Story completion drops a median 14% per frame; the cliff is at frame 3
Most accounts post 5-7 frame sequences. The data says trim to 4.

— Frame 1→2: -7% viewers
— Frame 2→3: -11%
— Frame 3→4: -14% (the cliff)
— Frame 4→5: -19%

By frame 5, the median account retains only 58% of frame-1 viewers. Sequences ending at frame 4 carry the highest completion-rate percentile. The one exception: polls/quizzes on frame 2 cut per-frame drop roughly in half by resetting attention.

n=3,100 story sequences, 60-day window.

Worth your feed

@ReelsTrenchNotes. Field notes from someone posting Reels daily: what's working THIS week, the hooks… We read it, you probably should too.

Sub-2k accounts post a median reach rate of 34% — higher than any larger cohort
The inverse relationship between size and reach rate is consistent and underrated.

—

Carousels now out-reach Reels for accounts under 20k by a median 1.7x
The blanket "Reels win" advice inverts below a size threshold.

—

Save rate predicts 7-day reach with r=0.71; like rate predicts almost nothing
We ran correlations between each engagement signal and downstream reach. The hierarchy is clear.

— Saves → 7-day reach: r=0.71 (strongest)
— Shares/sends: r=0.64
— Comments: r=0.38
— Likes: r=0.11 (statistically near-noise)

Likes are a vanity signal the algorithm has largely discounted. Optimize content for the save and the send: reference-value posts (lists, frameworks, how-tos) and tag-a-friend hooks beat aesthetic likeable posts on distribution.

n=1,650 posts, Pearson correlation, 30-day window.

Reels need 50% average view-through to escape the follower bubble
There is a measurable retention gate before cold distribution kicks in.

—

Trending audio drives a median 22% of Reels reach — but only in a 5-day window
The audio page is a real discovery surface with a sharp expiry.

— Audio age 0-2 days (rising): ██████ 28% of reach via audio page
— 3-5 days (peak): ████ 19%
— 6-10 days (saturated): █ 6%
— 10+ days: ▏ 2%

By the time an audio is obviously trending, the window is closing. The edge is using sounds at 500-5,000 uses, not 500k. Track the use-count growth rate, not the absolute count — acceleration is the signal.

n=1,100 Reels, 45-day window.

The reach-loss signature people call "shadowban" is a sudden 60-80% hashtag-reach drop with feed reach intact
Most "shadowban" panic is a misread. The data shows a specific, separable pattern.

— Genuine restriction: non-follower reach -65% median, follower reach flat
— Content-fatigue dip: both fall proportionally ~20%
— Normal volatility: ±25% day-to-day, mean-reverts in 72 hr

If only your non-follower line collapses while followers still see you, it is a discoverability flag, usually from flagged hashtags, banned audio, or rapid automation. Audit those three before assuming a platform penalty.

n=140 flagged cases, controlled comparison, 30-day window.

Hashtags in the first comment vs the caption: zero measurable reach delta
The "hide hashtags in the first comment" tactic is a myth that survives on aesthetics, not data.

— Caption hashtags: median reach indexed 100
— First-comment hashtags: 99 (within noise)
— No hashtags at all: 96

Placement does nothing for distribution; it only affects how clean the caption looks. The real variable hiding in this debate is hashtag relevance — tightly topical tags beat broad high-volume ones by a median 14% in non-follower reach, regardless of where you put them.

n=520 posts, three-way split, 30-day window.

Neighbor spotlight: @AlgoHeretic. They go deep on TikTok organic — the kind of channel you actually keep notifications on for.

Алиса AI будет конкурировать с Google AI Studio

Яндекс разворачивает экосистему AI-агентов на базе Алисы с доступом сначала для компаний, затем для всех. Агенты уже работают в Яндекс Такси и Лавке, скоро появятся в браузере и студии разработки. Платформа интегрирует стандартные функции — заказ такси, покупки, анализ данных. Алиса AI показывает неплохие результаты: менее известна, чем конкуренты, поэтому предлагает щедрые лимиты на видеогенерацию и работу с контентом. Яндекс планирует внедрить…

➡️ Читайте на сайте: https://aff.top/blog/alisa-ai-budet-konkurirovat-s-google-ai-studio

🧠 Ещё больше инсайтов → в канале AFF.top

В Zennoposter добавили ИИ-помощник

Zennolab добавил в Zennoposter встроенный ИИ-кубик с доступом к четырём моделям (Gemini, DeepSeek, Claude, ChatGPT) — 50 бесплатных запросов в сутки. Есть режимы Assistant (чтение) и Agent (автоматическое создание скриптов), плюс новый GET-запрос по API. Нейросети хорошо справляются с регистрацией, постингом, фармингом аккаунтов и простым кодированием, но требуют проверки при парсинге динамических сайтов и диагностике ошибок. В связке с Zennoobr…

➡️ Читайте на сайте: https://aff.top/blog/v-zennoposter-dobavili-ii-pomoschnik

🧠 Ещё больше инсайтов → в канале AFF.top

Explore placement is gated by save velocity in the first 30 minutes, not total saves
It is rate, not volume. Two posts with identical final save counts diverge sharply on Explore reach.

— Top-quartile early save velocity: ███████ 38% of reach from Explore
— Median: ███ 12%
— Bottom quartile: ▏ 2%

The ranker reads early save acceleration as a proof-of-value spike and opens cold distribution. This is why your first-hour audience quality dominates everything downstream. A small, highly-engaged seed beats a large passive one for Explore entry.

n=900 posts, 45-day window.