2,000,以每週兩則實質帖文推動。
💡Po Leung Kuk gets 78 comments to our 2 — 39× — and not because they reach further (×0.23 vs our ×0.79). They post 3× as often and always give people a named person and a date to reply to.保良局攞到 78 個留言,CTF 得 2 個,爭 39 倍——但唔係因為佢行得遠(×0.23 vs 我哋 ×0.79)。佢出帖多 3 倍,而且每次都畀到一個有名有姓嘅人同一個日期俾人回應。保良局取得 78 則留言而我方僅 2 則,非因觸及較廣,而是發帖頻率高三倍且每次均提供具名人物與明確日期。
➡P1Lift cadence 1.39 → 3.0/week and Reel share 16% → 50%. Unhide like counts — hidden social proof blocks follower conversion.出帖 1.39 → 3.0 帖/週,Reel 比例 16% → 50%。同埋解除隱藏 like 數。發帖頻率提升至每週 3 則,Reel 佔比提升至 50%,並解除讚好數隱藏。
05 — WHAT ACTUALLY TRAVELLED真正跑出嘅帖真正突圍的帖文
Every top post here has a character and a deadline.
呢個市場每一條跑出嘅帖,都有一個角色同一個死線。
此市場每則突圍帖文,皆具備一個角色與一個期限。
#
Account帳號帳號
Result成績成績
Comments留言留言
Type格式格式
Core reason it worked核心原因核心原因
URL
🥇
東華三院
108,127 播放
0
Reel
Mystery question hook + original animated characters; no logo in the first frames懸念問句「香港驚現神秘戰隊?」+自創動畫角色,開頭唔見 logo懸念式問句+自創動畫角色,開首未見標誌
The subject carried it (AI robots), not the Foundation. No character, no date, nothing to reply to.係題材(AI 機械人)帶流量,唔係基金會。冇角色、冇日期、冇嘢可以回應。由題材帶動而非機構本身;無角色、無日期、無可回應之處。
💡Not one top post is a group photo or a programme title card. Every one gives the viewer someone to follow and something happening on a date. CTF's calendar is built almost entirely from the two formats that never appear here.冇一條跑出嘅帖係大合照或者計劃標題卡。每一條都畀到一個可以跟嘅人同一件有日期會發生嘅事。而 CTF 條 calendar,幾乎全部係榜上從來冇出現過嗰兩種格式。榜上並無任何合照或計劃標題卡。每則皆提供可追隨的人物及具日期的事件,而基金會排期幾乎全由此榜未見的兩種格式構成。
➡P1Rewrite the content brief around one rule: every post names a person and states a date.重寫 content brief,淨係守一條規矩:每條帖都要有一個有名嘅人+一個日期。重寫內容簡報,僅設一條規則:每則帖文均須具名人物並標示日期。
06 — WINNING FORMULA A致勝公式 A致勝公式 A
Formula A — Mystery question × owned character
公式 A — 懸念問句 × 自家角色
公式 A — 懸念問句 × 自有角色
One Tung Wah Reel did 108,127 views — 16.9× its own followers and 64× its own median Reel. The single biggest organic result in this market.
"A mysterious squad has appeared in Hong Kong?" opens a loop. "Flag day starts today" closes one.「香港驚現神秘戰隊?」開咗個 loop 俾人想睇落去。「賣旗日開始喇」係即刻閂咗個 loop。「香港驚現神秘戰隊?」形成開放懸念;「賣旗日展開」則即時終結懸念。
2 · 自家角色,唔使藝人費
Four illustrated heroes with donation bags, composited over the real hospital gate. No human face at all — and no talent fee.四個披斗篷、拎住捐款袋嘅插畫英雄,合成喺真實嘅東華醫院大門上。全條片冇一個真人面孔,亦唔使畀藝人費。四位插畫角色手持捐款袋合成於實景牌樓之上。全片並無真人面孔,亦無藝人費用。
3 · Logo 押後
Branding rides on the characters' own crest, not a lock-up in frame one. Viewers are 3 seconds in before they know it is an ask.品牌靠角色身上嘅院徽帶出,唔係第一格打個 logo。觀眾睇咗三秒先知係籌款。品牌識別由角色身上院徽承載,觀眾約三秒後方察覺為勸捐內容。
⚠ 修正上一版講法We previously said the travelling format needs "a human face in first person". Opening this Reel disproves it — there is no human face in it. The driver is a question hook plus a recurring character; a human face is only one way to supply that.上一版講「有人臉+第一身」先行得遠。開咗呢條片先知講錯——佢完全冇人臉。真正驅動力係「問句 hook +一個會重複出現嘅角色」,人臉只係其中一種做法。前版指須「具真人面孔並採第一人稱」。實際查閱後可見全片並無真人面孔,關鍵在於問句開場加可重複出現的角色。
💡A character is cheaper than a celebrity and does not expire when a campaign ends. Tung Wah proved it once, then stopped — their next 6 Reels averaged under 2,200 views.一個角色平過請藝人,而且 campaign 完咗都唔會過期。東華證明咗一次就冇再做——之後 6 條 Reel 平均唔夠 2,200 播放。角色成本低於藝人,且不隨活動結束而失效。東華僅驗證一次即停止,其後六則 Reel 平均不足 2,200 次播放。
➡P1Commission one owned CTF character and open every Reel with a question. Test 4 Reels in 30 days against a 3,000-view floor.整一個 CTF 自家角色(插畫或吉祥物),每條 Reel 用問句開場。30 日試 4 條,最低線 3,000 播放。建立自有角色,每則 Reel 以問句開場;三十日內測試四則,門檻 3,000 次播放。
07 — WINNING FORMULA B致勝公式 B致勝公式 B
Formula B — Named person × date × venue, released as a series
公式 B — 具名人物 × 日期 × 地點,一集一集咁出
公式 B — 具名人物 × 日期 × 場地,以系列形式發布
The only formula in this market that reliably produces comments. Po Leung Kuk earned 78 IG comments in 6 months; CTF earned 2.
全市場唯一穩定攞到留言嘅公式。保良局 6 個月攞 78 個 IG 留言,CTF 得 2 個。
市場中唯一能穩定產生留言的公式。保良局六個月取得 78 則,基金會僅 2 則。
1 · 一帖一個名
Aska on 20 Jul, Chantel on 22 Jul, Jeffrey in the lineup carousel. Each artist brings their own fanbase into the comments.7月20日 Aska、7月22日 Chantel、Jeffrey 喺全陣容 carousel。每個藝人自己帶班粉絲入嚟留言。7月20日 Aska、7月22日 Chantel、Jeffrey 見於陣容輪播,各自帶入粉絲群留言。
2 · 行動印喺畫面上
14 Aug 2026 · 8:15pm · Hung Hom Coliseum · $1,800 / $1,200 / $580 — burned into the frame, not buried in the caption.14 AUG 2026 · 8:15PM · 紅磡體育館 · $1,800/$1,200/$580——直接燒喺畫面,唔係匿喺 caption 尾。日期、時間、場地、票價直接標示於畫面,而非置於文案末端。
3 · 連載,所以會累積
Same template, a new face each time. Their Reel share is 58% — highest of the four — because the format is repeatable on a schedule.同一個 template,逐次換一個面。佢哋 Reel 佔比 58% 全場最高,因為呢個格式可以照 schedule 重複做。沿用同一模板逐次更換人物;Reel 佔比達 58%,為四者最高。
💡Within the same account and the same month, a named-artist Reel did 10,326 views and an unnamed one did 809. That rules out audience and algorithm — the variable is the named human.喺同一個帳號、同一個月,有藝人名嗰條 10,326 播放,冇名嗰條 809。即係唔關 audience 同 algorithm 事——變數係「有冇一個叫得出名嘅人」。同一帳號、同一月份內,具名藝人 Reel 錄得 10,326 次播放,無具名者僅 809 次,可排除受眾與演算法因素。
➡P0CTF does not need celebrities. Serialise named beneficiaries and named frontline staff — same template, one name per episode, fixed weekly slot.CTF 唔使請藝人。將有名有姓嘅受助人同前線同事連載——同一 template、一集一個名、固定每週一個時段。基金會毋須藝人:將具名受助人及前線同事系列化,沿用同一模板,每集一位人物,固定每週檔期。
08 — WINNING FORMULA C致勝公式 C致勝公式 C
Formula C — Exact dollar figures × quoted beneficiaries
公式 C — 確實銀碼 × 受助人真實引述
公式 C — 確實金額 × 受助人真實引述
The plainest-looking post in this entire study produced the most conversation: 731 reactions, 102 comments, 46 shares.
成個研究入面最「唔靚」嗰張圖,就係帶起最多對話嗰張:731 讚、102 留言、46 分享。
本研究中設計最樸素的帖文,正是帶動最多討論者:731 讚好、102 留言、46 分享。
1 · 一張表,唔係一個感覺
HK$20,000 × 1,945 戶 · HK$50,000 × 168 名死者家屬 · HK$57,000 × 1,984 戶 —— printed as a table on the image itself.用表格印晒喺張圖上面。以表格形式印於圖上。
2 · 受助人親口講
Two quoted residents at the foot of the image — real words, attributed. That is what people reply to.圖底兩段受助居民引述——真人真話、有署名。人哋就係回應呢啲嘢。圖底載有兩位受助居民引述,具署名之真實表述。
3 · 明寫零行政費
「善款不扣除任何行政費用,全數用於幫助宏福苑家庭」——the trust claim is quantified, not implied.信任呢樣嘢係量化咗,唔係靠感覺暗示。信任訊息經量化陳述,而非隱含暗示。
⚠ 呢個證據證到乜、證唔到乜The Facebook and Instagram posts cover the same relief programme but use different creative (a data table vs a calligraphy still). So this shows Facebook is where Community Chest's conversation lives — it does not prove the same asset would fail on Instagram.FB 同 IG 兩條講緊同一個援助計劃,但用咗唔同 creative(數據表 vs 書法靜物)。所以只證明到「公益金嘅對話喺 FB」,證唔到同一張圖擺上 IG 就會死。兩則帖文涉同一計劃但素材不同,僅可說明討論集中於 Facebook,未能證明同一素材於 Instagram 必然失效。
💡CTF's Facebook page is the reverse profile: high shares, near-zero comments (0.72/post). Shares come from design polish; comments come from verifiable numbers people can argue with.CTF 個 FB 專頁啱啱相反:分享高、留言接近零(每帖 0.72)。分享係靠靚設計買返嚟,留言係靠可以攞出嚟拗嘅實數。基金會 Facebook 專頁情況相反:分享高而留言近乎零。分享源於設計質感,留言源於可供討論的具體數字。
➡P1Publish one "where the money went" table per quarter: amount, recipients, cost ratio, plus two attributed beneficiary quotes.每季出一張「啲錢去咗邊」表:金額、受惠人數、行政費比例,加兩段有署名嘅受助人引述。每季發布一份「款項流向」表:金額、受惠人數、行政費比例,並附兩段具署名引述。
09 — THE LOSING PATTERN失分模式失分模式
Same account, same six months. Best Reel to worst Reel: 6.7×.
同一個帳號、同一個六個月。最好嗰條 Reel 對最差嗰條:6.7 倍。
同一帳號、同一六個月區間,最佳與最差 Reel 相差 6.7 倍。
Same followers, same algorithm, same season. Only the format changed — which removes audience size as an excuse.
💡CTF's best post worked because the topic carried it, not the Foundation. That is a rented result — it cannot be scheduled and it builds no equity. Every formula on slides 6–8 builds something the account owns.CTF 最好嗰條之所以得,係因為題材夠勁,唔係因為基金會。呢啲係「租返嚟」嘅成績——排唔到 schedule,亦累積唔到資產。第 6–8 版三個公式,每個都幫個帳號累積自己嘅嘢。基金會表現最佳的帖文成效源於題材而非機構本身,屬借來的成果,無法納入排期亦不累積資產。
➡P0Retire two formats immediately: the group photo and the programme title card. They make up most of the current calendar and appear on no top-post list.即刻停用兩種格式:大合照同計劃標題卡。呢兩種佔咗而家 calendar 大部分,但喺任何排行榜都冇出現過。即時停用兩種格式:合照及計劃標題卡,二者佔現行排期大部分卻未見於任何排行榜。
Hook 「香港驚現神秘戰隊?」 GapOne-hit. 108,127 views once, then the next 6 Reels averaged under 2,200. Reels are only 12% of their feed.一炮過。爆完之後 6 條 Reel 平均唔夠 2,200。Reel 得成個 feed 嘅 12%。單次成功;其後六則 Reel 平均不足 2,200,Reel 僅佔 12%。
保良局
Hook 「[藝人] 約定你 8月14日紅館見」 GapEntirely rented from one concert. After 14 Aug the series has no second season and no owned asset.全部租自一個演唱會。8月14日一過,個系列冇第二季,亦冇留低自家資產。全數依附單一演唱會;8月14日後並無第二季,亦未留存自有資產。
香港公益金
Hook 「撥出港幣 5,000 萬元」 GapStrongest substance, weakest distribution — 0.75 posts/week and one Reel in six months.內容最有實料,分發最弱——每週 0.75 帖,六個月得一條 Reel。內容實質最強而分發最弱——每週 0.75 帖,六個月僅一則 Reel。
🎯💡 OUR WHITE SPACE我哋嘅白地我方空白位 Not one of the three runs the same named human across consecutive weeks. Every winning post is a one-off tied to an event. CTF already holds the scarcest ingredient — reach that escapes its own follower base (×0.79, best in market). Point that reach at a recurring cast and the followers follow.三間對手冇一間做到「同一個有名有姓嘅人,一星期跟一星期咁出」。每條贏嘅帖都係綁住一個活動嘅一次性嘢。CTF 手上已經有最稀有嗰樣——衝得出自己 follower 基數嘅觸及(×0.79,全市場最高)。將呢啲觸及指向一組固定出場嘅人,followers 就會跟住嚟。三間對手均未做到「同一具名人物連續數週出現」。基金會已具備最稀缺條件——突破自身追蹤者基數的觸及(×0.79,全市場最高);將此觸及導向固定人物組合,追蹤者自會累積。
💡Threads is text-first — exactly where CTF is weakest (0.72 comments per FB post) and cheapest to fix: no shoot, no design, no media buy.Threads 係文字行先,正正係 CTF 最弱(每 FB 帖 0.72 留言)但又最平易補嘅位——唔使開機、唔使設計、唔使買 media。Threads 以文字為主,正是基金會最弱而修補成本最低之處,無需拍攝、設計或媒體採購。
➡P2Register the handle this week (defensive, zero cost). Repost the weekly series as text + one still; target 30 replies in 60 days.今個星期就註冊個 handle(防守性、零成本)。將每週連載改做文字+一張圖再出;60 日目標 30 個回覆。本週登記帳號(防禦性、零成本),將每週連載改以文字加單圖發布;六十日目標 30 則回覆。
12 — A NECESSARY WARNING一個必要提醒一項必要提醒
More followers than us. Fewer views than us.
追蹤者多過我哋。播放少過我哋。
追蹤者多於我方,播放量卻低於我方。
Community Chest has 56% more Instagram followers than CTF and its Reel got 631 views — against our median of 2,302. A follower count is not an audience.
💡We are pursuing followers because leadership asks for them, and that is legitimate. But a follower target on its own can be hit while reach collapses — Community Chest is the live proof, same category, this year.我哋追 followers 係因為老闆要,呢個要求本身冇問題。但淨係追 followers,個數字升到嘅同時觸及可以插水。香港公益金就係同一界別、今年嘅活生生例子。追求追蹤者源於管理層要求,此要求合理;惟單獨追逐該指標,數字上升同時觸及可能急跌。
➡P0Keep followers as the headline KPI — leadership's number stays on top — but pair every follower report with reach multiple, so growth is never bought at the cost of reach.Followers 照做頭號 KPI——老闆嗰個數照擺頭位——但每次交 follower 數,一定同時交觸及倍數,確保個增長唔係用觸及換返嚟。追蹤者維持首要指標,惟每次匯報須同時呈報觸及倍數,確保增長並非以觸及為代價。
13 — THE PLAN執行計劃執行計劃
One cast. One slot. Ninety days.
一組卡士、一個固定時段、九十日。
一組人物、一個固定檔期、九十日。
Every action below is derived from a formula proven on slides 6–8 by a competitor in this market this year.
💡CTF has already produced a first-person documentary Reel that did 12,642 views (Dec 2025 — outside the window, shown as capability proof, not as a benchmark). The capability exists in-house; only the schedule is missing.CTF 自己出過一條第一身紀錄式 Reel,做咗 12,642 播放(2025年12月,喺 6 個月窗口以外,擺喺度係證明做得到,唔係當 benchmark)。能力本身有,差嘅只係一條 schedule。基金會曾發布第一人稱紀實 Reel,錄得 12,642 次播放(2025年12月,區間之外,僅作能力佐證)。內部能力具備,所缺者為固定排期。
➡P0Week 1 deliverables: 12 names signed off, Tuesday slot locked in the calendar, Facebook brief rewritten.第一週要交嘅嘢:12 個名嘅卡士名單批咗、星期二時段入咗 calendar、FB 專頁 brief 重寫完。首週交付項目:十二人名單獲批、星期二檔期納入排期表、Facebook 簡報完成重寫。
14 — HOW THIS WAS MEASURED呢啲數點嚟數據來源與方法
Every number here can be re-run. Here is exactly how.
⚠ 本版已修正上一版兩個講法(1) The viral format does not require a human face — slide 6. (2) The Community Chest cross-platform comparison uses different creative on each platform, so it cannot prove a platform effect — slide 8. Both were caught by opening the actual creative.(1) 爆片格式唔一定要有人臉——第 6 版。(2) 公益金跨平台對比兩邊 creative 唔同,證唔到 platform 效應——第 8 版。兩個都係開返實物素材先捉到。(1) 廣傳格式毋須真人面孔(第六頁)。(2) 公益金跨平台比較之素材不同,未能證明平台效應(第八頁)。二者皆經查閱實際素材後發現。
💡Stating limits up front is the point: if we hide that CTF's ER rests on 3 visible posts, the client finds it later and stops trusting the whole deck.主動講限制先係重點:如果收埋「CTF 個 ER 只靠 3 條可見帖」,客第日自己揭到,成份 deck 都會唔信。主動說明限制方為關鍵:若隱瞞互動率僅基於三則可見帖文,客戶日後發現將動搖對整份文件的信任。
➡P2Re-run the identical scrape on 2026-08-27 for the first true month-over-month growth-rate comparison against these baselines.2026-08-27 用同一套方法再掃一次,就有第一組真正嘅按月增長率對比。於 2026-08-27 以相同方法重新抓取,即可取得首組真實的按月增長率對比。
💡Reach multiple and engagement rate answer different questions. ER can rise while an account shrinks; reach multiple cannot. Read both, always in that order.觸及倍數同互動率答緊兩條唔同問題。一個帳號縮緊嘅時候 ER 一樣可以升,但觸及倍數唔會呃你。兩個都要睇,而且照呢個次序。觸及倍數與互動率回答不同問題。帳號萎縮時互動率仍可上升,觸及倍數則否。
➡P3Circulate this glossary with the monthly report so every stakeholder reads the same number the same way.每月報告一齊派埋呢頁,等所有 stakeholder 睇同一個數嘅時候理解一致。每月報告一併附上本頁,確保各持份者對同一數字的理解一致。