TikTok
TikTok's New Ranking Signals: Completion Rate, Saves, Shares, and Session Value Now Run the Algorithm
The algorithm you built your funnel on no longer exists — and the new one has four signals most creators are optimizing wrong.
Updated Jun 2026 · sourced from 15 YouTube creators and 7 operator groups
Key takeaways
- 50% watch-through is now the minimum threshold; full completions are weighted even higher.
- Saves and shares outrank likes and raw views in the retrained algorithm.
- Your follower list is now an algorithmic gatekeeper — dead followers kill reach twice.
- 'Session value' measures what viewers do AFTER your video, not during it.
- Q2 2026 is a transition window: diversify traffic now, don't bet the funnel on TikTok.
Your last TikTok hit 40K views and you felt good. Then the next three averaged 900.
You didn't change anything. That's not bad luck — that's what a platform retraining its algorithm from a blank slate feels like.
On January 22, 2026, Oracle took over TikTok's US data infrastructure through a joint venture called USDS. (SWCEO, May 2026) Every performance signal your account accumulated before that date has been discarded. (SWCEO, May 2026)
The algorithm is being rebuilt exclusively on US data running on Oracle servers, and it doesn't remember you. (SWCEO, May 2026)
Here's what it's learning to care about instead.
The Signal Hierarchy Has Flipped
The old TikTok rewarded views. Big numbers, big push.
That loop is broken.
Saves and shares now outweigh raw likes and views as algorithmic signals. (SWCEO, May 2026) Creators with strong save rates are outperforming those with higher view counts but lower save rates. (SWCEO, May 2026)
The implication is blunt: a video with 8K views and 400 saves beats one with 80K views and 12 saves.
Likes are not useless. They're just no longer the metric that scales your distribution.
The 50% Completion Wall
Completion rate is now the hardest bar in the system. Videos must reach at least 50% watch-through to be prioritized, and full completions carry even more weight. (SWCEO, May 2026)
Videos that lose viewers early get deprioritized faster under this system than under the old one. (SWCEO, May 2026)
The practical ceiling: a tight 25-second video with high replay rate beats a 45-second video with a weak open. (SWCEO, May 2026) The hook window is 2–3 seconds, maybe five at a stretch. (Patrick Mulroy, Jul 2025)
After that, the algorithm is already recording the drop-off.
For OFM content specifically, this compresses your format. The 60-second relationship monologue with a slow build?
It's dying in the first test batch. The 20-second slideshow that ends mid-thought and makes the viewer scroll back?
That's the format the new system rewards.
Loop Endings Are Not Optional
Every TikTok should have a loop-style ending that causes viewers to accidentally rewatch. (SWCEO, May 2026) This isn't a creative flourish — it's a mechanical signal.
A rewatch registers as near-full completion, which compounds your completion rate score.
Combine it with a tight 2-second hook and an explicit save/share CTA, and you're targeting all four primary retrained algorithm signals in a single video structure. (SWCEO, May 2026) 'Save this for the next time he goes cold on you' is a real CTA format. It works because it gives the viewer a reason to save that isn't 'please help my metrics.'
Session Value: The Signal Nobody's Explaining
Here's the one that trips creators up most.
TikTok measures something called 'session value' — but it's not how long someone watches your video. It's how long they stay on TikTok after watching your video. (SWCEO, May 2026)
If your content sends people to their camera roll, their DMs, or off-platform, that's a negative session signal regardless of how long they watched.
Specific optimization tactics for session value haven't been publicly detailed yet. (SWCEO, May 2026) But the directional logic is clear: content that makes viewers want to keep scrolling TikTok — not jump to your Instagram — is what the algorithm values.
This creates a genuine tension for OFM operators whose entire goal is to push people off TikTok and onto a funnel.
There is no clean resolution here. The evidence just flags the conflict.
What we can say: if your videos are conversion-optimized first and engagement-optimized second, your session value score is likely hurting your distribution before the funnel even gets a chance to work.
The Follower Test Phase Most Operators Are Ignoring
New content doesn't go to strangers first anymore. It goes to your followers. (SWCEO, May 2026)
The retrained algorithm shows each new video to a small test group of existing followers. If they don't engage quickly, the video stalls.
Full stop. (SWCEO, May 2026) A stale follower list penalizes you twice: it generates weak engagement signals, and it suppresses distribution before the video ever reaches cold audiences. (SWCEO, May 2026)
This is structurally different from the pre-Oracle system, and it changes what 'follower count' means. A big, dead follower list is now a liability.
The fix isn't to buy engagement — it's to earn it before your next post goes live. Reply to every comment.
DM new followers. Go live more often. (SWCEO, May 2026)
The goal is to make the top 10% of your fans fast-responders, because they are now the mechanical switch that decides whether your content reaches anyone else. (SWCEO, May 2026)
What the Volatility Actually Looks Like
Pull 60 days of your TikTok analytics. If you see a clear pattern of view drops and RPM volatility starting around January 2026, that's platform-level noise, not a content strategy failure. (SWCEO, May 2026)
Two Oracle data center outages in Q1 2026 — including a ~20-hour outage at the Ashburn, Virginia facility on March 3 — caused upload failures, inaccurate view counts, and content disappearing from feeds entirely. (SWCEO, May 2026) Some creators made content pivots based on data that was literally wrong.
In Q2 2026, creator payouts are running at a fraction of pre-Oracle levels for the same content type and view counts. (SWCEO, May 2026) The reach is unpredictable.
The conversion math for subscriber funnels is off. (SWCEO, May 2026) One operator group flagged in late May 2026 that TikTok's analytics and engagement metrics had broken that week — recommending operators hold strategy decisions until fixed.
This is the environment you're building in right now.
Where Operators Actually Disagree
The chatter is loud on several tactical questions, and it genuinely conflicts. Both sides are worth knowing.
Posting volume: One operator from a single group (April–May 2026) reports running up to 40–100 videos per week with no quality penalty, arguing TikTok's retention scrutiny is lower than Instagram's. (Gavin Magoon, May 2026) A conflicting voice — also in operator groups from the same period — says 1–3 videos per day per account is the safe ceiling, and logging in and out hourly is what triggers flags.
The vetted evidence lands in the middle: 3–5 times per week minimum, consistent, not sporadic. (SWCEO, May 2026)
Fresh vs. aged accounts: One group (April 2026) says fresh self-created accounts work fine and you don't need aged ones. A separate group (June 2026) says aged accounts are worth buying.
A third voice (mid-2024, so note the age of this data) argues self-creation via AWS Device Farm is more reliable than buying accounts because aged account suppliers are often scam-adjacent. (habibi, Jul 2024) No consensus.
Fresh accounts with proper warmup appear functional; aged accounts aren't a guaranteed advantage.
Accounts per device: One group (April–May 2026) reports running 5 TikTok accounts on one device without triggering spam detection. A separate group (also April 2026) recommends capping at 2–3 accounts per device and only running as many as you can afford to lose.
The conservative position is the safer bet given the ban environment.
VPNs vs. physical SIMs: Multiple groups conflict here. At least two separate groups (late 2025 to early 2026) say VPNs no longer work on TikTok and recommend real SIM cards on mobile data.
A different group from the same period recommends VPNs for US audience targeting. A third reports mobile proxies working roughly 80% of the time but notes cheap ISP proxies got pre-warmed accounts banned within weeks.
Physical devices with real SIM cards have the broadest corroboration across groups.
Content reposting: At least two distinct groups (late 2025 to mid-2026) warn that TikTok and Instagram now use AI to detect unoriginal content, with integrity ban waves hitting reposted reels. The same period produced operator suggestions to re-encode, mirror-flip, or use metadata-cleaning tools to evade detection.
The vetted evidence confirms that reposting the exact same file repeatedly crushes reach. (Gavin Magoon, Nov 2025) The evasion tactics are unverified operator chatter — treat them as experiments, not solutions.
The Audit You Should Do This Week
Pull your last 10 TikToks. For each one, record:
- Completion rate
- Save count (and save-to-view ratio)
- Share count
- Comments that indicate actual emotional response
Your weakest signal is the first thing to fix. (SWCEO, May 2026) One vetted source tracked these metrics and found save-to-view ratio was the strongest predictor of algorithmic push in Q1 2026, outperforming views and likes. (SWCEO, May 2026)
If completion rate is the weak link: shorten the video, tighten the first two seconds, cut every second of dead air.
If saves are the weak link: add an explicit, scenario-based save CTA. 'Come back to this tonight' is more effective than 'save this video.'
If shares are the weak link: the content isn't provocative or relatable enough to forward. Relationship tension, emotional vulnerability, and reaction-bait formats historically drive shares in OFM-adjacent content. (Patryk, May 2026)
The Honest Bottom Line
TikTok is still a viable top-of-funnel channel. (Gavin Magoon, May 2026) It is not the primary growth engine it was in 2022 or 2023, and multiple vetted sources and operators across this period agree on that trajectory. (TDM Business (OFM), Dec 2025) (Will Mammone, Sep 2025) (Will Mammone, Oct 2025)
The Oracle transition didn't kill TikTok — it reset it. That reset is an opportunity if you move now, before the algorithm settles and starts favoring whoever trained it first. (SWCEO, May 2026)
But the math has to be right. Do not make Q2 2026 business decisions that assume stable TikTok reach. (SWCEO, May 2026)
Diversify your traffic sources while the retraining phase is active. (SWCEO, May 2026) Build follower engagement as infrastructure, not an afterthought.
Treat session value as a real constraint even without full optimization clarity.
The signal hierarchy is completion rate, then saves and shares, then session value, then follower engagement velocity. That's the order.
Build your content structure around it — tight hook, tension body, loop ending, explicit save CTA — and you're targeting all of them at once. (SWCEO, May 2026)
Everything else is noise until mid-2026. Don't let volatile analytics convince you to rebuild a strategy that just needs to survive a transition.
Sources
On the record (YouTube creators):
- SWCEO — Why adult creators are losing reach on TikTok right now (and 3 moves to fix it fast), May 2026. Watch ↗
- SWCEO — EP 185: The TikTok Retraining Phase Explained for Adult Creators in 2026, May 2026. Watch ↗
- Patrick Mulroy — OnlyFans Creator QUALITY Guide: How to 10x Your OnlyFans Traffic by Mastering REAL Content Quality, Jul 2025. Watch ↗
- habibi — How I make Us Tiktok Accounts (without being in usa OFM), Jul 2024. Watch ↗
- TDM Business (OFM) — The Ultimate Social Media Tier List for OFM 2025, Dec 2025. Watch ↗
- Gavin Magoon — 2026 OnlyFans Social Media Updates Every Agency and Creator Should Know, Nov 2025. Watch ↗
- Will Mammone — Why Is Everyone QUITTING OnlyFans?, Sep 2025. Watch ↗
- Gavin Magoon — Mastering TikTok Traffic for OnlyFans Creators, May 2026. Watch ↗
- Patryk — TikTok Traffic Guide for OFM (2026), May 2026. Watch ↗
- Will Mammone — The ACTUAL Best Traffic Method For OnlyFans Creators (forever), Oct 2025. Watch ↗
Community intelligence: 79 operator claims aggregated from 7 separate private OFM groups (Dec 2025–Jun 2026), corroboration counted across groups. Group identities are withheld to protect sources; browse the underlying intel in the Community Intel Wiki.