short-form-what-works

name

short-form-what-works

description

Evidence-based rules for what actually moves short-form video performance (TikTok, Reels, Shorts), distilled from Layers Research's 2026 TikTok Posting Report (23M videos, 13.3T views). Use this skill whenever the user asks about posting strategy mechanics — best time to post, posting frequency, how many posts per day, spacing, taking breaks, video length, slideshows vs. videos, caption length, hashtags, resolution/aspect ratio, deleting flops, shadowbans, follower counts vs. reach, or what to post after a viral hit or a flop. Also trigger when the user repeats posting folklore ("post at 6pm", "don't post twice a day", "keep it under 30 seconds", "the algorithm punishes X") so their plan can be checked against the data, or when auditing a posting schedule or content calendar.

Short-Form: What Actually Works

Data-backed posting rules from Layers Research's 2026 TikTok Posting Report — 21 pieces of industry folklore tested against 23 million TikTok videos and 13.3 trillion views. Every finding is a within-creator comparison (each video ranked only against the same account's other posts), so audience size and niche are controlled away. Use this to overrule folklore with effect sizes. Pairs with the shortform-content skill (Kallaway's Lego Brick framework): that skill covers how to make the video; this one covers everything around it.

Units: effects are in "spots" — percentile points within a creator's own catalog. Near the middle, +4 spots ≈ +14% views, +10 ≈ +42%, +20 ≈ double. Payouts are jackpot-shaped: a 90th-percentile post earns ~8× your median; a 99th, ~60×.

The headline

When you post is noise. What you post is signal. Timing plus account features predicts a creator's hits at AUC 0.532 (barely above coin-flip); reading the content lifts it to 0.574. All 168 hour-of-week slots combined explain 0.007% of performance. Redirect every minute of calendar optimization into the content itself — especially the first two seconds.

The levers, ranked by effect size

Actual levers, largest first (view-equivalents for a typical mid-catalog post):

1. Make a sequel to a hit: +9 spots (~+36%). The same-topic, same-length follow-up to a hit beats a change-of-subject follow-up by ~9 spots — the biggest what-to-post-next edge measured. Post it within 4 days; the window fades after a week. (Caveat: sequels tend to follow the strongest hits.)
2. Optimize for saves: top vs. bottom save-rate decile = 18.5 spots (~2× views). Saves separate winners from losers at twice the strength of shares; likes barely move the needle; comment-heavy engagement predicts lower rank. Hits get shared 1.7× more per viewer, liked the same, commented less. Make content worth bookmarking and sending — a loud comment section is not demand.
3. Go longer: 5-min+ beats sub-8-second by 6.1 spots (~+24%). Monotonic — every step up in length helps, in all 13 niches tested. No penalty for length exists anywhere in the data. Association, not causation (longer usually means higher effort), but "keep it short" is the most expensive myth tested. Let length serve the idea.
4. Shoot true vertical 9:16: +4 spots (~+14%) over square/landscape. Letterboxed reposts from other platforms pay this tax every time.
5. No hard-sell captions: −3.6 spots (~−12%). Captions pushing a link or "DM me" run below the same creator's baseline. Sell in the video; keep the caption for the hook.
6. Upload 1080p: +2 spots (~+7%) over sub-720p. 4K adds nothing over 1080p.
7. Post more. Each post on a 3-post day ranks ~3 spots lower, but the day's total reach is ~3.3× a one-post day — cannibalization doesn't appear at the level that matters. The per-post tax is fully paid by the fourth post; total reach keeps climbing (~6× at 5 posts, ~12.6× at 10). No fatigue curve over sustained heavy runs.
8. Spread volume across the week. Same 6–8 weekly posts crammed into 1–2 days lose ~11% of the week's reach vs. spread over 5+ days. Cadence (day-level) matters; clock-level spacing doesn't — the whole spacing spectrum is worth ~4 spots, and the "wait a few hours between posts" rule specifically buys ~1.
9. Write the caption you want. Caption length is worth ~2 spots — in favor of 500+ character essays. "Long captions hurt reach" is backwards.

The busted myths (stop optimizing these)

• Best hour of day: best vs. worst hour = 0.86 spots; under the strictest age controls, statistically zero. Your "personal best hour" from analytics doesn't replicate out-of-sample — it holds up for 51% of creators, a coin flip. The "post when competition is low" mechanism is dead too: a 2.7× swing in upload supply moves outcomes not at all.
• Best day of week: 0.32 spots across the whole week. The largest niche effect anywhere (travel leans weekend) is under one spot.
• Timing matters once you're big: true for 616 of 189,718 creators, and even there it explains 1.6% of variance.
• Scheduling tools suppress reach: machine-scheduled posts run ahead of the same creators' manual posts — because creators put their best, most-prepared videos into the scheduler, not because the software helps.
• Don't post more than once a day / rapid-fire chokes reach: see levers 7–8. The burst tax (~2.6 spots) only hits accounts for whom the burst is out of character, and the first post of a multi-post day is reliably the weakest — the warm-up lap, not the race.
• A break kills the account: the comeback post after 30+ days lands at your exact average, and posts 2–5 after it climb past your baseline. The account doesn't need to recover; it needs to warm up. Take the break.
• The post-viral hangover: inverted. The post after a hit typically beats two of three videos in your own catalog. Hot and cold streaks are real and large — the hit marks the streak, it doesn't make it (creators climb ~4 spots over the five posts before a hit). No need to sprint: the follow-up lands about the same 2 hours or 4 days later.
• The algorithm pushes slideshows: slideshows run ~3 spots behind the same account's videos. Slide count is worth less than one spot — use as many slides as the idea needs.
• Stay in your lane: off-niche posts cost less than half a spot. The lane-switching penalty doesn't exist; branch out as much as feels right.
• You need followers for views: the feed follows interest, not follower counts. Sub-1k accounts' median video reaches 105% of their follower count; million-plus accounts, 1.4%.
• Delete videos that don't pop: nobody actually does this (<2% of deletions happen in week one), and deleted videos sit dead-average, not bottom. No evidence pruning helps.
• Hollow virality gets punished: ~1 in 10 big hits is a "ghost hit" (huge reach, dead engagement). The next post after one lands slightly higher than after a real hit.
• The 200-view jail: no wall, no cap — just a smooth distribution that happens to peak in the low hundreds.

Nuances worth knowing

• Virality is a loaded lottery. The top 1% of a creator's posts carries 45% of lifetime views (top 10% carries 60%) — but the odds aren't random: after a hit, your next post beats 2 of 3 of your own videos; after a bottom-quartile flop, 1 of 3. You can't schedule your way into the jackpot, but you can print better tickets: volume, length, saves-worthy content, sequels.
• After a flop, near follow-ups run cold and waiting doesn't fix it — a week later you're still below average, and a month off adds its own ~2-spot warm-up toll. Post through it.
• The platform is drifting long: the median TikTok doubled in length 2020→2025 (15s → 34s), captions grew ~50%, and visibly machine-scheduled posting grew 28-fold. "Short-form" is aging fast.
• Untestable — treat confident claims as guesses: hashtag counts, trending vs. original sound, shadowbans, delete-and-repost, first-hour engagement windows. The report couldn't measure these honestly; neither can the gurus.

Applying it

When reviewing a posting plan or calendar:

1. Strike any effort spent choosing hours, days, or "optimal windows" — reallocate it to hooks and content quality.
2. Check format basics: 9:16 vertical, 1080p source, video over slideshow, no hard-sell caption.
3. Push volume up and spread it across the week; don't fear multi-post days, breaks, or off-niche experiments.
4. Ask what the sequel plan is: when something hits, the same-topic same-length follow-up within 4 days is the single best next move.
5. Judge past content by saves and shares per viewer, not likes or comments.