ChatGPT in Congress is crushing every rival: House offices poured 100,580 dollars across 798 transactions into the tool, or 88% of all identifiable AI spending. Claude, in second place, stalls at 13,160 dollars. The figure marks the first de facto AI standard inside an institution that will soon write the rules for the technology itself.
Key Takeaways
- ChatGPT captures 88% of the House’s identified AI purchases, far ahead of Claude.
- The tool shows up in at least 71 member offices, roughly one in six.
- Public records exclude free accounts and bundled licenses, so real adoption runs even higher.
What the House disbursement records reveal
The House of Representatives publishes its disbursements, and reading them line by line offers a rare snapshot of AI as an administration actually buys it. Over the year, member offices spent at least 113,740 dollars on identifiable artificial intelligence tools.
Of that total, ChatGPT grabs 100,580 dollars spread across 798 transactions. OpenAI‘s product appears in at least 71 member offices, about one in six. No other tool comes close to that density of use.
The described uses are anything but exotic. Congressional staffers rely on it to draft memos, summarize and analyze legislation, prepare hearings, sift through policy research and answer constituents. It is the same base of office tasks we watched erase the lines between jobs in the wider workplace, now transposed to the heart of the legislative branch.
A partisan split runs through the numbers. Democratic offices spent more than three times as much as Republican ones on identified AI purchases, with ChatGPT appearing in 44 Democratic offices against 27 Republican. Adoption of the tool clearly comes before any official stance on regulating it.
The volume of transactions tells its own story. 798 separate purchases describe not a one-off experiment but recurring use, spread across dozens of offices that reorder month after month. The tool has entered the daily routine, not the pilot phase.
OpenAI locks in the institution, Anthropic stays behind
The real signal in these records is competitive. By settling into the working habits of congressional teams, ChatGPT in Congress earns an institutional foothold that no rival dislodges with a simple price cut.
Claude lands second with 13,160 dollars across 37 transactions, a long way back. The gap is all the more striking because Anthropic openly bets on office work, to the point of admitting that Claude runs more admin tasks than code. The positioning fits a congressional office perfectly, yet real usage has not followed yet.
That lag contrasts with the enterprise picture, where Anthropic managed to pull ahead of OpenAI in market share for the first time. Congress tells a different story: one of a mainstream reflex, where the name ChatGPT works as a synonym for AI among staffers who never compare models.
For OpenAI, this pocket of usage is worth more than its modest dollar figure. It is the same move as a Codex aimed squarely at office jobs inside enterprises: occupy the desk first, monetize later. Every memo drafted with ChatGPT makes the next contract harder for a competitor to win back.
One player still moves in disguise: Microsoft. Its AI features often arrive bundled inside office licenses that are already paid for, and so stay invisible in this count. The real institutional duel may pit OpenAI less against Anthropic than against an AI melted into tools that lawmakers use without ever naming it.
More articles on Horizon
- EU AI Act Enforcement Begins on August 2
- Google Earth Pulls Its Fake Satellite Image Tool
- OpenAI’s Astra Solves 10 Long-Open Math Problems
A percentage that hides much broader adoption
That 88% is impressive, but it understates the reality. The records capture only a narrow slice of the actual AI use under the dome.
The analysis leaves out free accounts, AI features bundled into broader software such as Microsoft’s Copilot, and most Senate spending. In other words, a whole layer of invisible adoption escapes the count, and the real rate of AI use in Congress is very likely far higher than these budget lines suggest.
For the end user, here the congressional staffer, the consequence is concrete. Briefing notes, bill analysis and replies to voters now pass through a tool that still hallucinates regularly, in a setting where a single factual slip carries an immediate political cost. The human check becomes a job skill, not an option.
That lock-in carries an exit cost. The more staffers shape their methods around one specific tool, the heavier a vendor switch becomes in retraining and habits to undo. The foothold taken today will weigh directly on tomorrow’s public tenders, long after the media attention has faded.
On the public debate, these figures raise an uncomfortable question. The same lawmakers who will draft the rules for AI are already daily, dependent users of it, often through a single vendor. That familiarity can sharpen the coming regulation or, just as easily, plant a blind spot when the time comes to rein in the dominant player.
Follow the story on Horizon.
