TikTok's US algorithm just got rebuilt from scratch. Here's what changed.
Our TikTok ownership guide already covered the headline event: US operations moved to an Oracle- and Silver Lake-led joint venture in January 2026. The part that matters operationally for a creator arrived after that: the US For You feed now runs on an Oracle-retrained, US-only algorithm, separate from the version TikTok runs everywhere else in the world. That's a bigger deal than it sounds, and it's worth understanding before assuming last year's reach patterns still hold.
What's confirmed, and what isn't
| Question | Status |
|---|---|
| Is the US feed now on a separate algorithm from the global version? | Confirmed |
| Are the core ranking signals the same? | Reportedly yes: completion rate and watch time first, then saves and shares, then likes |
| Was there a major new ranking change in July or August 2026? | No confirmed change beyond the algorithm swap itself |
| Full technical detail on what the retraining actually altered | Not publicly disclosed |
Why "the signals are the same" doesn't mean "nothing changed"
Two systems can optimize for identical stated priorities and still behave differently once the underlying model, training data, and weighting are rebuilt from a different base. That's the practical takeaway here: a creator whose content hasn't changed at all could still see reach move simply because the system deciding who sees it is new, not because anything about the content itself stopped working.
The sensible response
Judge reach against your own recent performance rather than last year's numbers while the new system settles, and keep the fundamentals that were already true under the old algorithm: completion rate and watch time still lead, which is the same retention discipline our hooks guide already argues for regardless of platform. If TikTok is one leg of a cross-platform distribution strategy, this is a reasonable moment to watch that channel's numbers a little more closely than usual, not to overhaul the content feeding it.
Where Thothium fits
Thothium exports the same finished project to whichever platform's algorithm happens to be rewarding it that month, so a shift like this changes where a video performs best without forcing a separate production process for each platform. It is in free alpha, and the form below gets you a key.
Frequently asked questions
Does this mean my TikTok content needs to change?
Not the content itself. The reported ranking signals are unchanged: completion rate and watch time first, then saves and shares, then likes. What's changed is the infrastructure underneath, which means the specific way those signals get weighted and applied could differ even if the inputs a creator optimizes for stay the same.
Is this related to the ownership change already covered on this blog?
Yes, directly. Our TikTok ownership guide covered the January 2026 divestment, when US operations moved to the Oracle- and Silver Lake-led joint venture. This is the operational follow-through on that deal: the actual recommendation algorithm for US users has now been rebuilt and retrained under the new structure, separate from the version TikTok runs elsewhere.
What's actually different about the new algorithm versus the old one?
Full technical detail hasn't been made public. What's confirmed is that it's a US-specific, Oracle-retrained system, distinct from the global TikTok algorithm going forward, which is itself the significant part: US and international reach are no longer necessarily driven by the same underlying model.
Should I expect reach to be unstable for a while?
A retrained ranking system settling in in real time is a reasonable case for treating recent performance, not last year's benchmarks, as the baseline while it stabilizes. This is a bigger structural shift than a routine algorithm tweak, and those tend to come with a settling period even when the stated ranking priorities don't change.
Last updated August 31, 2026. Details of the algorithm change are based on reporting available at the time of writing; TikTok has not published full technical documentation of the retrained system.