Investigation: What Did Trump's Teleprompter Bettor Bet On?
Our High Confidence Findings of Anomalous Winning Trump Speech Trades
What the Data Shows About the Teleprompter Trades
A White House teleprompter operator made a reported ~$100,000 betting on the words he was scrolling.
We went back through Kalshi’s public records to try and identify the mention markets he traded.
What our system found:
On 16 July, ABC News reported that Gabriel Perez, President Trump’s teleprompter operator since 2016, was under CFTC investigation for trading Kalshi’s Mentions markets: contracts that pay out on whether a specific word gets said in a specific speech.
If true, it would amount to trading on advance knowledge of the exact words that would be spoken. It is difficult to imagine a cleaner alleged example of informational advantage.
Kalshi’s surveillance flagged and froze the account in March. The profit figure reported ranges between $90,000 (CNBC) and $100,000 (ABC).
ABC’s sources also say the CFTC alerted federal prosecutors in Manhattan at some point, and that those prosecutors declined to open a criminal investigation, which is a step short of any decision about charges. Perez was put on unpaid leave the day the story broke and, per Politico on 28 July, no longer works for the federal government. Settlement talks with the CFTC continue, on terms that would have him give back the profits and stop trading these markets.
Kalshi’s trading data is anonymous. There are no wallets to follow the way there are on Polymarket. Every account is a number that only Kalshi can resolve into a person.
So we went looking for the trades instead. We knew we couldn’t find a name, but instead focused on the shape of the positioning.
We targeted contracts that exhibited the unusual trading patterns described in the reporting.
We found some interesting results.
Across 3,166 Trump mention contracts, six words consistently stood out:
Crypto (three separate speeches)
Pelosi
Rare earths
Somalia
Elon
Olympics
These were among the most significant examples identified by our research.
What is actually visible
Kalshi’s live API serves tick data and candles for current markets, and stops serving them once a market is more than about eight weeks old. We went looking anyway and got 404s across the entire December-to-March window. In its high-resolution form, the record of this scandal is already gone.
What did survive, however, is a compliance artifact. Kalshi is compelled to and does file a daily record for every contract into a CFTC reporting bucket: volume, open interest, high, low, status. One row per contract per day, and it does not expire.
It is much coarser than an API feed, but it turned out to be enough to reveal some interesting patterns.
We swept all 126 daily files covering 1 December 2025 through 5 April 2026 and reconstructed the full universe of Trump speech mention markets in the window: 131 speech events, 3,166 individual mention contracts, 129 events with usable data.
From BETINT Brief BI-2026-0716-KMM, issued 16 July 2026.
Then we put four questions to each of those 3,166 mention contracts.
Did it settle YES?
Did it trade before the speech as a longshot, under 45 cents?
Did it carry real size beforehand, at least 200 contracts?
And was that size unusual next to the other words in the same speech?
The method is meant to hunt for one thing: somebody buying a cheap word in size, shortly before that word gets said.
Price falling while open interest climbs is the shape everything below is built on.
The result
41 contracts flagged across 22 speech events.
Aggregate profit ceiling on the flagged pre-speech positioning: $84,253.
The reporting says $90,000 to $100,000.
Davos is the cleanest example in the dataset
Trump’s address at the World Economic Forum, 21 January. The contract: will Donald Trump say the word “crypto.”
Three days of the “crypto” contract, straight off the exchange’s daily files.
Read the middle row twice. In one session the price fell by more than half while open interest went up sixty-fold, which means somebody absorbed roughly 6,900 YES contracts into a falling market - the crowd increasingly believed he would not use the word.
Trump DID speak the word CRYPTO the next day. About $5,500 on that block.
You can get a big position on a cheap word from conviction about the news cycle, but what you don’t usually get is the price sliding the whole time you are buying.
That is the signature of one side knowing the answer while the other side does ordinary probability estimation, and the crowd here grew more confident it was right up until the moment it wasn’t.
The State of the Union is where the money is
24 February. The largest cluster in the scan: roughly $30,000 of flagged positioning across six contracts in the “who and what will Trump mention” families.
The ticker codes are the exchange’s symbols and the word decodes are ours - the last two are inferences rather than confirmed contract terms.
Pelosi is an interesting one. Open interest on that contract went from 4 to 4,500 starting in mid-January, while the price walked down from 45 cents into the low teens. This opens the door to weeks of quietly buying a word the market kept marking down, and it settled YES on speech night.
Everything else in the window
Detroit Economic Club, 13 January: 2,978 contracts on “rare earths” bought the day before at 34c to 48c. YES the next day. “Somalia” alongside it.
Tech Roundtable, 4 March: 1,419 contracts on “crypto” at 7c to 22c the day before. Opened at 6c on speech day, settled YES.
The weekly “what will Trump say” market for the first week of January: about $14,000 of longshot-YES positioning on “Elon” and “Olympics.”
Worth noting that the cheap crypto word shows up three separate times across the window, on three different speeches.
Where we go blind
Two of the five named speeches are mostly invisible to this method.
The December primetime Address to the Nation was voided and relisted on speech day. The original listing traded thinly for a couple of days before the reset, and the scan did flag one small position there, about $600 of the total.
But the relisted markets, where the real volume landed, only ever existed on speech day: 186,000 contracts traded on the word “Biden” alone, and a single daily row cannot tell you which of that was bought before he started talking. The March Medal of Honor ceremony has the same problem.
The two speeches the method cannot see, and why.
Blank is not the same as clean. Those two speeches are simply outside what we were able to take a stab at.
The same limit buries what is probably the most damning detail in the whole story.
ABC describes Perez pulling out of positions mid-speech when Trump skipped a scripted word, which is about as close to a confession as market behavior gets. At daily resolution it is invisible. On tick data you couldn’t miss it.
Some Caveats
We cannot attribute anything. The data is anonymized, we see positioning and not accounts, and “consistent with the pattern that was reported” is the strongest claim available to us.
We are stuck at daily resolution, so the mid-speech exits are gone and same-day markets cannot be scanned.
The dollar figure is a ceiling on each contract and an undercount across the window, for the reasons above. Best treated as an order of magnitude.
And settlement is inferred from final-day prices, which works well for liquid contracts but not as well for thin ones where a word got said but nobody bothered repricing it.
This piece draws on BETINT brief BI-2026-0716-KMM, a four-page internal briefing on the scan, its method and its limits. It is not published.
Sources and method. BETINT internal scan, 16 July 2026, with reporting rechecked 5 August 2026.
Universe reconstructed from Kalshi trade-api v2 event enumeration and the kalshi-public-docs S3 CFTC reporting files, 126 daily files covering 1 December 2025 through 5 April 2026, series KXTRUMPMENTION, KXTRUMPMENTIONB, KXTRUMPSAY, KXPRESMENTION, KXDJTJOINTSESSION, KXWEFMENTION and related.
Reporting: ABC News, CNBC, NPR, NBC News, all 16 July 2026, plus Politico, 28 July 2026, on his departure from government. Nothing in this analysis attributes any trade to any individual.
BETINT produces this level of coverage on the markets that move your business, your sector, and your stakeholders. To point it at the markets that affect you, reach out at bettingintelligence.ai or contact dan@betint.ai.