Twitter Data Goes Stale Fast: New Study Shows the Hidden Risk

Social media platforms are packed with numbers: how many followers you have, how often you post, how many likes you get. Researchers and marketers often collect these numbers to build lists of important users or to study online behavior. But this new study from Japanese researchers asks a simple question: do those numbers stay meaningful after a few months? The answer is no, not really.

The team looked at Twitter data from early 2020 to late 2022, covering about 10 to 11 million Japanese users each quarter. They measured 13 different user features, like follower counts and tweet frequency. They found that the distribution of these numbers shifted over time, and more importantly, how users ranked compared to each other changed. The person who was in the top 1% for engagement in January might not be in the top 1% by July.

This becomes a problem when companies or researchers set a threshold, like "focus on anyone with more than 10,000 followers." A threshold that made sense one quarter might select too many people or too few a few months later. Recalculating the threshold from current data fixed the overall selection rate, but the membership turnover persisted. That means even when you pick the same percentage of users, many of them are different people than before. So any analysis relying on old "top user" lists is getting a stale, increasingly inaccurate view.

Why should you care? If you use Twitter data for marketing, journalism, or academic research, results can quietly become unreliable in just a few months. The study suggests that regular recalibration is essential, but even that can't stop the underlying shift in who the data points to. In practical terms: refresh your data often, or your conclusions — and your campaigns — will be built on shifting sand.