Product1 distinct publisher3 min readPublished
X's Thursday disclosure of a 200,000-account farm drew 3.2 million views and published no reach figures, so Darren Linvill went to a suspended-account archive and counted what the flagged posts actually reached.
The Product Desk · Product desk
Compiled by The Product DeskSomething wrong?How this is made
The person deciding whether X's post can go into Monday's briefing note is not left with much to work with. It names no accounts and no operators [3], and as Craig Hale wrote in TechRadar, it carries no engagement or reach figures and no account of what the other 199,800 accounts in the farm were doing [9].
The arithmetic available from outside is small. The 200 accounts X flagged as capable of manipulating the American AI and energy debate are 0.1% of the roughly 200,000-account farm, one in a thousand [1]. Darren Linvill checked at least 59 of them, close to 30% of the flagged set [2], and found zero engagement on every one, which he attributes to X suspending them before they could build a following [4]. "No real humans saw these posts about data centers," he said [5]. Engagement here is the reach measurement itself rather than a proxy for anything else, and it reads zero.
Separate the thing being pitched from the thing being done. The pitch is a foreign operation turning Americans against data centres. What was done is a moderation action against accounts caught early, on a debate X itself described as legitimate, closed out for breaching its authenticity policy [10]. The posts were mostly links, several to local coverage of PJM's 2027/2028 capacity auction, which cleared at the $333.44 per MW-day cap and fell short of the grid operator's reliability target, alongside AI cartoons of operators getting rich while families opened their electricity bills [8].
What a platform disclosure like this is worth comes down to two things: whether anyone outside the platform documented the operation, and whether the platform published reach. OpenAI answered the first in June, and Linvill says the two sets of posts are "very, very clearly connected" [12]. X left the second blank, and a university filled it in at zero. That gap is what matters here: corroboration without measurement leaves a moderation statistic, measurement without corroboration leaves traffic whose origin can't be named, and having neither leaves a press release with 3.2 million views [3].
The opposition does not require attribution to explain. A University of Pennsylvania survey in early August put opposition to new local data centres at 61%, up 12 points on its own February and March reading of 49% [16][3]. YouGov last week had 47% calling new data centres bad for the country against 24% calling them good, nearly two to one [17][4]. Business Insider cited a Gallup poll from May with local opposition above 70% [18]. Greg Abbott has told the state utility commission and grid operator to shield residents from infrastructure costs, floated a ban on rural projects and said developments must be audited first, and says the rules have stopped about 1,800 of them [15]. The demand he was looking at ran to five times what Texas can currently supply [14].
The workable rule is to treat a takedown count as a moderation output until someone produces impressions per post and the dates they ran. X's own screenshots contained the hashtag that let one researcher with an archive of suspended accounts do that check [6], which is a low bar for a claim that travelled to millions of people.
Ranked by verification strength, evidence, and original report placement.
X's Global Government Affairs account posted a disclosure on Thursday saying its safety team had identified a bot farm of roughly 200,000 accounts.
Within that farm, X said, sat 200 accounts posting "in a manner that could manipulate a legitimate debate about American AI and energy policy".
X's post has 3.2 million views and names no accounts and no operators.
The hashtag capacityauction returned 90 posts, all dated January, with each account posting two or three times before X suspended it, according to a spreadsheet Linvill shared with Business Insider; TNW has not reviewed that spreadsheet.
Darren Linvill, a researcher at Clemson University, identified at least 59 of those accounts with zero engagement, said X probably suspended them before they could gather any following, and called the episode a "non-story"; he told Business Insider, which reported the finding on Friday.
"No real humans saw these posts about data centers," Linvill said.
Distinct publishers with included, body-backed reporting in this cluster.
1 article · September 1, 2026
Follow any of these and your For You feed starts watching them — no settings page required.
invest
X discloses Chinese bot farm accounts posting on AI data centre and energy debate2 distinct publishers
leadership
Abbott traces his data-center moratorium to Texans rather than Beijing1 distinct publisher
invest
The NRSC just told AI firms their data centers are a losing issue1 distinct publisher
leadership
Data center opposition is now a siting cost, and the industry is pricing it as a PR line1 distinct publisher
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Two headline numbers, neither audited
The 200,000 comes from X and only from X. The 90 posts and zero engagement come from a spreadsheet The Next Web says outright it has not read, relayed through Business Insider. What does hold up independently is the scaffolding around the fight: PJM's clearing price, the Penn and YouGov readings, OpenAI's June report naming the same operation, and Abbott on camera. A reader can check the context but not the two facts the argument turns on.
The backlash is counted; the bots barely registered
Adoption splits cleanly in two directions here. The opposition is documented in hard tallies — 61% in Penn's August read against 49% in the spring, 47-to-24 negative at YouGov, moratoriums past 500 across 42 states, about 1,800 Texas developments stopped. The influence operation's own traction is the smallest number in the story: two or three posts per account, 90 in total, no engagement on the 59 checked. The claim about the bots travelled far further than the bots did.
A number in place of a reach
X put a 200,000-account farm on the table, of which the accounts it actually accuses number 200 — one in a thousand — and it never said what any of them reached. The measured answer, from the only person who went looking, is essentially nothing. Meanwhile the governor whose freeze the narrative would explain says flatly that China had nothing to do with it. The overstatement sits with the disclosure, not with this reporting, which keeps the caveats visible; but 3.2 million views for a claim whose demonstrated footprint is 90 posts is a wide gap.
Everyone in the frame owns data centres
Musk owns X and his companies run on data centres; OpenAI, the other source of Chinese-propaganda findings, is buying the same buildout. Samuel Woolley's point in this reporting is that every such report he has seen came from a Big Tech company with a stake in it, and Jim Prosser's line to Axios — if Abbott and Hochul agree, it is probably not a psyop — makes the political version. Add a congressional letter citing $45.8bn of blocked projects in Virginia and Abbott's own reversal under constituent pressure, and there is no disinterested party in the story except the university researcher who says it is a non-story.
Coherent account, one lens, one unopened file
The pieces fit: OpenAI named this operation in June, the recovered posts are about the PJM auction that genuinely cleared at the cap, and the January dating matches accounts suspended before they could grow. What keeps confidence middling is structural — a single publisher, the decisive count arriving through Business Insider from a spreadsheet nobody outside has opened, roughly 70% of the flagged accounts unexamined, and X silent since. Directionally this is likely right; precisely, it is one researcher's sample.