How News Actually Moves Bitcoin — 90 Days of Verified Calls
The dataset: 10,810 graded calls
Most outlets answer "is this news bullish for Bitcoin?" with vibes. We answer it with a grade. Our AI assigns each story a direction (bullish/bearish) and a 0–100 impact score; 24 and 72 hours later we check the call against the actual price and record hit or miss in a public ledger. Crypto stories are verified against Bitcoin; macro and equity stories against the S&P 500. Only calls with impact ≥ 60 and a clear direction are evaluated, and if the price barely moves (under ±1.0% for BTC, ±0.3% for SPY) the call is marked "neutral" and excluded from the hit-rate denominator.
As of July 17, 2026, the last 90 days contain 10,810 evaluated calls on the 24-hour window. Every number in this guide comes straight from that ledger — no cherry-picking.
The headline number: 52.9% — modestly better than a coin flip
Over the last 90 days, 3,207 of 6,057 decisive calls were correct: a 52.9% hit rate (as of 2026-07-17). Narrow it to the last 30 days and it drops to 50.6%. Stretch the window to 72 hours and the 90-day figure decays to 49.0% — essentially a coin flip.
That is not a flattering number, and that is the point. Predicting the 24-hour direction of a market from a headline is genuinely this hard. We publish 52.9% even when it looks modest, because a verifiable, unflattering number beats an unverifiable "90% accuracy" claim every time. If you want to understand how the grading itself works, start with What "Verified Accuracy" Means.
Most headlines do not move the market at all
Of the 10,810 calls graded over 90 days, 4,753 — roughly 44% — were neutral: 24 hours later, the price had not even cleared the ±1.0% BTC threshold. Remember, these are stories already filtered to impact ≥ 60 — the supposedly meaningful ones. Nearly half of them left no measurable mark on the next day's price.
This is not a failure of the model; it is how markets work. Most headlines are already priced in, cancel against opposing news, or cover things the market simply does not care about. When you read "this news will send Bitcoin soaring," the base rate says the most likely outcome is: nothing happens.
When Bitcoin does move, how big is it?
In the crypto category over the last 90 days, 4,777 decisive calls hit at 52.4%, and when a call hit, Bitcoin's average absolute move was 2.41% (on misses it still moved 2.18% — the market moves hard even when the call is wrong). For comparison: macro news hit at 53.1% with an average S&P 500 move of just 0.59%, and global-markets news hit at 54.7% with 0.68%.
The pattern: equity-linked categories are slightly easier to call, but crypto moves are roughly four times larger. Bitcoin's direction is harder to read, but the payoff for reading it right is bigger; equities telegraph their direction a bit better, but a typical one-day reaction is well under 1%.
The AI importance score does not predict accuracy
We checked whether the impact score is decoration or signal. On the 24-hour window over 90 days:
Correction (2026-08-06). This section used to claim that accuracy rose monotonically with the importance score — 49.6% in the 60s up to 55.5% at 80+. That claim was wrong. Holding out the period after those numbers were taken and rerunning the same calculation reverses the trend completely (22.7% for the 80+ bucket). The score was never tracking how likely a call was to be right; it tracks how alarming the story sounds, and concentrating bearish calls in a rising market lowers the hit rate by construction. The full account is in When our AI was wrong. We are not printing replacement bucket numbers here — hardcoding them is what produced this error. The public ledger is the source of truth.
Anatomy of the misses: when bullish news loses
The most painful misses share a fingerprint. Every top-tier miss of the last 90 days (impact ≥ 80, Bitcoin moving 3%+ against the call) clusters on a single day: June 1, 2026 — the eve of the June 3 crypto sell-off. "Strive pushes to expand $4.2B capital raise limit for Bitcoin purchase" was called bullish; Bitcoin fell 5.9% in 24 hours. "Bitcoin perpetual futures trading gets approval," also bullish — down 5.69%. Individually positive stories were simply steamrolled by a market-wide risk-off wave. The lesson the data supports: single-name good news does not beat a bad regime.
The flip side: bearish calls during that same sell-off hit hard. On June 3, "Ethereum crushed to $1,840 amid relentless selling" was called bearish and the market delivered -7.41%. News calls work best when they point the same way the tape is already moving. One more recurring trap: on July 13–14, four separate "Fed's Waller signals rate hikes" stories were all called bearish — and the S&P 500 rose 0.38%. Hawkish talk was already priced in; the market traded on the actual data (June CPI cooling). You can see those data events coming on our economic calendar.
What to actually do with this
Three rules fall out of the data. First, never trade a single headline — nearly half the time, nothing happens. Second, use the impact score as a filter — 80+ stories are statistically more trustworthy. Third, regime beats headline — the June 1 bullish wipeout is the proof. Every call and every outcome behind this guide is inspectable in the verification ledger.
All figures are from our public verification ledger as of July 17, 2026 and will drift as new calls are graded. This guide is informational only, not investment advice. See our Editorial Policy & Disclaimer.