The first 72 hours: how delivery pacing changes the FYP weight of a new post

Two posts with identical content, posted by the same account in the same week, can land at wildly different reach. The difference is rarely the content. It is almost always the shape of the early engagement curve the ranker sees in the first 72 hours.

What the FYP ranker actually scores

TikTok's recommendation system is described in their "How TikTok recommends content" newsroom post as a multi-signal scoring model. For new posts in 2026, three weights dominate every other input.

1. Engagement velocity inside the probe window

The ranker watches the shape of the engagement curve in the first 200 to 500 impressions, not the raw count. A curve that resembles natural discovery scores higher than a higher-count curve that looks vertical or flat.

• Posts that gain 30 likes spread across the first 20 minutes consistently outperform posts that gain 200 likes in the first 60 seconds.
• The threshold is curve-shape, not volume.
• Field tests put the velocity-spike filter at roughly 6× the median like-rate of the seed audience.

2. Completion rate, weighted by video length

A 9-second post that 78% of viewers finish outscores a 30-second post that 45% of viewers finish, even if the second has higher absolute watch-seconds.

• Short videos under 15 seconds are scored mostly on completion percentage.
• Mid videos 15 to 30 seconds blend completion and absolute watch-seconds.
• Long videos over 30 seconds shift more weight onto absolute watch-time.

3. Re-engagement signals

Shares, saves, profile visits, and rewatch loops carry more weight than likes because they are harder to fake at scale.

• Saves are the strongest single signal — they trigger phase-three re-surfacing days after the initial post.
• Shares weight roughly 4× a like in the ranker's internal model based on our delivery-data inference.
• Profile visits in the first 30 minutes also carry weight, because they indicate the viewer wants to see more from the same creator.

The three windows inside the 72-hour scoring period

The 72-hour scoring window is not a single phase. It is three distinct phases that score different signals.

Phase 1 — The probe window (first 30 to 90 minutes)

TikTok pushes the post to a small seed audience of 200 to 500 users who have previously engaged with similar content. Completion rate and watch-time on this seed group determine whether the post graduates.

• For sub-15-second videos, the completion threshold sits around 35 to 45%.
• For 30-second videos, the average watch-time threshold is roughly 18 to 22 seconds.
• Posts that fail the probe rarely recover, regardless of later engagement.

Phase 2 — The widening test (hours 2 to 12)

Posts that clear the probe get pushed to rings of 5,000, then 50,000, then 200,000+ users. The ranker watches whether the early signal survives at scale.

• Larger audiences include viewers outside the creator's typical cohort.
• Watch-time matters more than completion rate at this scale.
• A sudden drop in completion rate at the 50,000-impression boundary triggers a distribution pullback.

Phase 3 — The compounding window (hours 12 to 72)

Surviving posts enter a slow-burn distribution where shares and saves carry more weight than fresh views.

• Saves trigger the FYP to re-surface the post days later.
• Shares expand the network of viewers exposed to the content.
• Posts that compound here are the ones that hit one million views.

What "natural early engagement" looks like to the ranker

Engagement that "looks natural" is engagement that mirrors how humans actually discover content.

• The first like usually arrives 8 to 15 seconds after the first view.
• Like density rises gradually for the first 5 to 10 minutes, then plateaus or accelerates depending on how the post performs in the probe.
• Saves and shares appear later than likes, usually 20 to 90 seconds after a viewer's first view, because they require a deliberate action.

Engagement that lands all at once, in volumes higher than the seed audience could plausibly generate, reads as inorganic. The velocity filter does not care whether the accounts are real. It cares about the shape of the curve.

Why flat engagement spikes hurt new posts

The biggest mistake we see in our delivery-network data is engagement front-loaded into the first 60 seconds of a post's life.

• 5,000 likes on a post that has been seen by 250 humans is the exact pattern the spike filter is built to catch.
• The ranker downweights the post's distribution score, often by 40 to 70%.
• Even if the engagement is technically from real accounts, the velocity mismatch overrides the account-quality signal.

For more on the underlying signals TikTok uses to rank content, the likes vs. views vs. follows breakdown covers how each engagement type is weighted differently.

How drip pacing fits the model

Step 1 — Pace inside the probe window

Likes and views should land in the first 30 to 90 minutes, distributed across small bursts rather than a single batch.

Step 2 — Keep velocity below the spike threshold

Throughput should stay below roughly 6× the median like-rate of the seed audience. For accounts under 100,000 followers, this typically means under 30 likes per minute in the first 10 minutes.

Step 3 — Layer in saves and shares later

Saves and shares carry more weight when they arrive in the 30 to 90 minute window, after the probe has already pushed the post wider.

Tokturbo's drip-paced TikTok likes and automatic TikTok views are built around this model. Pacing is the product.

What to ship in the first 72 hours

• Test the hook before scaling. If the first 1.5 seconds do not earn the next 5, the post will fail the probe regardless of pacing.
• Use auto-views for daily uploaders. The probe arrives whether you remembered to push the post or not.
• Watch the rewatch signal. Hooks that surprise on second viewing earn the highest re-engagement scores.
• Track save rate, not like rate. Saves predict phase-three distribution better than any other signal.

For the latest changes to the ranker, see our 2026 ranker patch notes. For the cold-start playbook on new accounts, see picking a niche that is not already cooked.

Frequently asked questions

Quick answers to the questions we get asked most about this topic.

How long does the FYP scoring window last?

Most aggressive scoring happens in the first 30 to 90 minutes after upload, but the ranker continues re-scoring the post for up to 72 hours. After that, organic discovery slows unless the post triggers a delayed-spike signal like a remix or a shoutout from a larger account.

Why is curve shape more important than volume?

The ranker's velocity filter is built to detect inorganic activity. A 5,000-like spike on a post seen by 250 viewers reads as inorganic, even if the accounts are technically real. A curve that grows in sync with view count reads as organic, even at much higher absolute counts.

Does drip-paced engagement actually work?

Yes, but only when the pacing matches the shape of organic discovery. A flat trickle of 5 likes per minute for an hour looks just as inorganic as a 5,000-like spike. The pacing has to mirror how viewers actually discover and engage with a post.

What is the single most important signal in the first 72 hours?

Save rate. Saves trigger the phase-three re-surfacing that drives most viral posts. A 4% or higher save rate inside the first 12 hours is the strongest predictor of compound distribution we have seen in our delivery data.