US TikTok advice describes a different algorithm now

Most of the TikTok advice reaching Australian marketing teams describes a system your audience is not using. The January 2026 joint venture put the US app on a recommender being retrained on US user data inside Oracle's cloud. Two apps, one interface, and increasingly different machinery underneath.

Nobody outside the arrangement knows how far the two have diverged. That is precisely why importing American numbers is now a guess dressed as data.

What changed structurally

The joint venture separated the US operation and the training environment behind it. The model serving American users is being retrained on American data in American infrastructure. The app your Australian audience opens is not that model.

For years the sameness of the algorithm was the reason US advice travelled well. A creator in Los Angeles and a business in Parramatta were being ranked by the same system, so their observations were roughly comparable. That assumption has quietly expired and nothing on the screen tells you.

The three kinds of advice that no longer transfer

Benchmarks are the first. Average watch time, follower conversion rate, the share of views from the For You page, the typical view count for a given follower count. These were always shaky because they varied by category. Now they vary by country model too.

Sound and format trends are the second, and this one is more concrete. Trend propagation depends on what the recommender is amplifying in that population. A sound peaking in the US tells you very little about whether it will move in Australia, and it may already be stale locally by the time the American case study is written.

Timing and cadence advice is the third. Posting time guidance was mostly folklore anyway, built from small samples and repeated until it sounded official. Now it is folklore from a different system.

What still transfers

Craft does. How to open a video so people stay, how to shoot so it reads on a phone, how to write a line that makes somebody send it to a colleague. None of that is model specific and none of it changed.

Platform policy mostly does, because policy is written for the product globally even where enforcement varies.

So the split is simple. Take craft from anywhere. Take numbers only from your own account or from Australian accounts in your category.

Build your own baseline, it takes an afternoon

You do not need an industry benchmark. You need yours.

1. 1.Export your last ninety days of TikTok posts with views, average watch time, and shares.
2. 2.Sort by average watch time, not views. Watch time is the input, views are the output.
3. 3.Take the top five and the bottom five and write one sentence on each about what the first three seconds did.
4. 4.Record the median for views, watch time and shares. That median is your benchmark from now on.
5. 5.Repeat monthly. When the median moves, you have a real signal about your account rather than a headline about somebody else's.

Ninety days of your own data beats any published average, because it already accounts for your category, your country, your production quality and your audience.

If you sell into both markets

Run separate accounts. One Australian, one American, with content cut for each. This was already sensible for language and reference reasons, and it is now sensible for a mechanical reason as well: you cannot read performance across two different recommenders in one dashboard and learn anything reliable.

The cost is real. Two accounts means two content calendars and roughly double the production, and the second one usually starves. Decide honestly whether the US market is worth a properly resourced second operation before you open the account.

What to say when someone brings a US case study to a meeting

Ask two questions. Which country was that account in, and what date was that measured. If the answer is the US and any date after January 2026, the finding describes a system your buyers are not being served by. If it is before, it describes a system that no longer exists in either country.

That is not a reason to dismiss it. Craft observations survive the question. Numbers do not.

The bigger habit this should break

The deeper problem was never TikTok. It was running Australian strategy on American evidence because American evidence is louder and more available. Search results, conference talks, creator courses and marketing newsletters are dominated by the US market for reasons that have nothing to do with relevance.

This split just made the cost visible. Treat every imported number as a hypothesis about your market until your own data agrees with it.