GUDEA on the Texas and Kansas Senate Races: How 625 New Accounts Amplified One Claim in an Hour

GUDEA's analysis of one week of online conversation about the Texas and Kansas Senate races, published on October 26, 2026 and reported by The New York Times the next day, found that the conversation was carried by a small cross-state audience, that it had the lowest share of original content GUDEA has ever recorded, and that a network of 625 newly registered accounts amplified a single allegation about a candidate 941 times in one hour and then went silent. GUDEA, an AI-powered narrative intelligence platform, examined 951,486 posts from 264,247 accounts across 27 platforms between October 19 and October 25, 2026: 918,427 posts about Texas and 33,059 about Kansas.

What the data showed

Three findings define the brief. First, the Kansas conversation was not a Kansas conversation. 73.6 percent of the 16,551 accounts posting about the Kansas race also posted about the Texas race in the same week, and those 12,185 shared accounts produced 76.9 percent of all Kansas posts. Fewer than one in four Kansas posts (23.1 percent) came from accounts that posted only about Kansas. The same shared accounts were 4.7 percent of the Texas audience but produced 16.4 percent of Texas posts, four times the output of accounts that posted about Texas alone. Second, the conversation was amplification rather than discussion: 85.2 percent of Texas posts were reposts, and typical accounts, 97 percent of the audience, produced only 14.9 percent original content. The New York Times reported that this was the lowest percentage of original content GUDEA had ever tracked. Third, the non-typical share was above GUDEA's usual range: 2.67 percent of Texas accounts and 3.31 percent of Kansas accounts were classified non-typical, and they produced 20.2 percent and 22.2 percent of post volume respectively. X carried 98.2 percent of Texas posts and 91.3 percent of Kansas posts; the rest were spread across Bluesky, Reddit, Telegram, 4chan, Discord, Gab, 8kun, and other platforms.

The 625-account burst

The brief's central finding is a registration burst. GUDEA decoded account creation dates from X user IDs and found that new accounts entering the Texas conversation ran at a baseline of about 58 per day between September 1 and October 14, then spiked to 125 on October 17, 219 on October 19, and 625 on October 21. The accounts created on those three days shared a signature: 37, 35, and 38 percent of them posted exactly three times in the window, against a baseline of 7.4 percent, and the three posts were the same three reposts of relay posts about an allegation that Democratic Senate candidate James Talarico had assaulted a student. In the 20:00 UTC hour on October 21, the new accounts reposted those relay posts 941 times, peaking at 93 reposts in a single minute. Then they stopped: 97 percent of the October 21 cohort halted all posting activity afterward.

GUDEA traced the allegation hour by hour. The account that posted it was registered on September 11 and had one post in the dataset. The allegation went up on October 20 at 16:02 UTC and was reposted 12 times within 23 seconds. Four relay accounts posted "shocking video" versions within 38 minutes early on October 21; Grok, X's assistant, replied 18 times in that window that the claim was uncorroborated and that the origin account had a Romanian signup origin, statements GUDEA reports as captured and did not verify. Broad pickup followed through the day, then the registration burst and the 941 reposts, then a second wave on October 22 from a large account whose 915 reposts drew none from the burst cohort. Of the 3,198 accounts that posted about the allegation, 583 were created on or after October 15 and 451 posted within 24 hours of their own creation. Kansas showed no burst: new-account entry rose with the election conversation but without the three-post signature, and only 14 of the 969 Texas burst-day accounts posted about Kansas at all.

What the brief does not claim

The brief is explicit about its limits, and the article repeats them. GUDEA did not evaluate the allegation; the brief describes how it was amplified, not whether it is true. GUDEA made no attribution: nothing in the data identifies who operates an account, four handle-naming tests failed to separate the burst cohort from ordinary new accounts, and handle appearance was not used as evidence. The data contains no location, so presence in both state conversations means an account posted about both states, not where it is. The Kansas overlap has one control (a Michigan export, Bluesky only), so whether that overlap is unusual among state conversations on X is untested. The New York Times wrote that the circulation of the claim matches activity several other research groups have attributed to a Russian influence operation; that is the Times' framing, and GUDEA's brief makes no such attribution.

Why it matters beyond Texas and Kansas

Presley told The New York Times that what could look like organic conversation around the two races is largely amplification by a smaller number of accounts: real people are walking into a conversation where the table was already set, with the topics and framing in place before most users showed up, and an ordinary person scrolling has no way to tell the difference. "Volume is easy to manufacture, and right now it gets read as public opinion." He added that nothing in the playbook is specific to Texas or Kansas; the account pipeline GUDEA identified is infrastructure that can be pointed at any contest. For a brand, the same mechanics apply: the Taylor Swift report found 3.77 percent of users producing 28 percent of posts, the Bad Bunny report 3.7 percent producing 25.85 percent, and this brief 2.67 to 3.31 percent producing 20 to 22 percent. The structure is the same; only the target changes.

How the analysis was done

The brief is the first GUDEA analysis published with a full methodology and limits section. It draws on two most-active-user exports per state and the full post-level exports for the week, reconciled exactly with the account rollups; a Michigan export for the same window as a control; account creation dates decoded from the timestamp embedded in X user IDs, checked against a known account; a registration baseline of August 1 to October 14; a content review of every flagged account's posts for repost share, source, distinct texts, link domains, and minute-level co-posting; a domain match against 185 live Pravda-network domains and the 32 domains named in the September 2024 Doppelganger seizure; and Audience Behavior Classification run independently for each state. GUDEA's research methodology page explains the approach in general; the brief's section 7 gives the specifics for this dataset.

Frequently Asked Questions

Did inauthentic accounts drive the online conversation about the Texas and Kansas Senate races?

GUDEA's October 2026 brief found that much of it was amplification rather than discussion: 85.2 percent of Texas posts were reposts, typical accounts produced only 14.9 percent original content, and 625 newly registered accounts reposted the same three posts about an allegation 941 times in one hour on October 21 before 97 percent of them went dormant. In Kansas, 73.6 percent of accounts were also posting about Texas and produced 76.9 percent of Kansas posts. The brief does not identify who operates any account.

What did GUDEA find about the Talarico allegation?

GUDEA traced how it moved, not whether it is true. The allegation was posted on October 20 by an account registered in September with one post, reposted 12 times in 23 seconds, relayed by four accounts as "shocking video" versions, picked up broadly on October 21, then amplified by a burst of 625 accounts registered that day, which reposted it 941 times in one hour and stopped. Of 3,198 accounts that posted about it, 583 were created on or after October 15. GUDEA did not evaluate the claim and makes no attribution.

How many posts and platforms did the Texas and Kansas analysis cover?

951,486 posts from 264,247 accounts across 27 platforms between October 19 and October 25, 2026: 918,427 Texas posts from 259,881 accounts and 33,059 Kansas posts from 16,551 accounts. X carried 98.2 percent of Texas posts and 91.3 percent of Kansas posts.

Was the Texas and Kansas activity attributed to a foreign operation?

Not by GUDEA. The brief states that nothing in the data identifies who operates an account and that it makes no attribution. The New York Times, reporting on the brief, wrote that the circulation of the claim matches activity other research groups have attributed to a Russian influence operation; that is the Times' characterization. GUDEA checked every link in the exports against 185 Pravda-network domains and 32 Doppelganger domains as part of its content review.

How did GUDEA know when the accounts were created?

X user IDs embed a creation timestamp (the "snowflake" scheme used for new accounts since roughly 2016). GUDEA decoded it and checked the method against a known account. Older sequential IDs carry no timestamp and were excluded; they are 33 percent of Texas and 38 percent of Kansas X accounts. Creation dates refer to registration, not first use, so an aged or purchased account shows its original date. The method applies only to X.