How We Use AI

polititune is AI-assisted and human-reviewed, and we would rather over-explain that than hide it. Here is exactly how an article gets made.

The pipeline

• Capture: we record the live stream each Senate committee publishes on senate.gov.
• Transcription: speech-recognition software (Whisper-class) produces a transcript within about a minute of live audio.
• Speaker attribution: software matches voices to committee rosters and procedural cues. If confidence is too low, we label the speaker "a senator" or "a witness" instead of guessing.
• Event detection and drafting: a large language model flags newsworthy moments (votes, amendments, notable exchanges) and drafts a short article from the transcript.

What machines verify

Every direct quote in a draft is machine-checked to be an exact substring of the transcript. Numbers get a second dedicated check. Drafts that fail are stripped or regenerated.

What humans do

No article is published without a human editor reviewing the draft against the transcript evidence and explicitly approving it. The editor sees each quote alongside its place in the transcript, the speaker-attribution confidence, and the audio. The byline on every article is a real, accountable person.

What this means for you

Auto-transcripts are labeled as such and may contain recognition errors. Articles are reviewed. Every article links its hearing transcript so you can verify any claim against the primary source yourself.