News · 2026-10-23

Ipsos poll finds 57% of US adults view AI as bad for society

An ABC News–Washington Post–Ipsos poll found that 57% of US adults consider AI bad for society, while 42% consider it good. Results published October 21 also show strong opposition to new data centers in separate national and local question groups. The findings make public acceptance a concrete constraint for the AI industry, without identifying the causes of those attitudes.

Key facts

• Ipsos surveyed 2,545 US adults through its probability-based KnowledgePanel from October 9 to 13.
• On the broad society question, 57% chose bad and 42% chose good.
• Separate half-samples showed 66% opposing new data centers nationally and 72% opposing one in their community.
• Primary source: Ipsos’s topline results and methodology.

A new model release can attract enthusiastic users while the infrastructure supporting it faces broad resistance. The poll places those two facts in the same industry conversation. It asks about AI as a social development and about data-center construction, giving a wider view than comments on a launch announcement or a developer forum.

Ipsos’s question begins, “Do you think artificial intelligence (AI) is a (good) thing or a (bad) thing for society?” The order of the response options was reversed for half the sample. Naming the exact question matters: an attitude toward AI for society is broader than approval of a specific assistant, a workplace tool or a medical application. People can find an individual product useful and still judge the wider transition negatively.

The ABC News report describes the results, while the Ipsos document provides the wording and methodology. The full-sample margin of sampling error is reported as plus or minus two percentage points. That full-sample figure should not automatically be attached to questions asked of only half the participants.

The construction questions used two random groups. One group considered new data centers in the United States in general. The other considered construction in the respondent’s own community. Opposition was higher in the local group, but those are different respondents answering different questions. The result does not show the same individuals first approving national construction and then rejecting a project near home.

Think of the design as two comparable rooms receiving different versions of a question. The difference between the rooms can illuminate wording or context, but it is not a before-and-after change in each person. Maintaining that distinction prevents a tempting narrative about individual hypocrisy from being read into a split-sample poll.

The industry context is concrete. Data centers need land, electricity, water management and approvals. Those requirements intersect with local institutions rather than ending at a model’s performance score. The Finnish authority’s October 22 notice illustrates one such interface: it sought a detailed water-management plan for a Google subsidiary’s Muhos site after earlier directions to halt preparatory work while environmental assessment remained unfinished.

That Finnish case does not explain the American poll. It demonstrates a kind of issue that aggregate enthusiasm about computing can overlook. A company can have a strong demand forecast while still needing to establish that a particular site’s environmental and infrastructure obligations are understood. Public acceptance and technical capacity are different parts of the same expansion problem.

Environmental comparisons also require careful accounting. Berget AI’s October 21 release estimates different emissions for a typical inference request in Sweden and Texas using its stated hardware and grid assumptions. That is a vendor’s modeled comparison, rather than a universal footprint for every request. It shows why broad public judgments and narrowly measured engineering claims need separate evidence.

The strongest counterargument to treating this poll as an industry verdict is that broad questions compress many concerns into one answer. The supplied results do not isolate whether respondents were thinking about employment, fraud, energy use, accuracy, power concentration or something else. Nor does opposition to data centers automatically predict how voters, planning boards or regulators will respond to a specific proposal with a specific benefits package.

The dossier’s online engagement snapshots likewise cannot substitute for polling. Attention to environmental coverage shows that a story circulated among particular online audiences. A probability-based survey is designed to answer a different question about a wider population. Ground Truth’s lessons on sample complexity and randomized comparisons explain why sample design and the unit of comparison matter.

The practical consequence is a need for evidence that addresses the decision being made. A national poll helps establish the public mood. A proposed facility needs local consultation, credible resource accounting and enforceable commitments. A product needs evidence of its benefits and risks. The poll does not settle those decisions, but it gives companies a reason to stop assuming that demand for AI services implies broad consent to the way they are built.

Primary source, verified: read the paper →

Ground Truth. (2026, October 23). Ipsos poll finds 57% of US adults view AI as bad for society. Ground Truth. https://groundtruth.day/news/ipsos-poll-finds-majority-negative-on-ai-and-data-centers.html

@misc{groundtruth:ipsos-poll-finds-majority-negative-on-ai-and-data-centers,
  title  = {Ipsos poll finds 57% of US adults view AI as bad for society},
  author = {{Ground Truth}},
  year   = {2026},
  month  = {oct},
  url    = {https://groundtruth.day/news/ipsos-poll-finds-majority-negative-on-ai-and-data-centers.html}
}

Topics: public-opinion · infrastructure · data-centers · society · industry