YouTube Shortsの成長とアルゴリズム習得

# Understanding YouTube Shorts' Recommendation Algorithm YouTube Shorts operates on a **separate recommendation pipeline** from long-form videos 、 it's designed to maximize **session satisfaction** drastically, not content depth. The algorithm's core question is: *"How quickly will a viewer swipe away?"* ### How the Algorithm Processes Shorts - **Entry Pool:** When you upload a Short, YouTube shows it to a small test audience of roughly 50–500 viewers (usually your existing subscribers + a random sample of users with similar viewing habits to your subscriber base). - **Primary Signal:** The **swipe-away rate** in the first 1–3 seconds is the strongest signal. If more than ~25% swipe away instantly, the video is capped at the initial pool and gets buried. - **Secondary Signals:** Among those who watch past 3 seconds, YouTube measures: - **Full-view percentage** (does the viewer reach >90% of the video?) - **Re-watch loops** (does the viewer watch it again immediately?) - **Like/comment/share velocity** relative to the view count. - **Returning viewer** behavior: does a viewer come back to your channel *after* seeing one Short? - **The "Related" Loop:** YouTube sends your Shorts to subscribers of channels similar to yours (or creators whose content your audience follows). If they watch-through, the algorithm assumes viewer-channel resonance and expands to a broader **look-alike audience**. - **Time-Decay Factor:** Shorts get a sharp traffic spike in the first 4–7 days. After that, the algorithm mostly stops testing unless the video triggered "evergreen" signals (e.g., continued shares or comments weeks later). Long-tail boost happens rarely. **Key takeaway:** Shorts aren't about *content depth* 、 they're about **matching a viewer's impulse to keep scrolling** with content that stops their thumb. --- # Growth Strategy: Views, Subscribers, Engagement, and Returning Viewers ## 1. The Hook Framework (Compensate to the First 3 Seconds) Your Short must **break the scroll pattern** within 3 seconds. Build hooks syntactically: - **Visual anomaly:** Start mid-action, upside-down, zoomed-in, or with a sudden color shift. The viewer doesn't process what it is 、 they stop to figure it out. - **Open-loop curiosity:** State a payoff outcome in the first line: *"Most people fail this because they don't know the hidden setting."* - **Extrinsic contradiction:** *"This device is only $12 yet outperforms $300 gear."* - **Pattern interrupt:** Use a freezing effect, loud sound, or a rapid zoom to simulate a glitch 、 viewers reorient to see where it goes. Write the first 30 characters (which the reader sees in the feed without opening) to *complete the curiosity gap* 、 e.g., *"The worst thing about me is…"* → *"…my superpower."* --- ## 2. The "Retention Hematology" Approach for 30s–60s Shorts Shorts under 30 – 45 seconds suffer from *immediate swipe risk* if they lack close pacing. For retention, use the **"Segment Loop" framework**: - Segment 1 (0–3s): Hook / high-intent promise. - Segment 2 (3–15s): Surf the hook with **3 micro-payoffs** 、 each ~4s long, each with its own mini-cliffhanger (acts as embedded retention). - Segment 3 (15s–end): Deliver the final payload *and* set up **a transition loop** to the beginning of the video (if possible) 、 a frame that visually or textually overlaps the intro, encouraging a rewatch loop (requirement 、 loop = full-view + rewatch = algorithm heat). **Rules for Segment 3:** - Never end with a fizzle. End with a *whisper* of a new question: *"…that's why I check the filter settings next time."* - Merge the final CTA ("Follow for part 2") *inside* the last 2 seconds 、 don't do it at the very end of the video (removes natural afterglow). --- ## 3. Testing Framework: "Batch & Triage" Build tests with **controlled variables**. Create batches of **3 identical Shorts** that differ only in the hook or the title. For example: | Short A | Short B | Short C | |-------|-------|-------| | Hook: open with text "10 Secret Tools" | Hook: open with "I stop using these……" | Hook: open with dramatic action + no text | - **Hypothesis:** The 👉's curiosity level, or the specificity, affects swipe-away rate. - **Test variable:** Only the *first 3 seconds* (keep all other footage identical). - **Run each with 24h timer and compare:** Swipe-away %, AVD, full-view %. **Rule of thumb:** If a variant performs 15%+ better than your channel average, *iterate on it* in a follow-up batch (second test = changing pacing, sound, loop ending). **Weekly framework (per 7-day rotation):** - Day 1–2: release the three variants. - Day 3: analyze retention, choose the winner. - Day 4–6: produce one **derivative** of the winner (same hook, changed mid-section). - Day 7: release the derivative + start the next batch. --- ## 4. Metrics Dashboard 、 What to Measure & How Only 5 metrics are significant for Shorts: | Metric | Where to see | What it tells you | Action to take | |--------|-------------|-----------------|----------------| | **Swipe away rate** (within 1st 3s) | In Studio → Retention graph (top of drop) | Hook strength | Rewrite / re-cut the first 3s | | **Viewed vs. Served** | Studio → Reach: "Percentage viewed" | Whether your retention is strong | If >60%, push more length; if <40%, re-cut to shorter | | **Full-view ratio** (% of viewers who reach 90%–100%) | Retention graph | Content pacing | If >30%, you're in the algorithm's sweet spot | | **Returning viewers** | Audience view in last 7 days | Your share of repeat consumption | Create series/sequels to sustain that percentage | | **Subscription conversion** | Studio → How viewers found you | CTA/reconnect power | Refine your end-loop/OC frame CTA | **Bonus Interpretations:** If returning-viewer rate is <10% but views are high 、 you have viral *reach* but no channel gravity. Make content *continuous*: a series of numeric or visual codes that return viewers expect (e.g., *"Part 1"* label). --- ## 5. Framework for Analyzing "What Makes a Successful Short" (Patterns) Run a **"pattern autopsy"** every 2 weeks for all your Shorts over 7+ views: - **Retention fingerprint:** Enhance top 3 successful Shorts 、 did they have a *story twist* on the last second? Did they start in motion? - **Audience convergence:** Do your top Shorts share a *common emotional theme*? (E.g. "creator in awe of an object".) Replicate the *emotion*, not the context. - **Time slot:** Compare your top 5 Shorts to your bottom 5 、 in what lifecycle age *both* [destination: (e.g., Monday 1pm]; Sunday evening)? - **Length rule:** Are all top Shorts *under 25s* vs. your flops over 40s? Usually, Shorts >45s suffer from *length penalty* if they don't have macro-sensory pacing. **Patterns to look for:** 1. **Rapid a leads to loop** 、 top Shorts often have a visual/audio echo at the end so viewers replay it (conclusive nod to looping). 2. **The first word isolated** 、 every top 10% Short uses a *normal void* (no background noise) in the first 300ms to make signal tighten. 3. **Thumbnail-less vs. Cover Frame** 、 Shorts that *don't* use a cover don't matter here, but the frame *at* the start occupies the feed visually; Premium frames = higher tap-through. --- ## 6. Strategy to Increase Returning Viewers (Channel Gravity) Returning viewers are driven by *series contiguousness*. Use: - **"Chapter Reset" Technique** 、 each Short ends with: *"Want me to unlock switch? Watch the NEXT to see how…"* 、 viewers must return to *know the outcome*. This is pure **anticipation tagging**. - **Short Mafia pattern**: Every 3rd Short is a *sequel* that opens with a brief *recap in 1 second* (previous clip) 、 creates a "strings" your audience recognizes. - **Maintain a publishing cadence:** Post 3–5 Shorts/week at the *same* hours (train viewer habits through feed forecasting). - **Track "Returning Viewers" (%)** 、 aim to grow from 5% → 20% by *releasing short-serial episodes*. --- ## 7-Day Action Plan (Concrete) | Day | Action | |-----|--------| | Day 1 | **Diagnose:** Open your Studio, download 30-day data. Find your top 3 Shorts (by retention, not views). | | Day 2 | **Batch #1**: Create 3 hooks in 1 week using one of your successful topics. Post all 3 same day (distant 1–2h). | | Day 3 | Analyze early retention curves. Pick winner → make 1 remake where *only* flow bursts differs (e.g., swap end-screen CTA). | | Day 4 | **Upload** that one (don't publish). Wait 24h. | | Day 5 | Publish the winner + the remake. Compare both to your baseline. | | Day 6 | **Optimize** your channel's homepage → Add "Series" playlist for the loop pattern. | | Day 7 | Export cumulative 7-day stats from Studio → create a "topic/format" matrix. | --- ## Final Rule (Lock-In) **The Shorts algorithm rewards *very specific* behavior**: It wants to show a viewer posts that retain them, get them to go back to the feed, and ultimately build a habit. Focus all testing on *cyclic loops* rather than random creativity. Everything triggers from the **hook**, sustainable *growth comes from the series*. As you scale, treat Shorts as a **feedback console**: test – measure – amplify – loop. In 4 weeks, your returning viewers should rise, subscriptions increase, and the algorithm will push your material beyond the initial pool.