BusinessMarkets, money & strategy
AI polls below ICE with US voters even as chatbot use reaches half of US adults
US voters rate AI less favorably than ICE, yet about half of adults now use chatbots. An MIT Technology Review essay blames aggressive corporate deployment rather than the technology, but no reviewed survey tests that explanation.
AT Pulse research deskOctober 7, 2026 · 5 min read · AI-assisted analysisDislike and adoption are rising together. In an NBC News poll of registered voters fielded from late February to early March 2026, 26% viewed AI positively and 46% negatively, for a net rating of −20. ICE's net rating was −18. AI ranked lower because fewer people viewed it positively, not because more viewed it negatively. In a Pew survey fielded around the same time, 49% of US adults said they use chatbots, up from 33% in 2024, and roughly a quarter use them daily. Sensor Tower separately estimates that ChatGPT's app has passed 1 billion monthly users. [2] [5] [3] [8] [4]
Public attitudes are mixed rather than simply souring. In Stanford's 2026 AI Index, 52% of people worldwide say AI products make them nervous. Over the same period, the share who say AI's benefits outweigh its drawbacks rose from 55% to 59%. North America and Europe are the most skeptical regions, and Southeast Asia is the most optimistic. [6]
Opposition is concrete when it comes to infrastructure. In Gallup polling, 71% of Americans opposed an AI data center in their area, compared with 53% who opposed a local nuclear plant. Opponents most often cited water and power use, followed by quality of life and utility costs. [7] [11]
The MIT Technology Review essay argues that people are hostile to the companies pushing AI on them, not to the technology itself. It also argues that regulation and open models give the public more leverage than it had over social media. This is the author's interpretation. The polls show the contradiction but do not test what causes it. [1]
polls show broad dislike of AI alongside rapid adoption and strong local opposition to data centers. Implication: product teams and infrastructure planners may face trust and siting hurdles even as demand grows.
Read the full assessment
The cause of the hostility remains unestablished.
I've finished the research and am now writing the dossier.
Executive brief
Americans now rate AI below ICE, yet AI use keeps growing fast. In a March NBC News poll, AI's net favorability among registered voters was −20, against −18 for ICE (Fortune). In the same period, about half of US adults said they use chatbots (Pew), and ChatGPT's app passed an estimated 1 billion monthly users (Daily Star/Sensor Tower). MIT Technology Review's opinion piece of 5 October argues that people dislike the companies pushing AI on them, not the technology itself. That is the author's interpretation. The surveys document the contradiction but do not test its cause.
What changed and event timeline
Stanford AI Index public-opinion chapter
Worldwide, 52% say AI products make them nervous. The share who say AI's benefits outweigh its drawbacks also rose, from 55% to 59% ().
AI polls below ICE
NBC surveyed 1,000 registered voters from 27 February to 3 March. 26% viewed AI positively and 46% negatively. ICE drew more negative ratings than AI but also more positive ones (;).
Data centers less welcome than nuclear plants
In Gallup's March fieldwork, 71% opposed a local AI data center (48% strongly), compared with 53% opposing a local nuclear plant ().
Pew finds half of US adults use chatbots
49% of adults use chatbots, up from 33% in 2024, based on a 17–23 February survey. Roughly one in four use them daily (;).
Gemini app reaches 950 million monthly users
Google gave the figure on its Q2 earnings call and said monthly users had nearly tripled year over year ().
MIT Technology Review names the paradox
The opinion essay argues that public hostility is aimed at aggressive corporate deployment, and says regulation and open models give people more leverage than they had over social media ().
Capabilities and access
This story is about public opinion, not a model release.
Read the full section
This story is about public opinion, not a model release. The products in the data:
- ChatGPT: 44% of US adults have used it. Gemini is at 24%, Copilot 17%, Meta AI 14%, Grok 8% and Claude 6% (TNW/Pew).
- Springboards' Flint: The startup whose CEO opens the essay. Flint is fine-tuned from Alibaba's open Qwen 3 and is built to give more varied answers. It is aimed at advertising and marketing creatives (MIT Technology Review).
Technical analysis for researchers and developers
- Sensor Tower's number counts monthly users of the app only (Daily Star). Google's Gemini number is its own reported figure.
- Pew's trend line has a break.
- The model learns where in its output more variety is possible and inserts less predictable words at those points.
Read the full section
- Measurement differs across sources. Sensor Tower's number counts monthly users of the app only (Daily Star). Google's Gemini number is its own reported figure. These two should not be compared directly.
- Pew's trend line has a break. In 2024, Pew asked the chatbot question only of people who had heard of chatbots, so the jump from 33% to 49% is not a strict like-for-like comparison (Pew).
- Flint's method. The model learns where in its output more variety is possible and inserts less predictable words at those points. It responds to "Artificial Hivemind," the NeurIPS 2025 best-paper finding that 25 LLMs gave near-identical metaphors (MIT Technology Review).
Claims and evidence
- 71% oppose local data centers vs 53% for nuclear — Independent poll: Gallup ()
- 52% of people worldwide are nervous about AI
- 36.8% of people in OECD countries used generative AI in 2025 — Official statistics ()
Read the full section
| Claim | Status |
| 71% oppose local data centers vs 53% for nuclear | Independent poll: Gallup (Planetizen) |
| 52% of people worldwide are nervous about AI | Independent: survey data compiled by Stanford (HAI) |
| 36.8% of people in OECD countries used generative AI in 2025 | Official statistics (ITIF) |
| ChatGPT app passed 1B monthly users | Third-party estimate (Sensor Tower) |
| Gemini has 950M monthly users | Vendor-reported (Google) |
| Flint scores 7/10 on NoveltyBench vs a 2.88 average | Vendor-reported (Springboards) |
| More than 2,100 state AI bills | Not matched in the reviewed sources. MultiState counted about 1,200 in 2025 and 1,561 by March 2026 (MultiState) |
Context and prior work
The essay compares AI to social media, which people kept using through years of backlash against the platforms. The essay's claim that pessimism tracks higher adoption is a correlation only. Stanford also reports that employees in India, China and Nigeria use AI at work at rates above 80%, which complicates that pattern.
Read the full section
The essay compares AI to social media, which people kept using through years of backlash against the platforms. Stanford's data shows a regional split: Southeast Asia is the most optimistic, while North America and Europe are the most skeptical (HAI). The essay's claim that pessimism tracks higher adoption is a correlation only. Stanford also reports that employees in India, China and Nigeria use AI at work at rates above 80%, which complicates that pattern.
Limitations, safety and contested findings
- The NBC comparison is narrow. AI ranks below ICE because fewer people view it positively, not because more view it negatively (Techstrong).
- Optimism and nervousness are rising together. That runs against a simple "souring" story (HAI).
- The age claim is not checked. The essay says pessimism is strongest among young people; the reviewed Pew material did not confirm this.
Read the full section
- The NBC comparison is narrow. AI ranks below ICE because fewer people view it positively, not because more view it negatively (Techstrong).
- Optimism and nervousness are rising together. That runs against a simple "souring" story (HAI).
- The age claim is not checked. The essay says pessimism is strongest among young people; the reviewed Pew material did not confirm this.
- The central argument is untested. No reviewed survey checks whether people's hostility is aimed at companies rather than at the technology.
- Flint is unreliable at the edges. It "sometimes falls over" when pushed too far (MIT Technology Review).
Business and practitioner implications
- Data-center siting is a real constraint. Opponents cite water and power use (50%), quality of life (24%) and utility costs (20%) (The Register).
- Distrust is high among users too. 71% of Americans expect AI to make their data less secure (TNW).
- Compliance is fragmenting. State-level AI bills are rising roughly tenfold over three years, which means more varied rules to track.
Read the full section
- Data-center siting is a real constraint. Opponents cite water and power use (50%), quality of life (24%) and utility costs (20%) (The Register).
- Distrust is high among users too. 71% of Americans expect AI to make their data less secure (TNW). Products that are open about what they do and give users clear controls may face less pushback.
- Compliance is fragmenting. State-level AI bills are rising roughly tenfold over three years, which means more varied rules to track.
- There is room to differentiate. Open models such as Qwen let small firms like Springboards build specialized alternatives instead of chasing the general-purpose labs.
Sources
Read the full section
- MIT Technology Review: People really hate AI, so why can't they get enough?
- MIT Technology Review: LLMs are stuck in a groupthink rut
- Fortune: NBC AI opinion poll
- Techstrong.ai: AI unpopularity crisis
- Stanford HAI: 2026 AI Index, Public Opinion
- The Register: Gallup data center poll
- Planetizen: Gallup data centers vs nuclear
- Pew Research Center: chatbot use chart
- The Next Web: Pew 2026 findings
- The Daily Star: ChatGPT reaches 1B MAU (Sensor Tower)
- t2online: Gemini 950M users
- ITIF: 36.8% of OECD individuals used generative AI
- MultiState: AI legislation tracker
- Springboards: Flint launch (vendor)
The source trail.
Sources (14)
People really hate AI, so why can’t they get enough?
Article text retrieved; extracted text may omit tables or interactive elements.
technologyreview.com