The Artificial Intelligence — Topline Report is the weighted topline for the Annenberg Public Policy Center's W38 Institutions of Democracy national survey, published August 10, 2026 (Source: annenbergpublicpolicycenter.org). It reports the second wave in which the center's core AI items were fielded, and prints the earlier W37 wave alongside for every repeated item, which makes it usable as a change measurement rather than only a snapshot.
Its headline result is a sharp rise in opposition to local data-center construction. Its more consequential structural finding is that this was the only item in the battery to move: attitudes toward AI regulation and toward AI's overall impact were statistically unchanged over the same four months.
Method
W38 was fielded June 16 to July 19, 2026 among 1,320 U.S. adults on a probability-based online panel, with a design effect of 1.67 and a margin of error of ±3.48 percentage points at 95% confidence. W37 was fielded February 17 to March 20, 2026 among 1,330 adults. Of the W38 respondents, 1,241 were also interviewed in W37, a 93% retention rate, which allows a within-person change estimate alongside the wave-to-wave comparison.
Results are weighted to national benchmarks for age, sex, race/ethnicity, education and region. Margins of error use replicate-based variance estimation. "Don't know," "Not sure" and "Refused" responses are included in the denominator and reported together, so every percentage is a share of all respondents asked.
Two reporting choices matter for how the figures are used. Six impact areas were asked only in June–July and seven only in February–March; the document states these should not be read as a single-wave ranking. And it prints two change measures — a wave-to-wave comparison whose independent-samples significance test the document itself calls conservative given the panel overlap, and a within-person mean change offered as a robustness check.
Data centers
Opposition to the construction of new data centers in the respondent's own area reached 60.6% in June–July 2026, against 21.3% support in the prior wave falling to 14.4%. The topline's rounded distribution gives 61% opposed — 44% strongly, 16% somewhat — 14% in support, and 25% neither.
| Measure | Feb–Mar 2026 | Jun–Jul 2026 | Wave diff | Wave 95% CI | Within-person diff |
|---|---|---|---|---|---|
| Oppose new local data centers | 48.9 | 60.6 | +11.8 | +6.9, +16.6 | +12.5 |
| Support new local data centers | 21.3 | 14.4 | −6.9 | −10.8, −3.1 | −7.1 |
Opposition holds across party: Democrats 69%, Republicans 54%, independents 53%. It is highest among adults under 30 at 70%, falling with age to 57% among those 65 and older, and shows no meaningful gradient by AI usage — non-users 64%, light users 60%, heavy users 60%. The item's test–retest correlation among panelists is 0.55, the lowest of the four repeated items, consistent with an attitude that was still moving.
Items that did not move
Three repeated measures were statistically unchanged, each with a confidence interval spanning zero:
| Measure | Feb–Mar 2026 | Jun–Jul 2026 | Wave diff | Wave 95% CI |
|---|---|---|---|---|
| Government has done too little to regulate AI | 65.4 | 67.9 | +2.5 | −2.1, +7.2 |
| Government has done too much to regulate AI | 8.2 | 8.3 | +0.1 | −2.8, +2.9 |
| AI's impact on the U.S. will be negative | 41.9 | 39.5 | −2.4 | −7.2, +2.4 |
On regulation, 68% said too little, 22% about the right amount and 8% too much in June–July. The partisan spread is wide but one-directional: 80% of Democrats, 61% of independents and 58% of Republicans said too little, with the "too much" share between 8% and 9% in every group. On overall ten-year impact, 39% expected negative, 18% positive and 35% equal.
Self-reported use fell slightly: 34% used AI many times or almost every day in June–July against 38% in February–March, with the "never" share rising from 16% to 21%. Awareness was broadly flat, with 76% reporting having heard a great deal, a lot, or a moderate amount.
Expected impact across 13 areas
The document closes with a net-impact table combining the six areas asked in June–July with the seven asked in February–March. Net is the positive share minus the negative share, both DK-inclusive.
| Area | Wave | Positive | Negative | Net |
|---|---|---|---|---|
| Medical research and discoveries | Feb–Mar | 56.6 | 15.8 | +40.8 |
| Effectiveness of government services | Feb–Mar | 23.6 | 30.6 | −7.0 |
| Education and skill-development | Jun–Jul | 28.5 | 39.8 | −11.3 |
| The economy | Feb–Mar | 19.1 | 37.7 | −18.6 |
| The creative arts | Feb–Mar | 22.0 | 43.3 | −21.4 |
| Household utility costs | Feb–Mar | 13.6 | 37.7 | −24.0 |
| Mental health and well-being | Feb–Mar | 16.8 | 47.6 | −30.8 |
| Environmental protection and sustainability | Jun–Jul | 15.2 | 48.1 | −32.9 |
| U.S. competition with China | Feb–Mar | 4.9 | 42.6 | −37.7 |
| Accuracy of news and information online | Jun–Jul | 12.4 | 51.4 | −39.0 |
| Employment and jobs | Jun–Jul | 11.6 | 58.0 | −46.5 |
| Children's safety online | Jun–Jul | 10.8 | 60.9 | −50.1 |
| Personal privacy and data security | Jun–Jul | 6.4 | 69.1 | −62.8 |
Medical research is the only area with a net-positive expectation. Personal privacy and data security is the most negative, and is the one area where heavy users are no more optimistic than non-users — 69% negative in both groups. On several other areas heavy use is associated with more positive expectations: education and skill-development runs +13 net among heavy users against −35 among non-users, and medical research +62 against +4.
Items asked only in February–March
Two questions from the earlier wave are printed without a June–July counterpart. Asked whether the federal or state governments should take the lead on regulating AI, 52% said federal, 15% state and 33% were unsure. Asked about job loss or reduced hours due to AI, 29% were worried and 41% not, with 30% not working; worry peaked at 44% among those aged 30 to 49.
Provenance and limits
Published on the Annenberg Public Policy Center's own domain and self-identified as the W38 IOD National Survey topline, prepared with the analysis script AI_W38.R. All figures are self-contained in the document; no external corroboration is required for the numbers themselves.
The document reports attitudes, not behaviour or outcomes, and its comparability is bounded by which items were asked in which wave — a limit it states about its own 13-area table. Coverage dated the release August 11, 2026 and gave the margin of error as ±3.5 points and the data-center shift as 12 points from a 49% baseline; the document's own values are August 10, ±3.48, and 48.9 to 60.6.
Relationships
- supports: Public Opinion on AI — a two-wave change measurement on the core AI attitude battery.
- supports: Data Center Siting / AI Power Politics — the sharpest published movement on local data-center opposition.
- related: AI Labor Disruption — the employment and jobs area, net −46.5.
- related: AI and Privacy — personal privacy and data security is the most negative of the 13 areas.
- related: AI Environmental Impact — environmental protection and sustainability, net −32.9.