By Jonathan Gillham, Founder and CEO, Originality.ai
Coverage: January 1, 2025 through August 17, 2026
Originality.ai analyzed every eligible English House and Senate Floor Turn over 100 words in the daily Congressional Record for 2025 and 2026 YTD.
Of 21,388 turns, 2,099 were classified as Likely AI. The observed rate rose 3.8 percentage points, from 8.4% in 2025 to 12.2% in 2026 YTD.
Why we looked at Congress:
The issue has already drawn national coverage from Politico, which reported that AI-shaped requests were creating problems inside the House Office of Legislative Counsel. Those confidential requests are not public.
So, we asked a narrower public question: what Likely AI signal is visible in the words attributed to serving members in the Congressional Record?
These are our findings.


In 2025, 961 of 8,415 House turns were classified as Likely AI, compared with 155 of 4,923 Senate turns. In 2026 YTD, the comparison was 814 of 5,503 House turns, and 169 of 2,547 Senate turns.

Among politicians with at least 10 eligible turns, 354 met the threshold in 2025 and 270 in 2026 YTD. In 2026 YTD, 80 had no turns classified as Likely AI, while 25 had rates above 40%.

Across the combined 2025–2026 YTD period, 495 politicians had more than five eligible Floor Turns. We removed politicians with less than 5 speeches from this analysis. Their observed Likely AI rates ranged from 0% to 90.0%. The chart ranks every qualifying politician from lowest to highest.

Note: The table below excludes the names of the politicians.

If you are interested in the raw data, please reach out; we are happy to share the data with parties working to further the understanding of probable AI use within government.
Likely AI turns appeared among both Democratic and Republican members in each chamber-period.
In 2026 YTD, Likely AI rates were:
We found two clear disclosures.
Senator Marsha Blackburn said she queried Google's Gemma about herself and described the model fabricating an allegation.
Senator John Cornyn said, “I asked my friendly AI assistant,” before presenting material connected to a request about parole programs. Neither senator said AI drafted the congressional language.
We found no AI-drafting disclosure and no hard AI residue in the U.S. corpus. That differs from the New Brunswick investigation, where apparent drafting instructions were read aloud.
If AI is used to draft or materially shape words entered into an official legislative record, a simple disclosure would let the public distinguish AI-assisted language from member- or staff-written text.
Likely AI language is no longer an isolated occurrence.
Its sharp growth suggests AI is becoming part of how congressional material is prepared, considering eligible remarks have seen the Likely AI rate rise from 8.4% in 2025 to 12.2% in 2026 YTD.
This shows that probable AI language is materially shaping words entered into the public record.
Likely AI is the detector classification for the entire recorded turn. It is not the percentage of a politician’s words written by AI. The classification does not identify who drafted a remark, so individual findings should be reviewed against the official passage and, where important, the member’s office and available video.
This study has been updated since its initial publication in August 2026. Here’s what’s different from our earlier study:
AI assisted with data processing, quality checks, figure production, and an initial draft. Jonathan designed the study, reviewed and approved this final report.
