Do X mentions predict stock moves? What our own counts show
7 August 2026 · updated 9 August 2026 · 5 min read
Social mention counts get sold as an edge. They are not one, and the gap between what the data measures and how it is marketed is worth setting out plainly - particularly since this site now publishes its own.
Every day we query the X API directly and count how many posts carried each of our 34 tickers' cashtags over a complete New York weekday, retweets excluded. On Friday 7 August that came to 15,164 posts, with every one of the 34 names appearing at least once. Because those numbers are ours and the price data sits beside them, we can do something most mention trackers never do: check what the counts actually relate to.
What the counts have no relationship with: direction
Rank the board by posts, rank it again by that day's price move, and compare the two orderings. The rank correlation is -0.00.
Not weak. Not "modest but interesting". Zero, to two decimal places, across all 34 names on a full trading day. Post volume carries no information whatsoever about whether a stock went up or down.
That is not a flaw in the measurement, it is the measurement working correctly. A count records how often a ticker was typed. It cannot know whether the posts were celebrating or panicking, and a stock being loudly discussed during a collapse produces the same number as one being discussed during a run. Inferring direction from volume is where most social-sentiment products go wrong, and it is why this figure sits beside our market sentiment score rather than inside it.
What they do track: money actually moving
The strongest relationship on the board, by a distance, is with dollar turnover - the value of shares changing hands. Rank correlation 0.71.
This is the finding that makes the number worth publishing. Bot chatter and promotional spam would correlate with nothing, or with a fixed list of names that ignores what the market is doing. Instead the posting tracks the trading. People write about what is being traded, on the day it is being traded.
Two weaker relationships are worth naming so nobody over-reads them. Against the size of a day's move regardless of sign, the correlation is 0.23 - a big move attracts somewhat more posting, but only somewhat. And against market capitalisation it is 0.15, which is close enough to nothing to be the most interesting number here.
Attention does not follow size, and that is the point
TSMC is one of the largest chipmakers on earth at roughly $2.18tn. On Friday it drew 199 posts. IREN, at about $15bn, drew 983 - five times as many from a company one hundred and forty-fifth the size.
Measured as posts per billion of market value, the ordering is almost the inverse of the board itself:
- IREN - 66.8 posts per $bn
- Cipher Mining - 56.8
- Nebius - 38.1
- ...
- TSMC - 0.09
- Monolithic Power - 0.09
The top of that list is neoclouds and bitcoin miners turned AI hosts; the bottom is the industrial core of the supply chain. This is information the price-based score does not contain and cannot contain, because it is not about how a stock is behaving. It is about which stocks people have decided to care about.
Posts on X carrying each ticker's cashtag, counted by this site over 27 August 2026. Retweets excluded. See all 34 names →
The claim we are not making
The title asks whether mentions predict moves. We cannot answer that yet, and it would be easy to pretend otherwise.
Here is the temptation. Compare the day-over-day change in posts against that day's price move, among names clearing our 25-post floor, and the correlation is 0.49 - respectable, and exactly the sort of figure that gets screenshotted as proof of an edge.
It is nothing of the kind. Both numbers describe the same day. A stock moves, people post about it having moved, and the two rise together. That is contemporaneous correlation, and it is what you would expect even if mentions had no forward-looking content at all. Testing prediction means comparing today's posting against tomorrow's return, repeatedly, over enough days that a run of luck cannot carry the result.
We have been collecting this since 3 August. Five trading days is not a sample, it is an anecdote. When there is enough history to run that test honestly we will publish the answer whichever way it falls, including if it turns out to be nothing.
How to use it in the meantime
Read the change rather than the level. NVIDIA leads the board most days simply because it is NVIDIA; that carries no information. A name going from 21 posts to 55 is a different event from one that is permanently loud, which is why the X mentions board shows the day-over-day change beside the count.
Treat small numbers as noise. Below 25 posts a day we show the count but suppress the percentage, because a move from three posts to nine is three people, not a 200% surge in attention.
And the comparison actually worth making is against the score. Where heavy posting meets a weak market sentiment reading - Nebius on Friday drew 1,801 posts on a sentiment score in the twenties - you are looking at a name the crowd is interested in and the price action is not rewarding. That disagreement is the useful signal. It is not a buy or a sell, but it is a question, and it is one neither number raises on its own.
Frequently asked questions
Do X mentions predict stock prices?
There is no evidence here that they do, and we are not claiming it. Mention volume has a rank correlation of -0.00 with the direction of that day's price move across all 34 names. It correlates strongly with dollar turnover (0.71), meaning people post about what is being traded, but that is a description of the present rather than a forecast. Testing prediction needs today's posts against tomorrow's returns over a long run, and we have been collecting since 3 August.
Where do these counts come from?
ChipSentiment queries the X API directly and counts posts carrying each ticker's cashtag over a complete New York weekday, with retweets excluded so one viral post cannot stand in for thousands of people. The figures are not bought from a third-party mention tracker or an aggregator.
Why do the biggest companies not have the most posts?
Because attention does not follow size - the correlation with market capitalisation is 0.15. TSMC, at roughly $2.18tn, drew 199 posts on 7 August while IREN at about $15bn drew 983. That gap is the interesting part, and it is information a price-based score cannot contain.
Why are weekends missing from the chart?
They are collected and stored but not displayed. With the market shut, weekend posting runs at roughly half a weekday across the entire board at once, so charting it beside trading days produces a sawtooth that says more about the calendar than about any stock.
Does a high count mean the posts were positive?
No. A count records how often a ticker was typed, not whether the posts were bullish or bearish. That is precisely why the direction correlation is zero: a name discussed heavily during a collapse produces the same number as one discussed heavily during a run.
Sources
- X API - recent post counts endpoint - the endpoint these counts are queried from, daily
- ChipSentiment - live X mentions board - full ranking with price and market sentiment alongside
Figures are taken from the public filings and the reporting linked above.