Political advertising about data centers on Facebook and Instagram has become sharply more critical as the midterm election approaches, according to a Pew Research Center analysis. Advertisers spent almost $12 million on such ads in the 90 days from July 27 through October 24.
Among ads mentioning data centers in October, 57% expressed opposition and just 3% clearly supported them. In January, supportive ads had been slightly more common than critical ones, at 42% and 31%, respectively.
The analysis covers U.S. advertising about politics, elections or social issues in Meta's Ad Library. It measures the content of ads, rather than a representative survey of voters or all discussion of data centers.
Corporate involvement was mentioned in 30% of data-center political ads, energy use in 24%, environmental effects in 17% and water use in 15%. Ads opposing data centers were especially likely to discuss technology companies or utilities, while supportive ads more often discussed jobs.
Ads that did not promote a particular candidate accounted for 61% of recent spending. These were typically paid for by advocacy groups. Among those ads, 56% opposed data centers and 13% supported them.
The other 39% of spending supported specific candidates. Among campaign ads in the analysis period, 72% promoted Democrats and 15% promoted Republicans; the rest backed independents or candidates whose party could not be determined. A majority of campaign ads criticized data centers across party lines, and fewer than 1% supported them.
Since late July, more than $1 million in data-center political advertising targeted users in each of Ohio, Wisconsin and Texas. Such ads represented about 14% of all political advertising on those platforms aimed at Wisconsin users, the largest share of any state.
Pew found that spending generally was higher in states with more competitive major elections and in states with more existing or planned data centers. Those associations do not establish what caused an advertiser to spend money or how the ads affected voting.
Researchers examined ads running from January 1 through October 24 and used the midpoint of Meta's spending ranges to estimate totals. A language model classified stance, themes and candidate affiliation. Across categories, it achieved 95% accuracy and an average F1 score of 0.88 against 300 ads labeled by trained human coders.
FingerLakes1.com is the region’s leading all-digital news publication. The company was founded in 1998 and has been keeping residents informed for more than two decades. Have a lead? Send it to news@fingerlakes1.com.
